The glossary of AI visibility, GEO and AI advertising

AI visibility is the degree to which a brand gets mentioned, recommended and cited when AI services like ChatGPT, Gemini and Perplexity answer questions. Where traditional visibility was about position in a list of links, AI visibility is about being part of the answer itself. The whole field is explained in our pillar guide to AI visibility.

GEO, generative engine optimization, is the work of making content visible and citable in AI-generated answers. It covers technical enablement, content structure and authority building aimed at language models rather than classic search engines. The difference between GEO, AEO and SEO is explained in our guide to the three terms.

AEO is optimization for answer engines: Services that answer the question directly instead of showing a list of links. The term arose with featured snippets and voice assistants, and today it is often used about optimizing for AI answers. In practice, AEO and GEO overlap heavily.

SEO is the work of making web pages visible in organic search results, primarily in Google. Solid SEO craft is still the foundation of AI visibility: The models draw much of their knowledge from the same indexes, and a page that is invisible to Google is usually invisible to AI as well.

AI search means search experiences where an AI-generated answer replaces or supplements the classic list of links. Examples are ChatGPT with search, Perplexity, Google's AI Overviews and AI Mode. What they share is that the user gets the answer directly, and sources compete to be cited rather than clicked.

An answer engine is a service that answers questions directly instead of returning a list of links. ChatGPT, Perplexity and Google's AI Mode are answer engines, and they read, interpret and synthesize sources before formulating an answer of their own. How such an answer comes together is explained in our guide to how an AI answer is read.

Generative AI is artificial intelligence that produces new content, such as text, images, code or video, rather than just classifying or sorting existing data. The language models behind ChatGPT, Claude and Gemini are generative AI, and this is the technology driving the entire shift from search results to AI answers.

A language model, often called an LLM for large language model, is an AI model trained on enormous amounts of text to understand and produce language. The model predicts text from context, and that mechanism is what decides which brands get mentioned in an answer. The GPT series, Claude and Gemini are language models.

A prompt is the instruction or question a user gives an AI model. The phrasing of the prompt steers both what the model answers and which sources it surfaces, which is why GEO work analyzes which prompts trigger mentions of your brand and your competitors.

A hallucination is when an AI model presents incorrect information as if it were true, for example products that do not exist or prices that are outdated. For brands, hallucinations are a concrete risk: The model can describe your business wrongly. Clear, updated and citable sources on your own website are the best defense.

RAG is a technique where the language model retrieves fresh sources, typically via search, before formulating its answer. RAG is what lets ChatGPT with search and Perplexity cite current web pages instead of relying on training data alone. For visibility it means both your presence in training data and your searchable sources count.

Training data is the body of text a language model learns from, drawn from web pages, books and licensed sources among others. Content that was available and widely referenced when the model was trained shapes what it "knows" about brands. That is why long-term, consistent presence on the web influences AI answers even without live search.

The knowledge cutoff is the point in time where a model's training data ends. Anything that happened after the cutoff the model only knows through live search. Newer content therefore reaches AI answers via the search indexes, while older content can be baked into the model itself.

ChatGPT is OpenAI's conversational service and the world's most used AI assistant, with over a billion weekly users in 2026. The service combines a language model with search, and it is both the most important surface for organic AI visibility and the home of the ad product ChatGPT Ads.

Claude is the AI assistant from Anthropic, known for long context windows and strong text comprehension, and widely used in professional and technical environments. Claude fetches web content via its own crawlers, including ClaudeBot, and is one of the surfaces where brands should measure how they are described.

Copilot is Microsoft's AI assistant, integrated into Windows, Bing, Edge and Microsoft 365. It builds on OpenAI models combined with the Bing index, so visibility in Bing affects visibility in Copilot. Copilot has shown ads in its answers from the start, managed through Microsoft Advertising.

Perplexity is an AI search engine that always answers questions with source citations, and it cites more aggressively than most other surfaces. That makes Perplexity a useful place to study which sources win in your category, even where its market reach is still limited.

Gemini is Google's language model family and AI assistant, and the engine behind the AI features in Google Search. Gemini builds on Google's own index and knowledge graph, so classic Google visibility, structured data and entity building feed directly into how Gemini describes your brand.

