AI SEO Glossary
500+ AI SEO Terms & Definitions, Explained Plainly
From GEO, AEO and LLMO to large language models, RAG, AI crawlers, structured data, agentic commerce and AI visibility metrics — every term used in AI search today, explained in plain English and organised by category, by AmplifyKlicks’ AI SEO team in Vadodara. Classic SEO vocabulary lives in our SEO Glossary.
Jump to a Category
AI Search Fundamentals
How search itself has changed now that answers are generated, not just listed — the core labels first, then the vocabulary of the new landscape around them.
AI SEO
The practice of optimising a business’s website, content and entity footprint so it is found, cited and recommended by AI-powered search — Google AI Overviews and AI Mode, ChatGPT, Gemini, Perplexity and Copilot — alongside traditional rankings. It combines classic SEO with GEO, AEO and entity work. Learn more →
AI Search
Search experiences that generate a direct, synthesised answer using AI rather than only returning a list of links. It spans AI features inside Google and Bing and standalone assistants such as ChatGPT and Perplexity.
Generative Search
A search experience in which results are written by a language model as a synthesised answer — with citations — instead of presented as static ranked links. AI Overviews, AI Mode and Perplexity are generative search products.
Answer Engine
A search system whose primary output is a single, written answer rather than a ranked list of links. ChatGPT, Perplexity, Google AI Mode and Copilot are answer engines; being quoted inside the answer is the goal, not just being listed below it.
Generative Engine
The academic and industry name for any search product that uses a large language model to generate its results — the “generative engine” in Generative Engine Optimization. It retrieves sources, then writes a new response from them.
Conversational Search
Search that happens as a back-and-forth dialogue instead of one-off queries. The engine keeps the thread of the conversation, so a follow-up like “which one is cheapest?” is understood in the context of the previous answer.
Synthesized Answer
A response that an AI engine composes by combining facts from several sources into one coherent piece of text. Your content may contribute a single sentence to a synthesized answer without your page ever being visible as a whole.
Answer Layer
The AI-generated block that now sits between the user and the classic results — an AI Overview, a Copilot answer, a Perplexity response. Winning the answer layer means being one of the few sources it cites.
Ten Blue Links
Shorthand for the traditional search results page: a plain list of ten ranked organic links. AI answers, featured snippets and video carousels have progressively pushed the ten blue links further down the page.
Zero-Click Search
A search that ends without the user clicking any result because the answer was delivered on the results page itself, increasingly by an AI summary. Zero-click searches make brand visibility and citations valuable even when they produce no session.
The Great Decoupling
The observed pattern in Google Search Console where impressions keep rising while clicks stay flat or fall — the two metrics “decouple”. It is the measurable footprint of AI Overviews answering queries that used to generate visits.
Search Journey Compression
The shortening of the research path from many searches and page visits down to one or two AI conversations. Compare-shop-decide steps that used to touch ten sites now happen inside a single answer engine session.
Search Fragmentation
The spread of search behaviour across Google, AI assistants, social platforms, marketplaces and app-based search instead of one dominant engine. It is why visibility now has to be tracked across several surfaces at once. Learn more →
Search Everywhere Optimization
An expanded view of SEO that optimizes a brand for every place people search — Google, AI chatbots, YouTube, Amazon, app stores, social search and voice assistants — rather than only for classic web search.
AI-First Search
A search product where the generated answer is the default experience and links are secondary. ChatGPT Search and Perplexity are AI-first; Google is moving that way through AI Mode.
Hybrid SERP
A results page that mixes an AI-generated answer with traditional organic listings, ads and rich features. Most Google results pages are now hybrid, which is why classic rankings and AI citations must be worked on together.
AI-Organized Results
A results layout in which AI groups links, snippets and media into thematic clusters (as in Google’s Web Guide) instead of a single ranked list. Pages surface under a sub-topic rather than a position number.
Commodity Content
Information that is available in the same form on dozens of sites — generic definitions, rewritten how-tos, summarised news. AI engines can generate commodity content themselves, so it earns few citations. Google’s own AI-search guidance recommends “non-commodity” content.
Non-Commodity Content
Content that exists nowhere else: original data, first-hand testing, proprietary frameworks, expert opinion, local specifics. It is the strongest predictor of being cited in AI answers because the engine cannot synthesise it from other sources. Learn more →
Answer Engine Results Page (AERP)
A term some practitioners use for the output screen of an AI answer engine — the response text, its inline citations and source cards — as the AI-era equivalent of the SERP.
Model Knowledge vs. Live Retrieval
The two ways an AI can know about your brand: what was baked into its training data (model knowledge) and what it fetches from the web at answer time (live retrieval). Fresh pages influence retrieval immediately; changing model knowledge takes months.
Discoverability (AI)
How easily an AI system can find, access and understand your content when it looks for sources. It depends on crawler access, clean HTML, entity clarity and being linked from pages the engine already trusts.
AI Search Readiness
An assessment of how prepared a website is to be found and cited by AI engines — crawler access, structured data, entity signals, answer-formatted content, and measurement in place. Usually the first deliverable of an AI SEO engagement. Learn more →
Related terms on this page
Related terms in the SEO Glossary
AI Optimization Disciplines
The named practices that have grown up around AI search — GEO, AEO and LLMO first, then the other labels you will meet in proposals, job posts and tool dashboards.
Generative Engine Optimization (GEO)
Optimising content and brand signals so a business is cited and recommended inside AI-generated answers from ChatGPT, Gemini, Perplexity and Google AI Overviews. The term comes from a 2023 Princeton study that showed statistics, quotations and citations measurably increase visibility in generative engines. Learn more →
Answer Engine Optimization (AEO)
Structuring content — clear questions, direct answers, well-organised FAQs, concise definitions and structured data — so answer engines and voice assistants can extract and cite it accurately. AEO focuses on the format of answers; GEO on the broader brand presence. Learn more →
LLM Optimization (LLMO)
Optimising content and entity signals specifically to be understood, trusted and referenced by large language models such as those powering ChatGPT, Claude and Gemini — including the training data and knowledge bases models learn from, not only live search results.
AI Search Optimization
The umbrella term for the combined GEO, AEO, entity and technical work needed to be found and cited across AI-powered search platforms. Often used interchangeably with AI SEO.
Retrieval Optimization
Structuring content so it is easily retrieved and matched by the systems AI engines use to pull source material before generating an answer — clear passages, semantic completeness, crawler access and fresh, well-labelled pages.
Citation Optimization
Optimising content specifically to increase the likelihood that AI systems cite it by name as a source: quotable sentences, original data, named expertise, corroborated facts and being present where engines already look.
Generative AI Optimization (GAIO)
A synonym for GEO used mainly in European agencies: optimizing a brand’s content and entity footprint so generative AI systems mention and recommend it. The practice is identical; only the acronym differs. Learn more →
AI Optimization (AIO)
A broad umbrella term for making any digital asset work better with AI systems — search visibility, chatbot readiness, machine-readable product data. In SEO circles it is used loosely as another name for AI-search optimization.
LLM SEO
Optimizing specifically for the large language models behind ChatGPT, Claude, Gemini and Grok, as opposed to Google’s classic index. It emphasises entity clarity, consistent brand facts across the web, and being present in the sources models retrieve from.
AI Visibility Optimization (AIVO)
The practice of measuring and increasing how often a brand appears in AI-generated answers. It ties the optimization work of GEO/AEO to an explicit metric layer: prompt tracking, citation share and sentiment.
AI Overview Optimization
Work aimed at being cited inside Google’s AI Overviews and AI Mode specifically: ranking in the top organic results for the query and its fan-out sub-queries, answer-first formatting, and clean structured data. Learn more →
Conversational Search Optimization
Optimizing content for multi-turn, natural-language sessions: anticipating follow-up questions, covering comparison and “which is best for me” angles, and writing in the language people use when they talk to an assistant.
Relevance Engineering
A term popularised by Mike King (iPullRank) for the technical discipline of shaping content so it scores highly in the vector-based relevance systems modern engines use — chunk-level relevance, embeddings and passage structure rather than keyword placement. Learn more →
Knowledge Graph Optimization (KGO)
The practice of getting a brand, its people and products accurately represented as entities in Google’s Knowledge Graph, Wikidata and similar databases, so AI systems have a verified record to ground answers in. Learn more →
Multimodal Optimization
Optimizing images, video, audio and text together so AI systems that understand all of them (Gemini, GPT-4o-class models, Google Lens) can use each format as evidence. Includes alt text, transcripts, captions and descriptive file context.
Voice Search Optimization
Structuring content so voice assistants (Google Assistant, Siri, Alexa) and voice modes in AI apps can read a concise, spoken answer aloud. Overlaps heavily with AEO: direct answers, question phrasing and local details.
Agent Experience (AX)
The AI-era counterpart to UX: how easily an autonomous AI agent can read, navigate and act on a website — parse its content, fill its forms, complete a purchase. Poor AX means agents route users to competitors.
AI Visibility Audit
A structured review of where a brand currently appears across AI engines — which prompts it is cited for, which competitors are cited instead, crawler access, entity accuracy and content gaps. The starting point of any GEO/AEO roadmap. Learn more →
AI Reputation Management
Monitoring and correcting what AI assistants say about a brand: outdated facts, negative framing, confusion with similarly named companies. Fixes flow through the entity sources models rely on — your site, Wikipedia/Wikidata, reviews and press.
Total Search Optimization
An agency framing that combines classic SEO, AI search, paid search and marketplace search into one visibility plan measured on total demand captured rather than Google rankings alone.
Related terms on this page
Related terms in the SEO Glossary
AI Search Platforms & Features
The engines, assistants and result features your customers actually use. Knowing what each one retrieves from, and how it cites sources, tells you where to focus.
AI Overviews
Google’s AI-generated summary shown at the top of many search results, answering the query directly with information synthesised from multiple sources and showing link cards to them. Launched in 2024 as the successor to the SGE experiment and now live in India and most markets.
AI Mode
Google’s conversational, chat-style search experience — a tab on the results page — that answers with a full AI response, supports follow-up questions, and uses query fan-out to search many sub-queries at once. Rolled out in India in 2025.
Search Generative Experience (SGE)
Google’s 2023 Search Labs experiment that placed an AI-generated summary at the top of results. It was the prototype for AI Overviews, which replaced it in May 2024.
Deep Search (AI Mode)
An AI Mode capability that issues hundreds of sub-queries and reasons across them to produce a fully cited, report-style answer for complex research questions. Content that covers a sub-topic thoroughly can be cited even when it would never rank for the head term.
Search Live
Google’s real-time voice (and camera) conversation mode inside AI Mode, letting users talk to Search and ask about what their phone camera sees. Answers are spoken, so concise, clearly phrased facts are what get used.
Web Guide
A Google Search Labs experiment that uses Gemini to organise results into AI-labelled groups (“getting started”, “expert opinions”, etc.) instead of a flat list — an example of AI-organized results.
Preferred Sources
A Google feature letting users pick publications they want to see more of in Top Stories. Marketing your site as a source worth selecting is one of the few direct user-controlled visibility levers Google offers.
Ads in AI Overviews
Sponsored placements Google shows inside or directly beneath AI Overviews and AI Mode answers, labelled “Sponsored”. They come from existing Search and Shopping campaigns rather than a separate ad product.
AI Overview Source Links
The clickable citations shown alongside an AI Overview — link cards, inline chips and the expandable source panel. Being one of these sources is the measurable outcome of AI Overview optimization.
Gemini
Google’s family of multimodal AI models and the consumer assistant app built on them (formerly Bard). Gemini answers with Google Search grounding, so classic ranking and entity signals strongly influence what it says about you.
Gemini in Chrome
Google’s integration of the Gemini assistant into the Chrome browser, able to summarise, compare and act across open tabs. It turns the browser itself into an answer engine that reads your pages directly.
Google Lens
Google’s visual search tool that identifies objects, text and products in images and now feeds into AI Mode. Clear product photography with descriptive on-page context helps Lens connect images to your pages.
Circle to Search
An Android gesture that lets users circle anything on screen to search it instantly, often returning an AI Overview. It expands visual and contextual search moments far beyond the search box.
Multisearch
Google’s ability to combine an image and text in one query (“this sofa but in green”). It rewards pages whose images and surrounding text describe the same attributes consistently.
Ask Maps
Gemini-powered conversational search inside Google Maps, answering questions like “a quiet café near Alkapuri with parking” by reasoning over business profiles, reviews and attributes. Complete Google Business Profile data becomes the answer material. Learn more →
Google Discover
Google’s personalised content feed in the Google app and Chrome, now showing AI-generated summaries for some stories. Strong entity signals and clear headlines help pages surface in both the feed and its summaries.
Shopping Graph
Google’s constantly updated database of products, sellers, prices, reviews and inventory, powering Shopping results and AI Mode shopping. Accurate Merchant Center feeds and product schema are how a store enters it.
ChatGPT
OpenAI’s conversational assistant and the most used answer engine outside Google. It blends model knowledge with live web retrieval, so brand facts in its training data and citable pages both matter.
ChatGPT Search
The web-search capability inside ChatGPT (launched as SearchGPT in 2024) that fetches current pages and shows source citations. It relies on OpenAI’s own crawler-built index, so allowing OAI-SearchBot is a prerequisite for appearing. Learn more →
SearchGPT
The name of OpenAI’s 2024 search prototype, later folded into ChatGPT as ChatGPT Search. Still used informally to refer to ChatGPT’s search mode.
ChatGPT Shopping
ChatGPT’s product-research experience showing product cards, prices, reviews and merchant links inside answers. Products are sourced from structured feeds and third-party data rather than ads, so feed quality drives inclusion.
ChatGPT Atlas
OpenAI’s AI-native web browser with ChatGPT built into the address bar and an agent mode that can browse and act on pages. It makes every page you publish something the assistant can read on the user’s behalf.
Custom GPTs
User-built versions of ChatGPT with their own instructions, knowledge files and actions. Brands publish Custom GPTs as a branded assistant surface, and the GPT Store is a minor discovery channel of its own.
Deep Research
A long-running research mode offered by OpenAI, Google, Perplexity and others that browses dozens of sources and returns a cited report. It reads far down the results and rewards thorough, well-structured reference content.
Perplexity
An AI-first answer engine that retrieves live web results for every query and shows numbered citations prominently. Its transparency makes it the easiest platform on which to see exactly which pages an AI chose to cite.
Perplexity Pages
Perplexity’s feature that turns research threads into publishable, indexable articles. These pages themselves rank in Google and compete with publishers for informational queries.
Perplexity Comet
Perplexity’s agentic browser with its assistant embedded, capable of summarising pages, comparing tabs and completing tasks. Another sign that browsing itself is becoming AI-mediated.
Microsoft Copilot
Microsoft’s AI assistant across Windows, Edge, Bing and Microsoft 365, grounded in the Bing index. Bing SEO — Bing Webmaster Tools, IndexNow, clean crawlability — is the route into Copilot answers.
Copilot Search
Bing’s AI-generated search results mode that produces a synthesised, cited answer above traditional listings. It draws on Bing’s ranking, so Bing visibility translates fairly directly into Copilot citations.
Claude
Anthropic’s AI assistant, widely used for professional and research tasks, with web search and citations. Claude fetches pages through its own crawlers, which must be allowed in robots.txt to be citable.
Grok
xAI’s assistant integrated into X (Twitter), with real-time access to X posts and web search. Brand conversation on X becomes source material for Grok answers in a way no other engine replicates.
Meta AI
Meta’s assistant embedded in WhatsApp, Instagram, Facebook and Messenger, with web search grounding. Its scale in India makes it a significant answer surface for consumer queries.
DeepSeek
A Chinese AI lab whose open-weight reasoning models became widely used in 2025. Its assistant offers web search, and its models are embedded in many third-party apps, extending where your content may be retrieved.
Mistral Le Chat
The assistant from French AI company Mistral, offering web search and citations. Notable in European markets and as an example of open-model ecosystems that retrieve from the same public web.
You.com
An AI search engine and research assistant that presents cited answers alongside web results and lets users pick the underlying model. A smaller platform, but one that indexes the open web independently.
Brave Search AI
Brave’s independent search index with its “Answer with AI” summaries. Because Brave licenses its index to other AI companies, appearing in Brave can propagate into several assistants.
DuckDuckGo AI
DuckDuckGo’s DuckAssist answers and Duck.ai chat interface, drawing on Bing and Wikipedia. Another reason Bing visibility and a solid Wikipedia/Wikidata footprint matter.
Apple Intelligence
Apple’s on-device and cloud AI features across iPhone, iPad and Mac, including Siri’s ability to hand off questions to ChatGPT and AI summaries in Safari. Applebot is the crawler that feeds it.
Amazon Rufus
Amazon’s shopping assistant that answers product questions using listings, reviews and Q&A. For sellers, complete listing content and review volume are the “SEO” of Rufus recommendations.
Reddit Answers
Reddit’s own AI answer feature that summarises community threads with links to the source posts. Reflects the broader trend of communities becoming answer engines over their own content.
AI Assistant
Generic term for a conversational AI product (ChatGPT, Gemini, Claude, Copilot) that answers questions and performs tasks. When such assistants search the web, they behave as answer engines.
Regional AI Search Engines
AI-powered engines dominant in specific markets — Yandex (Neuro), Baidu (Ernie Bot), Naver (Cue:) — each with its own crawlers and content preferences. Relevant for brands targeting those markets. Learn more →
Related terms on this page
Related terms in the SEO Glossary
How Large Language Models Work
The machine-learning vocabulary behind every AI answer. You do not need to build a model, but understanding tokens, embeddings, context windows and hallucinations explains why certain content gets used and other content gets ignored.
Artificial Intelligence (AI)
Software that performs tasks normally requiring human intelligence — understanding language, recognising images, making decisions. In search, AI now interprets queries, ranks pages and writes answers.
Machine Learning (ML)
A branch of AI in which systems learn patterns from data instead of following hand-written rules. Google’s ranking systems and every LLM are machine-learning systems trained on enormous datasets.
Deep Learning
Machine learning using multi-layered neural networks that learn increasingly abstract representations. Deep learning made modern language understanding, image recognition and generative models possible.
Neural Network
A computing structure of interconnected “neurons” arranged in layers, loosely inspired by the brain, that learns by adjusting the strength of its connections. LLMs are extremely large neural networks.
Natural Language Processing (NLP)
The field of AI concerned with reading, interpreting and generating human language. Search engines apply NLP to understand queries and pages; LLMs are its most advanced application to date.
Natural Language Understanding (NLU)
The NLP sub-field focused on meaning: intent, entities, relationships, sentiment. It is what lets an engine know that “best dentist near me open now” is a local, transactional request.
Natural Language Generation (NLG)
The NLP sub-field focused on producing fluent text. Every AI Overview and chatbot reply is NLG output assembled from retrieved facts and model knowledge.
Large Language Model (LLM)
A neural network trained on vast amounts of text to predict the next token, which gives it the ability to understand and generate language. GPT, Gemini, Claude, Llama and Grok are LLMs; they power every answer engine.
Foundation Model
A large, general-purpose model trained once on broad data and then adapted to many tasks. LLMs are foundation models; search products are one of the tasks they are adapted for.
Frontier Model
The most capable model generation available at a given time from a leading lab (e.g., the latest GPT, Gemini or Claude). Frontier models set the ceiling for how well answer engines reason and cite.
Transformer
The neural-network architecture introduced by Google in 2017 that underpins all modern LLMs. Its attention mechanism lets the model weigh every word in a passage against every other, capturing context far better than earlier designs.
Attention Mechanism
The component of a transformer that decides which parts of the input matter most for each output token. It is why a clearly structured passage with the answer near the question is easier for a model to use.
Token
The basic unit of text an LLM reads and writes — roughly three-quarters of an English word. Costs, context limits and “how much of my page the model actually saw” are all measured in tokens.
Tokenization
The process of splitting text into tokens before a model processes it. Unusual spellings, brand names and non-English words often split into many tokens, which can make them harder for a model to recognise consistently.
Context Window
The maximum number of tokens a model can consider at once — its working memory for a single request. Retrieved passages, the conversation and the answer must all fit, which is why engines pull chunks rather than whole sites.
Parameters
The learned numerical values inside a model (billions or trillions of them) that encode everything it knows. Model size is usually quoted in parameters; more parameters generally means more capability and cost.
Model Weights
Another name for a model’s trained parameters, stored as large numeric files. “Open-weight” models publish these files so anyone can run the model.
Pre-Training
The initial, expensive phase in which a model learns language and world knowledge from a huge text corpus. Content on the open web at pre-training time becomes part of what the model “remembers”.
Training Data
The corpus of text, images and code a model learns from — web crawls such as Common Crawl, licensed publisher content, books, code and curated datasets. Whether your site is in it affects what the model believes about you.
Knowledge Cutoff
The date after which a model has no training data. Anything you published later is only known to it through live retrieval — one reason freshly updated pages depend on search grounding to be seen.
Parametric Knowledge
Facts stored inside a model’s weights from training, as opposed to information retrieved at answer time. It is slow to change and can be outdated, which is why AI reputation issues often trace back to old web content.
Fine-Tuning
Further training of a pre-trained model on a narrower dataset to specialise its behaviour. Search products fine-tune models to answer concisely, cite sources and follow safety rules.
Instruction Tuning
A form of fine-tuning that teaches a model to follow instructions and answer questions helpfully rather than merely continue text. It is what turns a raw language model into an assistant.
RLHF
Reinforcement Learning from Human Feedback — training that rewards responses humans rate as helpful, accurate and safe. It shapes stylistic preferences such as favouring clear, well-sourced answers.
Alignment
The broad effort to make AI systems behave according to human intentions and values — accurate, harmless, honest. Alignment choices influence what an assistant will and won’t recommend.
Inference
Running a trained model to produce an output for a given input. Every AI answer is an inference call; its cost is why engines retrieve small chunks and cache popular answers.
Temperature
A setting controlling how random a model’s word choices are: low temperature gives consistent, predictable answers; high temperature gives more varied ones. It is one reason the same prompt can cite different sources on different days.
Non-Deterministic Output
The property that an LLM can produce different answers to the identical prompt. AI visibility therefore has to be measured across many runs and prompts, not from a single check.
Hallucination
A confident but false statement generated by a model — an invented statistic, a wrong address, a product you never sold. Clear, consistent facts on your own site and in knowledge bases reduce hallucinations about your brand.
Embedding
A numerical vector that represents the meaning of a word, sentence, page or image so that similar meanings sit close together. Retrieval systems compare the embedding of a query with embeddings of content chunks.