AI Overviews are the AI summaries at the top of Google's search results, generated by Gemini with links to sources. They reduce clicks to organic results significantly, making it more important to be the source cited in the overview than the blue link below it. Google also shows ads in and around AI Overviews in several markets.

AI Mode is Google's pure conversational search, a separate tab where the whole search experience is a dialogue with Gemini. It represents Google's answer to ChatGPT, and ads are already being tested in the surface. For content owners the same logic applies as in AI Overviews: The sources that get cited win the traffic.

A citation is when an AI answer points to your website as a source, typically with a link. A citation is stronger than a plain mention because it both drives traffic and builds authority in the models' source evaluation. The difference between being mentioned and being cited is explained in our guide to how an AI answer comes together.

A mention is when your brand appears in an AI answer, with or without a link. Mentions are measured across prompts and platforms, and they are the core metric of AI visibility: Do you get named when someone asks for recommendations in your category, and in what context?

Share of voice in an AI context is the share of relevant AI answers where your brand is mentioned, measured against competitors. If the model names you in three out of ten relevant prompts and a competitor in five, you have 30 percent and they have 50. The number makes AI visibility comparable over time and between players.

A visibility score is a composite measure of how visible a brand is in AI answers, calculated from factors like mention frequency, position in the answer, sentiment and citations. The score makes progress measurable month over month, and it is among the core metrics in the Synlig.ai platform.

AI traffic is website visits coming from AI services, for example when a user clicks a citation in ChatGPT or Perplexity. It is identified in analytics tools via referral sources like chatgpt.com and perplexity.ai, and it often converts unusually well because the visitor arrives pre-qualified from a conversation.

Referral traffic is visits arriving via links from other websites rather than search or direct entry. AI traffic is reported as referral traffic in Google Analytics, and a dedicated channel group for AI sources makes it easy to follow separately.

Branded search is a search where the user types your name, for example "Synlig.ai" instead of "GEO tools". Many who see a brand in an AI answer or AI ad look it up afterwards instead of clicking. Growth in branded search is therefore an important indirect effect of AI visibility that otherwise hides as organic traffic.

A zero-click search is a search where the user gets the answer without clicking any result, for example in an AI overview. The share of zero-click searches grows in step with AI answers, and the prize moves from the click to being the source and the recommendation inside the answer itself.

An AI crawler is a bot that fetches web content for the AI companies' models and search features. The most important ones are GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot and Google-Extended. Block them in robots.txt and you disappear from the source pool; the setup is explained in our guide to technical SEO for AI search.

GPTBot is OpenAI's crawler for collecting training data for future models. If you allow GPTBot in robots.txt, your content can become part of what the models "know" about your industry. It is separate from OAI-SearchBot, which powers live search, and the two can be controlled independently.

OAI-SearchBot is OpenAI's search crawler, the bot that makes web pages visible when ChatGPT searches and cites in real time. It must be allowed in robots.txt for you to be citable in ChatGPT answers, and OpenAI also requires access for OAI-SearchBot and OAI-AdsBot on landing pages used in ChatGPT Ads.

ClaudeBot is Anthropic's crawler, collecting content for the Claude models. As with the other AI crawlers, access is controlled in robots.txt, and an open setup is a prerequisite for being visible in Claude's answers and source references.

robots.txt is a text file at the root of a website telling crawlers what they may and may not fetch. It has gained renewed strategic importance because it now also governs the AI crawlers: An overly strict robots.txt can effectively remove a business from AI answers.

llms.txt is a proposed standard for a file that gives language models a clean, prioritized overview of a website's most important content, placed at /llms.txt. The idea is to help models find and understand the core content without interpreting the whole site. The standard is young, but cheap to adopt.

Structured data is machine-readable markup that tells search engines and AI models explicitly what a piece of content is: An article, a product, an FAQ, an organization. It reduces the room for misinterpretation and increases the chance of correct citation, and it is covered in our guide to structured data for AI.

Schema.org is the shared vocabulary for structured data, developed jointly by the search engines. It defines the types and fields content is marked up with, such as Article, Organization, Person and FAQPage, and it is the standard both Google and the AI models understand.

JSON-LD is the format structured data is usually written in: A JSON block in the page's HTML head, separate from the visible content. Google recommends JSON-LD over other formats, and it is how schema.org markup is implemented in practice on modern websites.