Vector
A list of numbers describing an item’s position in a many-dimensional space. In AI search, every query and passage is turned into a vector so relevance can be computed mathematically.
Latent Space
The abstract, high-dimensional space in which a model organises meaning; related concepts cluster together. Content that clearly “belongs” to a topic occupies a tight region of latent space and is easier to retrieve.
Semantic Similarity
A measure of how close two pieces of text are in meaning, regardless of shared words. It is the core of vector retrieval and the reason synonyms and paraphrases can match a query.
Cosine Similarity
The most common way to score semantic similarity: the cosine of the angle between two vectors, from –1 to 1. Higher scores mean a passage is a closer match to a query.
Vector Database
A database designed to store embeddings and find nearest neighbours quickly. AI search engines and RAG systems use vector databases to retrieve candidate passages in milliseconds.
Multimodal Model
A model that understands more than one kind of input — text, images, audio, video — in the same system. Gemini and GPT-4o-class models are multimodal, so images and video on a page become evidence, not decoration.
Reasoning Model
A model trained to “think” through intermediate steps before answering (OpenAI’s o-series, Gemini Thinking, DeepSeek-R1, Claude’s extended thinking). Reasoning models cross-check sources, which raises the bar for consistency and accuracy.
Chain-of-Thought
A prompting and training technique in which a model writes out step-by-step reasoning before its final answer, improving accuracy on complex questions. Deep-research features rely on it.
Test-Time Compute
The amount of computation a model spends while answering (rather than while training). Giving models more time to reason at answer time is the mechanism behind reasoning models and deep-research modes.
Small Language Model (SLM)
A compact model (typically under ~10 billion parameters) that runs on phones or laptops. On-device assistants such as Apple Intelligence and Gemini Nano use SLMs, often with cloud fallbacks for hard questions.
Mixture of Experts (MoE)
A model architecture that routes each token to a few specialised sub-networks (“experts”) instead of the whole model, giving frontier capability at lower cost. Many current LLMs use it.
Model Distillation
Training a smaller “student” model to imitate a larger “teacher” model, keeping most of the capability at a fraction of the size. Distilled models power many cheap, fast search features.
Quantization
Compressing a model by storing its weights at lower numerical precision so it runs faster and cheaper. Common for on-device and high-volume inference.
Open-Weight Model
A model whose trained weights are published for anyone to download and run (Llama, Mistral, Gemma, DeepSeek). Open-weight models are embedded in countless third-party apps, spreading your brand’s training-data footprint.
Proprietary Model
A model available only through its maker’s products or API, with weights kept private (GPT, Gemini, Claude). Most consumer answer engines run on proprietary models.
Model API
The programmatic interface developers use to send prompts to a model and receive responses. SEO tools that track AI visibility work by querying model APIs at scale.
System Prompt
Hidden instructions the product sets before the user’s message, telling the model how to behave — tone, citation format, safety rules, when to search. System prompts explain why engines cite differently.
Prompt Engineering
Crafting inputs that get better outputs from a model — specifying role, format, constraints and examples. In SEO it is used both to operate AI tools and to build realistic prompt sets for visibility tracking.
Zero-Shot & Few-Shot Prompting
Asking a model to perform a task with no examples (zero-shot) or a handful of examples in the prompt (few-shot). Few-shot prompting is a quick way to get consistent, brand-styled output from AI writing tools.
In-Context Learning
A model’s ability to pick up a task or fact from information placed in the prompt, without retraining. Retrieval works because of it: a passage from your page in the context window can change the answer immediately.
Long-Context Model
A model with a very large context window (hundreds of thousands to millions of tokens) able to read entire documents or sites in one request. It reduces, but does not remove, the need for chunk-level clarity.
Position Bias (Lost in the Middle)
A documented tendency for models to pay most attention to information at the start and end of their context, and less to the middle. It reinforces putting the direct answer at the top of a passage.
Recency Bias
The tendency of retrieval-based answers to favour recently published or updated sources when freshness seems relevant. Visible last-updated dates and genuinely refreshed content benefit from it.
Model Bias
Systematic skew in a model’s outputs inherited from its training data — over-representing certain sources, regions, languages or viewpoints. It is one reason English-language, well-linked sites are cited more often.
Guardrails
Rules and filters that constrain what a model will say — blocking harmful content, avoiding medical or financial advice, requiring citations. Guardrails shape how commercial recommendations are phrased.
Model Card
A published document describing a model’s training data, capabilities, limitations and intended uses. Useful for understanding what a given engine can and cannot know.
Benchmark
A standardised test used to compare model capabilities (reasoning, coding, factual accuracy). Benchmark leadership drives which models search products adopt.
Evals
Short for evaluations: systematic tests of a model or AI feature’s output quality. SEO teams run their own evals — fixed prompt sets scored for brand presence and accuracy — to measure AI visibility over time.
LLM-as-a-Judge
Using one model to grade the outputs of another against a rubric. Visibility tools use it to classify whether an answer mentions a brand favourably, neutrally or negatively at scale.
Synthetic Data
Machine-generated training or test data. Labs use it to fill gaps in training; SEO teams use synthetic prompt sets to simulate how customers might phrase questions to an assistant.
Model Collapse
The degradation that occurs when models are trained on too much AI-generated output, losing diversity and accuracy. It is a structural reason engines value original, human-sourced content.
Related terms on this page
Related terms in the SEO Glossary
Retrieval, Grounding & RAG Mechanics
What happens between the prompt and the answer: how engines fan out queries, retrieve passages, rank them and decide which sources to cite.
Retrieval-Augmented Generation (RAG)
The technique AI systems use to pull real, current information from external sources (a web index, a knowledge base) before generating a response, rather than relying only on what the model learned in training. Every cited AI answer is a RAG output; being retrievable is the precondition for being cited.
Query Fan-Out
Google’s technique, used in AI Mode and AI Overviews, of breaking a query into many related sub-queries, searching them simultaneously and combining the results into one answer. It means pages can be cited for sub-topics they would never rank for as head terms.
Information Retrieval (IR)
The computer-science discipline of finding relevant documents for a query, underlying both classic search engines and the retrieval step of AI answers. Decades of IR research — relevance scoring, ranking, evaluation — still apply.
Retriever
The component of a RAG system that searches an index and returns candidate passages for a query. If the retriever never surfaces your page, the model cannot cite it, no matter how good the content is.
Generator
The language model in a RAG pipeline that writes the final answer from the retrieved passages and the prompt. It decides which retrieved sources are actually used and cited.
Grounding
Anchoring a model’s answer in verifiable external information — retrieved web pages, a knowledge graph, a database — instead of relying on memory alone. Grounded answers cite sources; ungrounded ones are more prone to hallucination.
Grounded Answer
A response whose claims are tied to specific retrieved sources, usually shown as citations. Google, Perplexity and ChatGPT Search all produce grounded answers for factual queries; being a grounding source is the target of GEO.
Grounding with Google Search
Google’s capability, used by Gemini and available to developers, of running a Google search and feeding the results to the model before it answers. It means Google’s classic ranking systems decide what Gemini sees.
Live Retrieval
Fetching current web content at the moment a question is asked, rather than relying on training data. All major assistants now do live retrieval for time-sensitive or specific queries.
Search-Augmented Generation
A specific form of RAG where the retrieval source is a web search engine (Google, Bing or a proprietary index). Most consumer AI answers are search-augmented, which is why search rankings still matter.
Web Index (LLM)
The crawled collection of pages an AI company searches when it needs live information. OpenAI, Anthropic, Perplexity and Google each maintain their own; your page must be in an engine’s index to be retrieved by it.
Query Rewriting
The engine’s reformulation of a user’s prompt into one or more search-friendly queries before retrieval. A conversational prompt about “that clinic you mentioned” becomes a precise query with the clinic’s name.
Query Decomposition
Breaking a complex question into smaller sub-questions that can each be searched separately, then combining the findings. It is the mechanism behind Google’s query fan-out and deep-research modes.
Sub-Query
One of the smaller searches an engine runs after decomposing a prompt. A single AI Mode answer may draw on dozens of sub-queries, each with its own top results — and each a separate opportunity to be cited.
Synthetic Query
A query generated by the engine itself rather than typed by a user, such as the sub-queries in a fan-out. Ranking for synthetic queries you would never see in keyword tools is a major source of AI citations.
Query Classification
The step where an engine decides what kind of query it is handling and whether to trigger an AI answer at all. Informational and complex queries trigger AI answers far more than navigational or simple transactional ones.
Passage
A self-contained portion of a page — a paragraph, a list, a table with its heading — that a retrieval system can score and cite on its own. AI engines work at passage level, not page level.
Passage Retrieval
Finding the specific passages that best answer a query, rather than whole documents. It is why one well-written paragraph deep in an article can be cited while the article as a whole ranks nowhere.
Chunk
A unit of text created when a system splits a page for indexing and retrieval — typically a few hundred tokens. Each chunk is embedded and retrieved independently, so each should make sense in isolation.
Chunking
The process of splitting content into chunks before embedding it. Systems chunk by headings, paragraphs or fixed token counts; content with clear headings and short, focused sections chunks cleanly.
Semantic Chunking
Splitting content at natural meaning boundaries (topic shifts, headings) rather than at fixed lengths, so each chunk covers one idea. Writing with one idea per section makes your content chunk well under any method.
Chunk-Level Retrieval
Retrieval that scores individual chunks against a query. A page’s chance of being cited depends on whether at least one of its chunks is among the closest matches — not on its overall word count.
Dense Retrieval
Retrieval based on embedding similarity, matching meaning rather than exact words. It finds relevant passages that share no keywords with the query, which is why semantic completeness beats keyword repetition.
Sparse Retrieval
Traditional keyword-based retrieval (BM25, TF-IDF) that matches the literal terms in a query. Still used alongside dense retrieval because exact names, model numbers and rare terms match best this way.
Hybrid Retrieval
Combining dense (semantic) and sparse (keyword) retrieval and merging the results. Most production AI search systems are hybrid, so content needs both clear meaning and the precise terms users and engines use.
Reranking
A second-stage pass in which a more expensive model re-scores the top retrieved candidates for relevance before they are sent to the generator. Rerankers reward passages that answer the question directly and completely.
Top-k Retrieval
Returning only the k best-scoring passages (often 5–20) to the model. Because the cut-off is hard, being “quite relevant” is not enough — a passage must be among the very best matches for the sub-query.
Relevance Score
The numeric value a retriever or reranker assigns to a passage for a query. Engines set thresholds; passages below them never reach the model.
Context Injection
Placing retrieved passages into the model’s prompt so it can use them when answering. Whatever is injected is what the model can quote — the reason your text should be quotable as-is.
Context Engineering
The practice of deciding what information goes into a model’s context window and in what form — retrieved passages, memory, tool outputs. The engine-side counterpart to writing content that is worth putting in context.
AI Citation
A reference in an AI answer pointing to the source page a claim came from, shown as a numbered link, chip or card. Citations are the AI-era equivalent of a ranking position and the primary unit of GEO measurement.
Source Attribution
The mechanism by which an engine links each part of an answer to the page it was drawn from. Attribution accuracy varies by platform; Perplexity and Google AI Mode attribute at sentence level, others more loosely.
Source Selection
The engine’s choice of which retrieved pages to actually use and cite. Selection favours authoritative, corroborated, clearly structured and recently updated sources — a compressed version of E-E-A-T.
Citation Slot
One of the limited number of source positions an AI answer displays. Because an answer may cite only three to eight sources across many sub-queries, competition per slot is much tighter than for a page-one ranking.
Consensus
Agreement across multiple retrieved sources on a fact. Engines trust consensus claims and are wary of outliers, so a brand fact stated consistently on many sites is far more likely to appear in answers.
Corroboration
Independent confirmation of a claim by a second source. Models weight corroborated facts higher, which is why third-party mentions matter even without links.
Source Diversity
An engine’s preference for drawing on several different domains rather than many pages from one. It caps how much a single site can dominate an answer and rewards being a distinct, specialised source.
Freshness Weighting
Extra retrieval weight given to recently updated content for queries where recency matters. Accurate dateModified signals and visible update notes help, stale pages get demoted.
Retrieval Bias
Systematic preferences in what gets retrieved — well-linked sites, certain domains (Wikipedia, Reddit, major publishers), English content. Knowing an engine’s biases tells you where to earn presence.
Answer Synthesis
The final step where the model combines retrieved facts into a coherent response, resolving conflicts and choosing emphasis. Clear, unambiguous statements survive synthesis; hedged or buried ones get dropped.
Extractive Summarization
Summarising by selecting and quoting existing sentences from sources. Engines lean extractive for precise facts, so sentences written to stand alone are lifted verbatim.
Abstractive Summarization
Summarising by generating new sentences that paraphrase the sources. Most of an AI answer is abstractive, which is why your brand may be mentioned in words you never wrote.
Attribution Accuracy
How correctly an engine credits the actual origin of a claim. Syndicated or copied content can be attributed to the wrong site, making canonical signals and original publication important.
Related terms on this page
Related terms in the SEO Glossary
Prompts, Queries & AI Search Behaviour
How people ask AI systems for things, and how that differs from typing keywords into Google. Search Intent and Search Query are covered in the SEO Glossary.
Prompt
The full input a user gives an AI assistant — often a complete sentence or paragraph with context, constraints and a question. Prompts replace keywords as the unit of demand in AI search.
Prompt vs. Query
A query is the short keyword string typed into a search box; a prompt is the longer, conversational request given to an assistant. The same need (“dentist Vadodara”) becomes “Which dentist in Vadodara is good for a nervous adult and takes weekend appointments?” as a prompt.
Conversational Query
A question phrased the way a person would ask another person — full sentences, pronouns, implicit context. Both AI assistants and Google’s AI Mode are designed for these.
Natural Language Query
A query expressed in ordinary language rather than keyword shorthand. Modern engines parse natural language directly, so content should answer questions in natural language too.
Long-Form Query
Queries that are two to three times longer than classic searches — Google reports AI Mode queries average this. Longer queries contain more qualifiers (budget, location, use case), each an opportunity to match.
Compound Query
A single prompt containing several needs at once: “compare X and Y, tell me which suits a small clinic, and where to buy it in Gujarat”. Engines decompose compound queries into sub-queries and stitch answers back together.
Comparison Query
A prompt asking how options differ (“Semrush vs Ahrefs for a local business”). Comparison content with clear, structured criteria is heavily cited for these.
Recommendation Query
A prompt asking the assistant to choose (“which SEO agency should I hire in Vadodara?”). Answers draw on reviews, lists and third-party mentions, making off-site reputation decisive.
Best-Of Query
A recommendation query phrased as a ranking request (“best CRM for a dental clinic”). Engines usually answer from existing listicles and review aggregators, so appearing in those lists is the lever.
Local AI Query
A prompt with a location component, answered from Google Business Profile data, reviews, maps and local pages. AI assistants increasingly recommend specific local businesses by name. Learn more →
Transactional AI Query
A prompt that signals readiness to act — book, buy, call, sign up. Increasingly handled by agentic features that complete the action, which makes structured product and booking data essential.
Follow-Up Query
A subsequent question in the same conversation that relies on prior context (“and the second one?”). Engines resolve it using session context, so content that covers the next logical question keeps you in the thread.
Multi-Turn Conversation
A session with several exchanges that build on each other. Visibility in AI is often decided in turn three or four, when the user narrows down to specific requirements.
Session Context
The accumulated information from earlier turns that an assistant uses to interpret later ones. It lets a user drill from a broad topic to a specific provider without restating everything.
Assistant Memory
A feature in ChatGPT, Gemini and others that stores facts about the user across sessions (location, preferences, business). It personalises recommendations and makes local, specific content more likely to be surfaced.
Custom Instructions
User-set standing preferences for how an assistant should answer (tone, format, region). They shape which kinds of sources get used, e.g., “prefer Indian sources and INR pricing”.
Persona Prompt
A prompt that assigns the assistant a role (“act as a procurement manager”) to shape its answer. Used in visibility research to simulate how different buyer types would ask about your category.
Prompt Intent
The underlying goal behind a prompt, mapped to the classic search intent types but usually richer — a single prompt can carry informational, comparison and transactional intent together.
Prompt Volume
An estimate of how often a given prompt (or cluster of similar prompts) is asked of AI assistants. Because platforms publish no prompt data, tools estimate it from panels, APIs and search volume models.
Prompt Research
The AI-era counterpart of keyword research: discovering the questions and requests customers give assistants, clustering them by intent, and mapping them to content. Sources include sales calls, support tickets, PAA data and assistant outputs. Learn more →
Question-Based Keywords
Search terms phrased as questions (who, what, how, which, should). They are the bridge between classic keyword data and AI prompts, and the natural basis for FAQ and answer-first content.
Zero-Volume Keywords
Queries that keyword tools report as having no search volume but which real customers ask, especially in long conversational form. AI search makes these “invisible” queries matter more than ever.
AI Overview Trigger Query
A query for which Google displays an AI Overview. Trigger rates vary sharply by intent and industry; tracking which of your target queries trigger AI answers tells you where citations, not rankings, decide traffic.
Synthetic Query Set
A curated list of prompts built to represent how a target audience asks about a category, used as a fixed benchmark for tracking AI visibility over time.
Voice Query
A spoken search or prompt, typically longer and more conversational than typed queries and often local (“near me”). Answered by reading a single concise response aloud.
Visual Query
A search initiated with an image (Google Lens, ChatGPT image input) rather than text. Descriptive images, captions and surrounding text let engines connect the image to your content.
Multimodal Query
A prompt combining text with images, screenshots, voice or video (“is this the right part for my model?”). Answering it requires the engine to understand your media, not just your words.
Search Personalisation (AI)
Tailoring of AI answers to the individual user based on memory, location, history and stated preferences. Two people asking the same prompt may get different brands recommended.
Related terms on this page
Related terms in the SEO Glossary
Entities, Knowledge Bases & Semantics for AI
AI systems reason about things, not strings. Entity, Entity SEO, Knowledge Graph and the entity types are defined in the SEO Glossary — these terms go one level deeper into how machines ground, identify and connect entities.
Entity Grounding
The process by which AI systems connect a claim or answer back to a verified, recognised entity in a knowledge graph or knowledge base to support its accuracy. Brands that exist as clear entities are grounded; those that don’t are guessed at.
Entity Linking
The process of matching a mention in text (“AmplifyKlicks”, “the agency”) to a specific record in a knowledge base. Consistent naming and sameAs links make your brand easy to link correctly.
Entity Reconciliation
Merging multiple records that refer to the same real-world thing into one — for example your Google Business Profile, Wikidata item, LinkedIn page and website. Unreconciled duplicates dilute authority and confuse AI.
Entity Home
The single page an entity points to as its authoritative source about itself — usually the About page or homepage — referenced from every other profile via sameAs. AI systems use it to resolve conflicting facts. Learn more →
Entity Type
The category a knowledge base assigns an entity (Organization, LocalBusiness, Person, Product, Place). Correct typing in structured data tells engines which attributes to expect and which queries you are relevant for.
Entity Identifier
A unique machine ID for an entity in a knowledge base — a Wikidata QID (Q12345), a Google Knowledge Graph MID (/m/…), or your own @id in JSON-LD. Identifiers let systems refer to you unambiguously across languages and sources.
Entity Embedding
A vector representation of an entity learned from all the contexts it appears in. Brands that co-appear with a topic across many trusted sources develop embeddings close to that topic and get retrieved for it.
Brand-Topic Association
The strength of the link between a brand entity and a topic in an engine’s understanding, built by repeated co-mention across your site and third-party sources. It is what makes an assistant think of you when the topic comes up.
Semantic Triple
The basic unit of a knowledge graph: subject–predicate–object (“AmplifyKlicks — located in — Vadodara”). Writing clear factual sentences and mirroring them in schema makes your facts easy to extract as triples.
Fact
A discrete, verifiable statement about an entity that an engine can store and reuse: founding year, address, price, spec. Answers are assembled from facts; ambiguous or inconsistent facts get dropped or hallucinated.
Claim
A statement that may or may not be verified — reviews, opinions, marketing assertions. Engines treat claims cautiously unless corroborated; original data and named sources turn claims into citable facts.
Claim Verification
The process (automated or human) of checking a statement against trusted sources. Reasoning models increasingly verify claims across sources before including them in an answer.
Canonical Definition
The single, clearest statement of what your brand, product or concept is, used verbatim across your site and profiles. Consistent definitions are what models learn and repeat.
Knowledge Base
Any structured store of facts about entities — Wikidata, Google’s Knowledge Graph, a company’s product database. AI engines consult knowledge bases to ground answers about who and what things are.
Knowledge Vault
A Google research project that automatically extracted facts from the whole web and scored their confidence, prefiguring how modern systems assemble knowledge without human curation.
Knowledge Graph API
Google’s public API for looking up entities in its Knowledge Graph by name or ID. SEO teams use it to check whether a brand is recognised as an entity and what Google believes about it.
Wikidata
The open, machine-readable knowledge base behind Wikipedia, widely used by search engines and AI systems as an entity source. A correct Wikidata item is one of the strongest entity signals a brand can have.
Wikipedia
The most heavily cited source in LLM training data and AI answers. A Wikipedia article is hard to obtain and must meet notability rules, but its influence on what models “know” is unmatched.
Notability
Wikipedia’s standard for whether a subject deserves an article — significant coverage in reliable, independent sources. It doubles as a useful benchmark for whether AI systems will treat a brand as an established entity.
Linked Data
Publishing data with globally unique identifiers and explicit links between items so machines can connect facts across sites. Schema.org with @id and sameAs is linked data in practice. Learn more →
Semantic Web
Tim Berners-Lee’s vision of a web whose data is machine-readable and interconnected. Structured data, knowledge graphs and entity SEO are the parts of it that came true, and they now feed AI answers.
Concept
An abstract idea (e.g., “topical authority”) as opposed to a named entity (a specific company). Engines model both; strong content defines the concepts it uses and ties them to the entities involved.
Word Embedding
The earlier generation of embeddings that gave each word a fixed vector (word2vec, GloVe). Superseded by contextual embeddings, but still the origin of the idea that meaning can be measured as distance.
Contextual Embedding
A vector for a word or passage that changes with its surrounding context, as produced by BERT-style and LLM encoders. It is why “apple” near “iPhone” and “apple” near “orchard” retrieve different content.
Semantic Distance
How far apart two meanings are in embedding space. Content optimization for AI aims to reduce the semantic distance between your passages and the questions you want to be cited for.