FAQ schema is structured data of the FAQPage type, marking up questions and answers on a page explicitly. The format suits AI answers exceptionally well, since the models themselves work in questions and answers, and a good FAQ block increases the chance of being used as a source for question-shaped prompts.

A canonical URL tells search engines which version of a page is the original when the same content exists at several addresses. It consolidates signals at one address and prevents duplicates from competing with each other, which also keeps the source picture clean for AI models.

Hreflang is markup telling search engines which language versions of a page belong together, and which one to show to which users. For multilingual websites it ensures the right language version appears in both search and AI answers.

Indexing is the process where a search engine takes a page into its database so it can appear in results. A page that is not indexed does not exist for search, and therefore not for AI features built on the index either. Fast, error-free indexing is the foundation of all visibility.

An embedding is a numerical representation of text that captures the meaning of the content, letting machines compute similarity between texts. Embeddings are the engine of semantic search: They are how a model knows that "tool for AI visibility" and "GEO platform" are about the same thing.

Semantic search matches on meaning instead of exact words. The AI surfaces are thoroughly semantic: They understand synonyms, paraphrases and context, which makes clear, well-structured content more important than keyword repetition. The same mechanism, incidentally, drives the matching of context hints in ChatGPT Ads.

An entity is a uniquely identifiable "thing" in the worldview of search engines and models: A company, person, product or place. Strong entities with consistent information across the web become safer for an AI model to recommend, which is why entity building is core GEO work.

A knowledge graph is the database of entities and the relationships between them, such as Google's Knowledge Graph. It tells the systems that Synlig.ai is a product from SEG Solutions AS, who works there and what the company does. The richer and more consistent the graph is about you, the more precisely the AI surfaces describe you.

E-E-A-T stands for experience, expertise, authoritativeness and trustworthiness. It is Google's framework for content quality, and the AI models reward the same signals: Named authors with real domain backgrounds, first-hand experience, sources and verifiable claims.

A pillar page is the thorough main page on a core topic, the hub every related subpage links up to. The pillar gives both users and models one authoritative place to start, and the structure is explained in our guide to pillar and cluster in practice.

A topic cluster is a pillar page plus the subpages covering each part of the topic, bound together with internal links. The cluster demonstrates breadth and depth in one field in one place, and it is the most effective structure for building topical authority in both search and AI answers.

Topical authority is the status a website earns when it covers a field broadly, deeply and coherently over time. Models and search engines prefer sources with documented authority on the topic over single pages that happen to match, and the authority is built through topic clusters and consistent publishing.

Internal linking is the links between pages on your own website. They show search engines and AI crawlers how the content fits together, which pages matter most and what elaborates on what. Good internal linking is among the cheapest and most underrated GEO measures.

Anchor text is the clickable text of a link. Descriptive anchor texts, like "the guide to technical SEO for AI search" instead of "read more", tell users, search engines and language models what the target page is about, and they strengthen the semantic coherence of the topic cluster.

Search intent is what the user is actually trying to achieve: Learn something, compare options or complete a purchase. In AI conversations the intent surfaces even more clearly than in short searches, and content that matches the intent precisely wins both citations and conversions.

Long-tail keywords are long, specific search phrases with low volume per phrase but high combined volume and clear intent. AI conversations are, in practice, extreme long tail: People ask complete, detailed questions, and content that answers specific questions thoroughly becomes the sources the models reach for.

Evergreen content is content that stays relevant over time, such as guides, definitions and process descriptions. It accumulates authority year after year and keeps getting cited long after publication, unlike news content that decays. A strong GEO strategy builds on an evergreen core with news as seasoning.

A featured snippet is the excerpt Google lifts to the top of the search result as a direct answer to a question. Snippets were the forerunner of AI answers, and the recipe is the same: Concise, self-contained answers high on the page, preferably in lists and tables, with the depth below.

Thin content means pages with little substance: Short texts without depth, auto-generated pages or content duplicating what already exists. Thin content does not get cited by models, drags down the impression of the whole site, and should be strengthened, merged or removed.

Duplicate content is the same or nearly the same content available at several URLs, internally or across websites. It splits signals and creates uncertainty about which version is the original. Canonical URLs and deliberate content structure are the remedy.

A content audit is a systematic review of existing content: What performs, what is outdated and what is missing. In GEO work the audit includes how the content is cited in AI answers, and updating good older pages often delivers faster results than new content.