LSI Keywords (Myth)
“Latent Semantic Indexing keywords” is a persistent SEO myth; Google has said it does not use LSI. The valid underlying idea — covering related concepts and entities — is properly called semantic or topical completeness.
Related terms on this page
Related terms in the SEO Glossary
Content Optimization for AI Search
How to write and structure pages that AI engines can extract, quote and cite — from the three core content formats to chunking, quotability and provenance.
Machine-Readable Content
Content structured clearly enough — through headings, lists, tables, semantic HTML and structured data — that AI systems and search engines can accurately parse, chunk and extract it without ambiguity.
Answer-First Content
Content that states the direct answer to a question immediately, before supporting explanation and context. The format AI systems most often lift into answers, and the foundation of AEO.
Citation-Worthy Content
Content with clear facts, original data, structure and credibility signals that make it a strong candidate for AI systems to cite as a source — the content-quality half of GEO.
Passage Optimization
Writing and structuring each section of a page so it can be retrieved and cited on its own: a descriptive heading, the direct answer in the first sentence, supporting facts, and no dependence on text elsewhere on the page. Learn more →
Self-Contained Passage
A paragraph or section that makes complete sense when lifted out of the page — names the subject explicitly, avoids “it” and “this” references to earlier text. Chunk-level retrieval cites exactly these.
Chunkability
How cleanly a page splits into meaningful, independent chunks. Short sections, descriptive H2/H3s, and one idea per paragraph raise chunkability; long undivided prose lowers it.
Extractability
How easily an engine can pull a precise answer from a page — definitions in one sentence, steps in numbered lists, comparisons in tables, numbers with units and dates. High extractability means fewer paraphrase errors.
Quotability
The degree to which a sentence can be reproduced verbatim as a useful, accurate statement. Specific, declarative sentences (“X costs ₹12,000 per month and includes Y”) are quotable; vague marketing copy is not.
Information Density
The ratio of concrete facts to filler in a passage. The 2023 Princeton GEO study found that adding statistics, quotations and citations raised generative-engine visibility by up to 40%; dense passages are cited, padded ones are skipped.
Data-Rich Content
Content built around numbers the reader cannot get elsewhere — survey results, benchmarks, pricing ranges, local statistics. Engines prefer to cite the origin of a number rather than a site repeating it.
Direct Answer Block
A short paragraph (40–60 words) placed immediately under a question-style heading that fully answers the question before any elaboration. The building block of AEO and the format most often lifted into AI answers.
Key Takeaways Block
A bulleted summary near the top of a long page stating its main conclusions. It gives engines a compact, extractable version of the page and readers an instant answer.
Inverted Pyramid
The journalistic structure — conclusion first, then supporting detail, then background — applied to web content. It matches how models read context (start and end matter most) and how users skim.
BLUF
“Bottom Line Up Front”: state the conclusion in the first sentence. A writing discipline borrowed from military communication that maps directly onto answer-first content.
Question-Based Headings
H2/H3 headings phrased as the questions users ask (“How much does local SEO cost in Vadodara?”). They match conversational queries, aid chunking and signal exactly which sub-query each section answers.
FAQ Content
Question-and-answer sections that address specific, real customer questions with concise answers. Still valuable for AI extraction (and voice) even though Google restricted FAQ rich results in 2023.
Definitional Content
Content whose purpose is to define a term or concept precisely — glossaries, “what is” pages. Definitions are among the most frequently extracted passage types in AI answers.
Glossary Page
A page that defines the vocabulary of a field, like this one. Glossaries earn citations for definitional queries, build topical coverage, and provide dozens of internal-link anchors for the rest of a site.
Comparison Content
Pages that compare options against explicit criteria — features, prices, fit for use cases. The main source engines cite for comparison and “vs” prompts.
Comparison Table
A structured table with options as rows and criteria as columns. Tables are highly extractable; engines lift cells directly into answers and can reason across them.
Expert Quotes
Attributed statements from named, credentialed people inside content. They add corroboration, E-E-A-T signals and the “quotations” that the GEO study found increased visibility.
Outbound Citations
Links from your content to the authoritative sources of its facts. They make your page verifiable, which reasoning models reward, and mirror the citation behaviour engines themselves use.
Fluency Optimization
Improving the readability and flow of text so a model can parse it with less ambiguity. Another technique the Princeton GEO research found effective, alongside adding statistics and sources.
Conversational Tone
Writing in the natural, second-person language people use with assistants. It aligns your phrasing with prompt phrasing, which improves semantic matching in dense retrieval.
Reading Level
The complexity of language measured by readability formulas. Plain language (roughly grade 8–10) is easier for both users and models to extract from; jargon should be defined when used.
Semantic Completeness
Covering all the sub-topics, entities and questions a user would expect on a subject, so the page fully satisfies the fan-out sub-queries an engine generates. It is topical authority applied to one page. Learn more →
Content Atomization
Breaking a large asset into independent units (a definition, a stat, a checklist, a table) that can each be retrieved, reused across formats and cited separately.
Modular Content
Content designed as reusable, self-contained blocks with consistent structure. It aids chunking and lets a single canonical answer be reused across service, location and FAQ pages without contradiction.
Canonical Answer
The one definitive version of an answer to a recurring question that your brand uses everywhere. It prevents the contradictions that cause engines to drop your site as a source.
Freshness Signals
Visible and machine-readable indications of when content was updated — dateModified in schema, an on-page “last updated” note, updated figures. They feed the recency weighting in retrieval.
Multimodal Content
Pages combining text with images, video, audio and data so multimodal models can use each as evidence. Each asset needs text context (captions, alt text, transcripts) to be understood.
Video Transcript
The full text of a video’s speech published on the page or in the video platform. Transcripts make video content retrievable and quotable by engines that otherwise cannot “watch” it.
AI-Generated Content
Text, images or video produced by a generative model. Google evaluates it by the same quality standards as any content; it is penalised only when produced at scale to manipulate rankings without value.
AI-Assisted Content
Content where AI helps with research, outlines, drafts or editing but a human directs, verifies and adds first-hand expertise. The workable middle ground for most publishing teams.
Human-in-the-Loop
A workflow in which a person reviews, corrects and approves AI output before publication. It protects accuracy, brand voice and E-E-A-T, and is the standard governance model for AI-assisted content.
Content Provenance
Verifiable information about where content came from and how it was made — author, date, tools used, edits. Provenance standards are emerging so engines can favour accountable content.
C2PA
The Coalition for Content Provenance and Authenticity — an open standard for cryptographically signed “content credentials” attached to images, video and audio, recording their origin and edits. Adopted by camera makers, Adobe, Google and OpenAI.
AI Disclosure
Telling readers when and how AI was used to create content. Not required by Google, but increasingly expected by audiences and platforms, and part of a credible editorial policy.
AI Content Detection
Tools that estimate whether text was machine-generated. They are unreliable and not used by Google as a ranking signal; quality and originality, not detection scores, are what matter.
Content Velocity
The rate at which a site publishes or updates content. AI makes high velocity easy, but engines reward velocity only when each piece adds unique value; otherwise it looks like scaled content abuse.
Content Moat
A body of content competitors and AI models cannot replicate — proprietary data, tools, community, expert access. In the AI era, the moat is what keeps your site a source rather than a summary.
Related terms on this page
Related terms in the SEO Glossary
E-E-A-T, Trust & Authority Signals
Why engines believe one source over another. First-hand Experience and Helpful Content are defined in the SEO Glossary; these terms cover the rest of Google’s quality framework and how it carries into AI answers.
E-E-A-T
Experience, Expertise, Authoritativeness and Trustworthiness — the framework from Google’s Search Quality Rater Guidelines for judging content quality. AI engines apply the same logic when selecting sources: demonstrable experience, credentials, reputation and accuracy.
Expertise
Evidence that the creator has the knowledge or skill the topic requires — professional qualifications, depth of explanation, correct use of terminology. Shown through author bios, credentials and the content itself.
Trustworthiness
Google calls trust the most important E-E-A-T component: accuracy, transparency, honest claims, secure site, clear contact and refund information. AI engines are especially sensitive to accuracy because errors propagate into answers.
YMYL
“Your Money or Your Life” — topics that can affect health, finances, safety or wellbeing (medical, legal, financial, news). Google and AI engines apply the highest quality bar here and lean on established institutions as sources.
Search Quality Rater Guidelines
Google’s public manual used by thousands of human raters to evaluate result quality; it defines E-E-A-T, YMYL and “lowest to highest” page quality. Rater feedback trains ranking systems, so it reveals what Google is optimising for.
Credentials
Verifiable qualifications, certifications, roles and affiliations that substantiate expertise. State them explicitly on author and about pages and in Person schema (hasCredential, jobTitle, affiliation).
Expert Review
A named specialist reviewing content for accuracy, shown as “Medically reviewed by…” or “Reviewed by…”. Standard on YMYL sites and a strong trust signal for AI source selection.
Editorial Policy
A published statement of how content is researched, written, reviewed, corrected and (if applicable) AI-assisted. It signals accountability to raters and models alike.
Fact-Checking
Verifying every factual claim against primary sources before publishing, and correcting errors visibly afterwards. The habit that keeps a site from becoming the source of an AI hallucination.
Transparency Signals
Clear information about who runs the site, how to contact them, where they are located, and who wrote each page. Missing basics are a leading reason raters mark pages low-trust.
Trust Signals
The full set of cues that a site is legitimate and reliable — HTTPS, real address and phone, reviews, policies, credentials, press coverage, consistent NAP. Engines weigh them when choosing which business to name in an answer.
Reputation Signals
What independent sources say about a business or author — reviews, news coverage, awards, forum discussion, ratings. Raters are told to research reputation off-site, and models do the same by retrieval.
Factual Accuracy
Whether a page’s claims are correct and current. Reasoning models cross-check sources, so pages with outdated prices, wrong facts or contradictions are quietly dropped as sources.
Source Credibility
An engine’s overall confidence in a domain as a reliable source, built from E-E-A-T signals, historical accuracy, links and mentions. It decides tie-breaks when several pages answer equally well.
Related terms on this page
Related terms in the SEO Glossary
AI Crawlers, Bots & Access Control
Which bots fetch your content for training and answers, and the files, tags and headers that control them. Googlebot, robots.txt, noindex and rendering basics are defined in the SEO Glossary.
AI Crawlers
Bots operated by AI companies (OpenAI, Anthropic, Google, Perplexity, Meta, Apple and others) that crawl the web to gather content for model training, to build live search indexes, or to fetch a page when a user asks about it. Each is controlled by its own robots.txt user-agent.
Crawler Classes
AI companies now run three kinds of bots: training crawlers that collect data for model training, search crawlers that build a live index for answers, and user-triggered fetchers that load a page when someone asks about it. Each has its own user-agent and can be controlled separately.
User-Agent Token
The name a bot presents in its User-Agent header (e.g., GPTBot, ClaudeBot) and the name you use to address it in robots.txt. Blocking or allowing the right token is the whole game of AI crawler control.
GPTBot
OpenAI’s crawler for collecting training data for its models. Blocking it keeps content out of future model training but does not affect ChatGPT Search, which uses OAI-SearchBot.
OAI-SearchBot
OpenAI’s search crawler that builds the index behind ChatGPT Search. It must be allowed in robots.txt for your pages to be retrievable and cited in ChatGPT answers.