AI advertising means paid ads in AI surfaces like ChatGPT, AI Overviews and AI Mode, targeted on the conversation's context instead of keywords. The ads sit next to the AI answer, never inside it, and the whole landscape is covered in our pillar guide to AI advertising.

ChatGPT Ads is OpenAI's ad product, where clearly labeled ads appear below ChatGPT answers for users on the free and Go tiers. The ads are managed in OpenAI Ads Manager and targeted with context hints instead of keywords. The full setup is in our step by step guide to ChatGPT Ads.

Google Ads is Google's advertising platform for search, YouTube, the display network and increasingly the AI surfaces: Ads in AI Overviews are drawn from regular search and shopping campaigns. A properly rigged Google Ads account is thus the entry ticket to Google's AI ads, as described in our Google Ads guide.

Performance Max is Google's fully automated campaign type, where machine learning places ads across all of Google's surfaces based on the goals and assets you provide. It is one of the campaign types that qualifies for AI Overviews placement, and it requires solid conversion tracking to work.

Quality Score is Google's 1 to 10 assessment of how relevant a keyword, an ad and a landing page are to each other. Higher quality yields lower click prices and better placements. The principle behind it, alignment between intent, message and landing page, applies just as strongly in AI advertising.

Remarketing is advertising aimed at people who have already visited your website or are customers, based on tracking lists. ChatGPT Ads has no website-based remarketing, and over 80 percent of ad-driven ChatGPT traffic is new customers. Customer lists there are better used to exclude existing customers.

Context hints are free-text descriptions in ChatGPT Ads of which conversations and needs an ad is relevant for, and they replace keywords. They are matched semantically against the whole conversation and guide the system rather than control it. The craft is covered in our guide to context hints.

Custom audiences in ChatGPT Ads are uploaded customer lists with email addresses, phone numbers or advertising IDs, matched against signed-in users and usable for inclusion or exclusion in campaigns. The upload is a privacy decision, and the choices are covered in our guide to custom audiences.

The OpenAI pixel is the tracking code that connects ChatGPT ads to actual conversions on your website. It captures the click ID from the ad link, stores it in first-party cookies and sends conversion events back to Ads Manager. The full setup is in our pixel guide.

A click ID is a unique identifier the ad platform attaches to the link when someone clicks an ad, such as oppref at OpenAI and gclid at Google. The tracking code on the website stores the ID in a cookie, and it is what connects a later conversion back to the right ad. If the parameter gets stripped along the user journey, attribution breaks.

UTM parameters are standardized tracking parameters in URLs, like utm_source and utm_campaign, telling your analytics tool where the traffic came from. They give you your own channel numbers independent of the ad platforms' reports, and a consistent UTM taxonomy is the prerequisite for comparing AI channels with established ones.

A conversion is a completed, valuable action: A booked demo, a submitted form or a purchase. Conversions are the currency all ad optimization counts in, and they should be measured when the action is actually completed, not when someone clicks a button.

Conversion tracking is the setup that registers conversions and ties them to the channel and ad that created them, via pixels, click IDs and events. Without it, neither you nor the platforms can separate profitable campaigns from money pits, and conversion-optimized bidding becomes impossible.

A Conversions API sends conversion events server to server, directly from your systems to the ad platform, instead of via the browser. It is more robust against ad blockers and necessary for conversions happening outside the website, for example in a CRM. Both OpenAI and Google offer such APIs.

Consent Mode is Google's framework for passing the user's consent choice to Google's tags, so measurement and advertising cookies adapt to the consent given. In version 2 it is effectively mandatory for advertisers in the EEA. The same principle applies to every tracking tool: No consent, no marketing cookies.

CPM, cost per mille, is the price for a thousand ad impressions, and the buying model for reach campaigns. In ChatGPT Ads, US CPM levels have ranged from 25 to 60 dollars in the early phase. The buying models are compared in our guide to buying models and budget.

CPC, cost per click, is the price per click on an ad, and the standard model for traffic campaigns. In ChatGPT Ads, OpenAI recommends click bids of 3 to 5 dollars in the early phase, and CPC buying is the natural starting point while conversion data builds up.

oCPC, optimized cost per click, is click-based buying where the system optimizes delivery toward conversions instead of clicks. The model requires working conversion tracking with sufficient data volume, and it is the step you take after CPC once the pixel has collected real conversions.