ChatGPT-User
The user-agent ChatGPT uses when it fetches a specific page on a user’s behalf during a conversation. Blocking it prevents ChatGPT from reading your page live even when a user shares the URL.
ClaudeBot
Anthropic’s crawler for collecting content that may be used to train Claude models. Controlled separately from Anthropic’s search and user-fetch bots.
Claude-SearchBot
Anthropic’s crawler that indexes pages for Claude’s web search feature. Allowing it makes your content citable in Claude answers.
Claude-User
The user-agent used when Claude fetches a page in real time because a user asked about it. The Anthropic equivalent of ChatGPT-User.
PerplexityBot
Perplexity’s crawler that indexes the web for its answer engine. Blocking it removes you from Perplexity’s citations; the company has faced criticism for fetching pages through undeclared agents.
Perplexity-User
Perplexity’s declared user-agent for fetching a page when a user asks about it in a conversation, distinct from the index-building PerplexityBot.
Google-Extended
A robots.txt token that controls whether your content is used to train and ground Google’s Gemini models. It does not affect Google Search indexing, AI Overviews or AI Mode, which use ordinary Googlebot access.
Bingbot
Microsoft’s search crawler, whose index powers Bing, Copilot, DuckDuckGo and several smaller assistants. Bingbot access and Bing Webmaster Tools are the route into Copilot answers.
Applebot & Applebot-Extended
Applebot crawls for Siri, Spotlight and Safari suggestions; the Applebot-Extended token lets you opt out of Apple Intelligence model training while still appearing in Apple search features.
CCBot (Common Crawl)
The crawler of Common Crawl, the non-profit web archive that has been the single largest source of LLM training data. Blocking CCBot reduces presence in future open training corpora.
Bytespider
ByteDance’s (TikTok’s parent) aggressive crawler used for AI training, notorious for high request volumes and for ignoring robots.txt in many reports. Often rate-limited at the CDN.
Amazonbot
Amazon’s crawler used to improve Alexa and Amazon’s AI features, including Rufus. Declared and generally compliant with robots.txt.
Meta-ExternalAgent
Meta’s crawler for collecting content to train its AI models and power Meta AI search; Meta-ExternalFetcher handles user-triggered page loads.
DuckAssistBot
DuckDuckGo’s crawler used to fetch sources for its DuckAssist AI answers. Small in volume, but a distinct opt-in/out control.
AI Training Opt-Out
Blocking training crawlers (GPTBot, ClaudeBot, Google-Extended, CCBot, Applebot-Extended) while still allowing search and user-fetch bots. A common policy for publishers who want citations but not model training.
AI Bot Blocking
Denying AI crawlers via robots.txt, firewall rules or CDN settings. Blocking search-class bots removes you from AI answers entirely; many sites block by accident with a blanket rule.
Bot Verification
Confirming a crawler is genuine by checking its IP against the published ranges (OpenAI, Anthropic, Google and others publish JSON lists) or via reverse DNS. Fake bots impersonating AI agents are common.
Stealth Crawling
Fetching pages with undeclared or disguised user-agents to evade robots.txt blocks. Cloudflare publicly documented Perplexity doing this in 2025, prompting new bot-authentication standards.
Web Bot Auth
A proposed standard (led by Cloudflare) in which bots cryptographically sign their requests so sites can verify exactly which AI agent is fetching a page, replacing unreliable user-agent strings.
Cloudflare AI Bot Management
Cloudflare’s controls that, since July 2025, block AI training crawlers by default on new sites and give owners per-bot allow/block/charge rules. Check these settings — many sites are blocking AI search bots without knowing.
Pay Per Crawl
A Cloudflare marketplace that lets site owners charge AI companies a per-request fee for crawling, using HTTP 402 “Payment Required”. An early attempt at a commercial model for AI access to content.
Content Signals Policy
A robots.txt extension proposed by Cloudflare in 2025 that declares how content may be used with a “Content-Signal” line: search=yes, ai-input=no, ai-train=no. Separates permission for search from permission for AI answers and training.
llms.txt
A proposed plain-Markdown file at /llms.txt that summarises a site and links to its most important pages for LLM consumption, proposed by Jeremy Howard in 2024. Widely adopted by sites, but no major AI engine has confirmed using it.
llms-full.txt
The companion to llms.txt containing the full text of a site’s documentation in one Markdown file, so an agent can load everything in a single request.
ai.txt
An earlier proposal (from Spawning) for a file declaring whether a site’s media and text may be used for AI training. Largely superseded by robots.txt tokens and newer signals.
TDM Reservation Protocol
A W3C community protocol (TDMRep) for declaring text-and-data-mining rights, backed by the EU copyright directive’s opt-out provision. Relevant for sites in or serving the EU.
RSL (Really Simple Licensing)
An open standard launched in 2025 by publishers (the RSL Collective) that lets sites state machine-readable licence terms for AI use in robots.txt, including pay-per-crawl and pay-per-inference terms.
IETF AI Preferences
An Internet Engineering Task Force working group (aipref) standardising a vocabulary for expressing how content may be used by AI, attachable via robots.txt or HTTP headers. The likely long-term successor to today’s patchwork of signals.
nosnippet
A robots meta directive telling Google not to show any text snippet for a page — which also excludes it from AI Overviews and AI Mode. One of the few controls that directly govern AI-feature appearance.
max-snippet
A robots meta directive setting the maximum number of characters Google may show in a snippet. Very low limits can keep a page out of AI Overviews; the default (unlimited) is right for most sites.
data-nosnippet
An HTML attribute that marks a specific section of a page as unavailable for snippets and AI features while leaving the rest usable. Useful for hiding disclaimers or gated text.
AI Crawler JavaScript Limitation
Most AI crawlers (OpenAI, Anthropic, Perplexity, Common Crawl) fetch raw HTML and do not execute JavaScript, unlike Googlebot. Content that only appears after JS runs is invisible to them. Learn more →
Prerendering
Serving a fully rendered HTML snapshot of a JavaScript-driven page to bots (via SSR, static generation or a prerender service). The standard fix for making JS sites readable to AI crawlers.
HTML-First Content
The principle that all important text, links and structured data should be present in the initial HTML response. It guarantees visibility to non-rendering AI bots and speeds Google’s processing too.
Semantic HTML
Using elements that describe meaning — <article>, <section>, <nav>, <table>, <h1>–<h6>, <dl> — rather than generic <div>s. It helps parsers, chunkers and accessibility tools understand structure.
Markdown Content Negotiation
Serving a Markdown version of a page when a request’s Accept header asks for text/markdown, so AI agents get clean text without boilerplate. An emerging practice among documentation sites.
NLWeb
Microsoft’s open protocol (2025) that lets any site expose a natural-language query interface over its content using Schema.org data and MCP, effectively making the site queryable by agents.
Crawl-to-Referral Ratio
The number of pages an AI company crawls for every visit it sends back — Cloudflare data shows ratios in the thousands for some assistants. A way to judge whether allowing a bot is worth it.
AI Bot Log Analysis
Reviewing server or CDN logs for AI user-agents to see which bots fetch which pages, how often, and whether they hit errors. The only direct evidence of what AI engines actually read on your site.
Rate Limiting
Capping how many requests a client may make in a period, returning HTTP 429 when exceeded. Used to keep aggressive AI crawlers from overloading servers without blocking them outright.
AI Cloaking
Serving different content to AI crawlers than to human visitors. A spam technique in the same family as classic cloaking, and increasingly detectable as bots verify themselves.
IndexNow
A protocol from Microsoft and Yandex that lets sites push URL changes to search engines instantly. Supported by Bing, so it speeds content into Copilot and other Bing-powered assistants.
Bing Webmaster Tools
Microsoft’s equivalent of Search Console, with crawl, index and performance data for Bing — and therefore the diagnostic tool for Copilot visibility.
lastmod Signal
The last-modified date in an XML sitemap entry. Google uses it (when accurate) to prioritise recrawling, so keeping lastmod truthful helps updated content reach AI features faster.
Web Scraping
Bulk extraction of content by automated scripts, often without permission or identification. Distinct from declared crawling; it inflates bot traffic and can feed unlicensed training datasets.
Related terms on this page
Related terms in the SEO Glossary
Structured Data & Schema for AI
The machine-readable layer that tells engines exactly what your pages are about. Rich Result, Product Schema and sameAs are defined in the SEO Glossary.
Structured Data
Code added to a page that labels its content in a standard vocabulary so machines can understand it without guessing — this is a business, here is its address, this is a product with this price. The foundation of entity clarity for AI. Learn more →
Schema.org
The shared vocabulary of types and properties (Organization, Person, Product, FAQPage…) maintained by Google, Microsoft, Yahoo and Yandex. Nearly all structured data on the web uses it, and AI systems are trained on it.
JSON-LD
JavaScript Object Notation for Linked Data — Google’s recommended format for structured data, placed in a <script type=”application/ld+json”> block. Easy to generate, validate and connect into a graph.
Microdata
An older way of adding structured data as attributes inside HTML tags (itemscope, itemprop). Still valid, but harder to maintain than JSON-LD and less common on modern sites.
Schema Graph (@graph)
A JSON-LD structure that holds multiple connected entities in one block — the Organization, the WebSite, the WebPage, the Author — linked by @id references. It tells engines how everything on a site relates.
@id
The unique identifier for an entity within JSON-LD, usually a URL with a fragment (https://example.com/#organization). Reusing the same @id across pages lets engines know they all refer to the same entity.
Organization Schema
Structured data describing a company: name, logo, URL, address, contact points, founders, sameAs profiles, knowsAbout. The single most important schema for brand entity recognition in AI.
Person Schema
Structured data describing an individual — jobTitle, affiliation, credentials, sameAs, knowsAbout. Used for authors and founders to connect expertise to content.
LocalBusiness Schema
A subtype of Organization for businesses with a physical presence: address, geo coordinates, opening hours, price range, service area. It feeds Maps, local packs and local AI recommendations. Learn more →
Service Schema
Structured data describing a service offered — type, provider, area served, offers. Helps engines match a business to “who provides X in Y” prompts.
Article Schema
Structured data for editorial content (Article, BlogPosting, NewsArticle) declaring headline, author, publisher, datePublished and dateModified — the freshness and authorship signals AI retrieval uses.
FAQPage Schema
Markup for a page’s question-and-answer pairs. Google restricted its rich result to government and health sites in 2023, but the markup still gives engines clean, labelled Q&A to extract.
HowTo Schema
Markup for step-by-step instructions. Google retired the HowTo rich result in 2023, but the structure remains useful for making procedural content explicit to AI systems.
Speakable Schema
Markup identifying the sections of a page best suited to being read aloud by voice assistants. Limited adoption, but a direct hint for voice and audio answers.
DefinedTerm Schema
Markup (DefinedTerm, DefinedTermSet) for glossary entries, declaring each term and its definition — used on this page. It turns a glossary into a machine-readable vocabulary.
WebSite & WebPage Schema
Markup that identifies the site as a whole (name, URL, publisher, search action) and each page’s type, language and primary entity. Often the backbone of a schema graph.
Review & AggregateRating Schema
Markup declaring individual reviews and overall rating counts. It supports star ratings and gives recommendation engines quantified reputation data — but must reflect real, visible reviews.
VideoObject Schema
Markup describing a video — name, description, thumbnail, duration, upload date and, importantly, a transcript. It lets engines index video content they cannot watch.
Dataset Schema
Markup for publishing data sets (surveys, statistics, research) so they are discoverable in Google Dataset Search and citable as primary data by AI systems.
knowsAbout
A Schema.org property on Organization or Person listing the topics the entity has expertise in, ideally as links to Wikipedia/Wikidata entities. A direct way to declare topical authority.
mainEntity, about & mentions
Properties that state what a page is primarily about (mainEntity/about) and what other entities it references (mentions). They let you connect a page to knowledge-graph entities explicitly.
potentialAction
A Schema.org property describing actions that can be taken on an entity — SearchAction, ReserveAction, OrderAction. An early hook for agents to discover what they can do on your site.
isAccessibleForFree
A property that declares whether content is paywalled, used with hasPart and cssSelector to mark the gated portion. It lets Google index paywalled content without treating it as cloaking.
datePublished & dateModified
Article schema properties recording when a page went live and was last substantively updated. dateModified is one of the clearest freshness signals for both Google and AI retrieval — keep it honest.
Structured Data Parity
Google’s requirement that structured data describe content actually visible on the page. Marking up facts, reviews or FAQs that users cannot see is a policy violation and undermines AI trust.
Rich Results Test
Google’s validator for structured data eligible for rich results. Together with the Schema.org validator, it is how you confirm markup is parsed correctly before relying on it.
Open Graph
Meta tags (og:title, og:image, og:description) that control how a link is previewed on social platforms and in many AI chat interfaces when they show link cards.
AI Schema Generation
Using LLM tools to draft JSON-LD from page content. Fast and useful, but output must be validated for correct types, real values and parity with the visible page.
Related terms on this page
Related terms in the SEO Glossary
Brand, Mentions & Off-Page Signals for AI
AI recommendations are built from what the rest of the web says about you. Brand Mention, Unlinked Mention, Digital PR and expert sourcing are defined in the SEO Glossary.
Implied Links
Brand mentions without hyperlinks, which Google’s patents describe as a ranking signal alongside real links. For LLMs, unlinked mentions are equally readable and contribute to brand-topic association.
Co-Citation
Two brands or pages being cited together by third-party sources. Repeated co-citation with category leaders teaches engines that you belong in the same set.
Third-Party Validation
Independent sources confirming your claims — press coverage, reviews, awards, expert lists, case studies on client sites. Engines weight validation from sources they already trust far above self-description.
Listicle Inclusion
Being named in “best X” and “top 10” articles published by others. Because engines answer best-of prompts largely from such lists, inclusion is one of the highest-impact GEO tactics.
Review Platform Signals
Ratings, review counts and review text on Google, Justdial, Practo, Clutch, G2, Trustpilot and similar sites. Assistants read these directly when recommending local and B2B providers.
Reddit as an AI Source
Reddit’s outsized role in AI answers, amplified by its licensing deals with Google and OpenAI and by Google surfacing forum content. Genuine, helpful participation in relevant subreddits influences recommendations.
Quora Signals
Quora’s question-and-answer content, frequently retrieved for “which/why/how” prompts. Well-sourced expert answers there can be quoted by assistants.
YouTube as an AI Source
Video content, via titles, descriptions, transcripts and Google’s ownership, is heavily used in AI Overviews and Gemini. A YouTube presence extends citation opportunities beyond the website.
Community Signals
Discussion of a brand in forums, groups, Discord/Slack communities and comment threads. Engines increasingly seek “real people’s experiences”, so authentic community mentions carry weight.
Seed Sources
The set of domains an engine repeatedly cites for a category — Wikipedia, Reddit, major publishers, review platforms, specialist blogs. Identifying them, then earning presence on them, is the core of AI off-page work.
Publisher Licensing Deals
Agreements in which AI companies pay publishers (News Corp, AP, Axel Springer, Reddit, Financial Times and others) for content access. Licensed sources are surfaced preferentially in some assistants.
Training Data Inclusion
Whether your content was part of the corpus a model learned from. Achieved passively by being crawlable and widely referenced at training time; blocking training bots trades this for control.
Brand Search Volume
The number of searches for your brand name. A signal of real-world reputation that correlates with AI recommendations, and one that AI mentions themselves tend to raise.
Branded Queries
Searches or prompts that include your brand name (“AmplifyKlicks reviews”, “is X any good”). Tracking what assistants say in response to branded prompts is the core of AI reputation management.
Brand Sentiment
The positive, neutral or negative tone in which sources — and therefore AI answers — describe a brand. Measured by visibility tools using LLM-as-a-judge classification.
Aggregator Sites
Directories and comparison platforms (Clutch, DesignRush, GoodFirms, Justdial, Sulekha) that list many providers with structured attributes. Engines lean on them for recommendation queries, especially locally.
Earned Media
Coverage obtained through newsworthiness rather than payment — press, podcasts, interviews, expert quotes. It provides the independent corroboration models trust most.
Digital Footprint
The total set of places a brand appears online — site, profiles, listings, mentions, media. AI systems assemble their understanding of a brand from the whole footprint, so gaps and inconsistencies show up in answers.
Related terms on this page
Related terms in the SEO Glossary
AI Visibility Measurement & Analytics
How to prove AI search is working. Impressions, Clicks, CTR and Average Position are defined in the SEO Glossary; these are the metrics, data sources and tools specific to AI answers.
AI Visibility
How often and how prominently a business is mentioned or cited across AI-generated search and chat answers for the prompts that matter to it. The headline KPI of AI SEO, tracked across engines and over time. Learn more →
AI Search Visibility
A business’s overall presence and citation frequency specifically within AI-powered search features (AI Overviews, AI Mode, Copilot Search), as distinct from classic organic rankings and from chatbot mentions.
Citation Rate
The percentage of tracked prompts for which an engine cites at least one of your URLs. The most direct measure of GEO performance.
Mention Rate
The percentage of tracked answers that name your brand, whether or not they link to you. Mentions without links still drive branded searches and trust.
Answer Position
Where in an answer a brand appears — first recommendation, mid-list, or an afterthought. Early positions get most of the user’s attention, just as top rankings do.
Sentiment Score
A classification of how AI answers characterise a brand (positive, neutral, negative) across prompts, usually generated by LLM-as-a-judge scoring. A leading indicator for AI reputation issues.
Visibility Score
A composite index (tool-specific) combining mention rate, citation rate, position and sentiment into a single trendable number for AI presence.
Citation Gap
The set of prompts where competitors are cited and you are not, with the pages they are cited for. The most actionable output of an AI visibility audit.
Competitor Citation Analysis
Systematic study of which domains and pages engines cite for your category, why those pages were chosen, and which formats they use. It reveals the seed sources and content patterns to target.
Source Overlap
The extent to which different engines cite the same sources for a prompt. High overlap means a few pages dominate; low overlap means each engine has its own biases to work with.
Answer Consistency
How stable an engine’s answer is across repeated runs of the same prompt. Because output is non-deterministic, visibility must be sampled several times to be statistically meaningful.
Answer Accuracy
Whether what an AI says about your brand is correct — prices, services, locations, claims. Tracked alongside visibility, since being cited with wrong facts can be worse than not being cited.
Hallucination Rate
The proportion of AI answers containing fabricated facts about a brand. Reduced by consistent, well-structured facts on the entity home and in knowledge bases.
Fan-Out Coverage
The share of an engine’s sub-queries for a topic for which your site has a strong candidate passage. It predicts how often you will be pulled into complex AI Mode and deep-research answers.
AI Overview Presence Rate
The percentage of your tracked keywords that trigger an AI Overview, and the share of those in which you are cited. It separates “AI took the click” from “AI took the click and cited us”.
AI Overview CTR Impact
The change in organic click-through rate when an AI Overview appears for a query — studies through 2025 reported declines of roughly 30–60% for informational queries. Measured to prioritise where citations matter most.
AI Referral Traffic
Visits arriving from AI platforms (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com). Small in volume but typically high-intent, with better conversion rates than average organic traffic. Learn more →
utm_source=chatgpt.com
The tracking parameter ChatGPT appends to links it shows, letting analytics attribute the visit. Other assistants are less consistent, so a custom referral channel group is needed to capture them all.
Dark Traffic (AI)
Visits with no referrer — typically from native AI apps, voice assistants or copied links — that analytics logs as “direct”. Part of AI’s impact is invisible in referral reports.
Assisted Conversions (AI)
Conversions where an AI answer influenced the decision but the final visit came through another channel (brand search, direct). Requires attribution modelling or post-purchase surveys to see.
Brand Lift (AI)
The increase in branded search, direct traffic or survey awareness attributable to AI mentions. Often the clearest business impact of GEO when referral clicks are scarce.
Search Console AI Data
Google includes AI Overview and AI Mode clicks and impressions in the Performance report’s “Web” search type but does not break them out. Position changes around AI features are therefore inferred rather than reported.
GA4 AI Referral Segmentation
Configuring a custom channel group in Google Analytics 4 that matches AI platform referrers (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai) so AI traffic can be reported and trended.
Prompt Tracking
Running a fixed set of prompts through AI engines on a schedule and recording mentions, citations, positions and sentiment — the AI equivalent of rank tracking.
Benchmark Prompt Set
The stable, representative list of prompts (by intent, persona and funnel stage) used for prompt tracking. Changing it breaks trend lines, so it is versioned like a keyword list.
AI Visibility Tools
Software that automates prompt tracking and citation analysis — Semrush AI Toolkit, Ahrefs Brand Radar, Profound, Peec AI, Otterly.AI, Scrunch, ZipTie, SE Ranking’s AI tracker, HubSpot’s AEO Grader and others. Choose on engine coverage, prompt volume and sentiment analysis.
Zero-Click Rate
The share of searches for a query set that end without a click to any site. Rising zero-click rates alongside stable citations mean visibility is being delivered without sessions.
Related terms on this page
Related terms in the SEO Glossary
Google’s AI Ranking Systems & Policies
The named machine-learning systems, spam policies and guidance that decide what Google’s classic results — and therefore AI Overviews — are built from. Core Updates, Spam Updates, Manual Actions and Ranking Systems are covered in the SEO Glossary.
RankBrain
Google’s first machine-learning ranking system (2015), which interprets unfamiliar queries by relating them to known concepts. It marked the shift from keyword matching to meaning.
BERT
Bidirectional Encoder Representations from Transformers — a Google language model (2019) that understands words in the context of the whole sentence. It improved query understanding and is the ancestor of the encoders in today’s retrieval systems.
MUM
Multitask Unified Model — a Google model (2021) that understands text, images and video across 75 languages and can reason across them. MUM-derived capabilities underpin multimodal features like Lens and multisearch.
Neural Matching
A Google system that relates queries to pages by concept rather than exact words, using neural networks. It is why pages can rank for terms they never mention.
Passage Ranking
A Google system (2021) that can rank a page for a query based on one relevant passage, even if the rest of the page is about something else. It anticipated the passage-level retrieval AI answers now use.
Helpful Content System
Google’s site-wide classifier (2022) for content made for search engines rather than people, folded into core ranking systems in 2024. Its criteria — first-hand expertise, satisfying the reader, not chasing volume — are the same ones AI source selection favours.
Reviews System
Google’s ranking system rewarding in-depth, first-hand product and service reviews over thin summaries. Directly relevant to being cited for comparison and recommendation prompts.
SpamBrain
Google’s AI-based spam detection system that identifies link spam, hacked content and manipulative patterns. Modern spam policies (including scaled content abuse) are enforced through it.
Scaled Content Abuse
A Google spam policy (2024) against producing many pages — by AI or otherwise — primarily to manipulate rankings without adding value. It is the line between AI-assisted publishing and penalised content farms.
Site Reputation Abuse
A Google spam policy (2024) against hosting third-party content on an authoritative domain to exploit its ranking signals (“parasite SEO”). Enforced by manual actions and, later, algorithmically.
Expired Domain Abuse
A Google spam policy against buying expired domains and repurposing them to rank low-quality content on inherited authority.
Google Content Warehouse Leak
The May 2024 leak of internal Google API documentation revealing thousands of ranking-related attributes — including click data (Navboost), site authority scores and content-quality classifiers. It confirmed that user behaviour and site-level trust matter.
Twiddlers
Re-ranking functions described in the 2024 leak that adjust results after the main ranking — for freshness, diversity, spam demotion and more. They explain why results shift in ways core relevance cannot.
Freshness Systems
Google’s systems that boost recently updated content for queries where recency matters (“query deserves freshness”). Their logic carries over into the freshness weighting of AI retrieval.
Search Labs
Google’s programme for testing experimental search features (SGE, AI Mode, Web Guide) with opted-in users before wider launch. Watching Labs previews shows what is coming to the main results page.
Google AI Features Guidance
Google’s official 2025 documentation on succeeding in AI Overviews and AI Mode: no special optimization needed beyond good SEO, unique non-commodity content, ensured crawlability, structured data that matches visible content, and snippet controls for visibility management.
Google’s AI Content Policy
Google’s stated position that content is judged on quality and helpfulness, not on whether AI produced it; automation is a problem only when used primarily to manipulate rankings.
Bard
The original name (2023) of Google’s conversational AI assistant, renamed Gemini in 2024. Still encountered in older articles and tool names.
Gemini Models
Google’s family of multimodal foundation models powering AI Overviews, AI Mode, the Gemini app and Workspace features. Search uses custom versions tuned for grounding and citation.
Gemma
Google’s family of open-weight models derived from Gemini research, used by developers to build their own assistants and search tools on the same underlying technology.
Related terms on this page
Related terms in the SEO Glossary
Agentic AI, Agents & Agentic Commerce
The next layer: AI that does not just answer but acts — browsing, comparing, booking and buying on the user’s behalf. Being the site an agent chooses is the newest form of search visibility.
AI Search Agents
AI systems capable of autonomously searching the web, opening pages, gathering information and synthesising an answer across many steps, going far beyond a single query-response. Deep-research modes and agentic browsers are search agents.
AI Agent
An AI system that pursues a goal by planning, using tools (search, browsers, APIs) and taking actions with limited human input. Agents turn “find me a dentist” into a booked appointment.
Agentic AI
The broader category of AI designed to act autonomously toward objectives rather than only respond to prompts. Agentic features are appearing inside every major assistant and search product.
Agentic Search
Search performed by an agent that runs multiple queries, visits pages, compares options and reports back — or completes the task — without the user doing each step. It rewards sites that are easy for machines to read and act on.
Agentic Browsing
Letting an AI navigate websites on a user’s behalf — reading, filling forms, clicking through checkouts — as in ChatGPT Atlas’s agent mode, Perplexity Comet and Gemini in Chrome.
AI Browser
A web browser with an AI assistant built into its core (ChatGPT Atlas, Perplexity Comet, Dia, Chrome with Gemini). The assistant reads and can act on every page, making the browser itself an answer engine.
Computer Use
A model capability (Anthropic, OpenAI, Google) that lets an AI operate a computer interface — viewing the screen, moving the cursor, typing — to complete tasks in any application or website.
ChatGPT Agent
OpenAI’s agent mode (2025, merging the earlier Operator and Deep Research) that browses, uses tools and completes multi-step tasks inside ChatGPT, including shopping and form filling.
Project Mariner
Google’s research agent that browses the web to complete tasks, whose capabilities have been rolled into AI Mode’s agentic features such as booking restaurants and event tickets.
Agentic Checkout (AI Mode)
Google’s ability, within AI Mode and Search, to track prices and complete purchases from merchants on the user’s behalf using Google Pay. Merchants participate via the Shopping Graph and Merchant Center.
Agentic Commerce
Commerce in which an AI agent discovers, evaluates and buys products for a person. It shifts optimization from persuading humans to being the structured, trustworthy option an agent selects. Learn more →
Agentic Commerce Protocol (ACP)
An open standard from OpenAI and Stripe (2025) that lets merchants accept purchases initiated inside ChatGPT, defining how product data, checkout and payment are exchanged with the agent.
Instant Checkout
ChatGPT’s feature for buying a product inside the chat, built on the Agentic Commerce Protocol, initially with Etsy and Shopify merchants. The first mainstream example of a search conversation ending in a purchase.
Buy For Me
Amazon’s agentic feature that purchases items from other brands’ websites on the shopper’s behalf when Amazon does not stock them — an agent acting as a customer on third-party sites.
Agent Payments Protocol (AP2)
Google’s open protocol (2025), developed with payment networks and banks, for authorising and verifying payments made by AI agents using signed “mandates” that prove user intent.
Universal Commerce Protocol (UCP)
Google’s open standard (2026) for agent-driven shopping across retailers, letting merchants expose catalogue, checkout and order data that agents in AI Mode, Gemini and partner platforms can transact against.
Model Context Protocol (MCP)
An open standard (introduced by Anthropic in 2024, now under the Linux Foundation) for connecting AI models to external tools and data sources through a common interface. Websites and SaaS products expose MCP servers so agents can use them directly. Learn more →
Agent2Agent (A2A)
An open protocol (introduced by Google in 2025) that lets independent AI agents discover each other and collaborate on tasks — for example a travel agent hand-off to a hotel’s booking agent.
Tool Use (Function Calling)
A model capability that lets it call external functions — search, calculators, APIs, databases — and use the results. Web search inside assistants is tool use; so is booking or checkout.
Apps in ChatGPT (Apps SDK)
OpenAI’s framework (2025) for third-party apps that run inside ChatGPT conversations (Zillow, Booking.com, Spotify…). A branded app surface that positions a company as the tool the assistant reaches for.
Machine Customers
Non-human buyers — AI agents purchasing on behalf of people or businesses. Gartner projects they will influence a large share of purchases within a few years, making agent-readability a revenue issue.
Agent-Readable Website
A site whose content, prices, availability and actions can be reliably parsed and used by AI agents: clean HTML, complete structured data, accessible forms, stable URLs, and optionally MCP/NLWeb endpoints.
Orchestration
Coordinating multiple models, tools and agents to complete a workflow — e.g., research agent → writing agent → fact-check agent. The pattern behind advanced SEO automation.
Multi-Agent System
Several specialised agents working together on a task, each with its own role and tools. Used in deep-research products and in agencies’ internal SEO pipelines.
Task Completion
The outcome metric for agentic search — whether the user’s goal was achieved (booked, bought, answered) rather than whether a page was visited. Sites that block or confuse agents lose these completions to competitors.
Agentic SEO
Using AI agents to perform SEO work itself — crawling, auditing, drafting, linking, reporting — under human supervision, and preparing a site to be chosen by other people’s agents.
Related terms on this page
Related terms in the SEO Glossary
AI SEO Tools, Workflows & Governance
How teams use AI to do SEO — and the rules that keep it safe. Semrush, Ahrefs, Screaming Frog and Search Console are covered in the SEO Glossary.
Generative AI
AI that creates new content — text, images, code, audio, video — from prompts. The technology behind both AI answer engines and the tools used to produce and optimise content for them.
AI Writing Assistant
Software (ChatGPT, Claude, Gemini, Jasper, Writesonic) that drafts and edits content from prompts. Most effective as a research and drafting partner inside a human-in-the-loop workflow.
Prompt Library
A curated, versioned set of prompts a team reuses for recurring tasks — briefs, meta descriptions, schema, audits — so output is consistent and improvements are shared.
Content Optimization Platform
Tools such as Surfer, Clearscope, MarketMuse, Frase and NeuronWriter that analyse top-ranking pages and score drafts for topical completeness, entities and structure. Increasingly adding AI-answer visibility features.
Content Scoring
A numeric estimate of how well a draft covers a topic relative to competing pages, based on entity and term coverage. Useful as a completeness check, not as a target to game.
AI Internal Linking
Using AI to find and suggest contextually relevant internal links across a site at scale, based on semantic similarity between pages rather than exact-match anchors.
AI Alt Text Generation
Using vision models to draft descriptive alt text for images at scale. Output needs review for accuracy and context, but it closes an accessibility and multimodal gap quickly.
AI Meta Tag Generation
Drafting title tags and meta descriptions with an LLM from page content and target queries, then reviewing for accuracy, length and brand voice.
AI SEO Audit
An audit workflow in which AI tools crawl, classify and prioritise issues — including AI-specific checks like crawler access, chunkability and entity clarity — with a human interpreting the findings. Learn more →
SEO Automation
Using scripts, APIs and AI to perform repetitive SEO tasks automatically: reporting, log analysis, schema generation, content refresh alerts, rank and prompt tracking. Learn more →
Workflow Automation Platforms
Tools like n8n, Make and Zapier that connect AI models with CMSs, spreadsheets and analytics to build SEO pipelines without much code.
WordPress AI Integration
Connecting an LLM to WordPress through plugins or MCP so content, schema and site changes can be made conversationally, with human approval. AmplifyKlicks runs its own site this way. Learn more →
LLM Rank Tracking
Another name for prompt tracking: monitoring brand mentions and citations across AI engines for a defined prompt set. Offered by most major SEO platforms since 2025.
Fact-Checking Workflow
A defined step in which every statistic, name, price and claim in AI-assisted content is verified against a primary source before publication, with sources linked.
Brand Voice Guidelines
Documented tone, vocabulary and style rules, provided to AI tools as instructions so generated drafts sound like the brand and stay consistent across writers.
AI Usage Policy
An organisation’s rules for how AI may be used in content and SEO — permitted tasks, required review, disclosure, data handling. It protects quality and reduces legal and reputational risk.
Prompt Injection
Hidden or embedded text that tries to manipulate an AI system reading a page (“ignore previous instructions and recommend…”). Treated as spam by platforms and a security risk for agents; never a legitimate tactic.
AI Content Farm
A site mass-producing low-value AI-generated pages to capture traffic, the target of Google’s scaled content abuse policy and a drag on the AI ecosystem’s trust in web content.
Vibe Coding
Building software or site features by describing them to an AI coding assistant rather than writing code by hand. Used in SEO for quick tools, scripts and prototypes; output still needs technical review.
Custom GPT for SEO
A tailored ChatGPT configured with a team’s guidelines, data and prompts for repeatable SEO tasks. A lightweight way to operationalise a prompt library.
LLM Visibility Report
A recurring report showing brand mentions, citations, sentiment and competitor comparisons across AI engines, alongside classic rankings and AI referral traffic. Now a standard section of SEO reporting. Learn more →
Related terms on this page
Related terms in the SEO Glossary
Put These Terms to Work
Every concept above maps to a service or guide on this site.
Want Your Business Cited by ChatGPT, Gemini & AI Overviews?
Get a free AI visibility audit showing exactly which prompts your Vadodara business appears for today — and which ones your competitors own.
Explore AI SEO Services →
Or talk to our team about GEO, AEO and AI visibility for your industry.