23 definitions · updated 27 September 2026

Glossary of GEO and AI visibility

GEO jargon explained simply, with what each term means for a business.

GEO (Generative Engine Optimisation)

GEO (Generative Engine Optimisation) covers the practices that help a business get cited, recommended and accurately described in the answers of AI assistants such as ChatGPT, Gemini, Claude or Perplexity.

The term was popularised by a research paper published in late 2023 by researchers from Princeton and IIT Delhi, among others. Where SEO aims for a good position in a list of links, GEO aims for the answer itself: being the name that comes up when a customer asks “which plumber should I call in [town]?”. For a small business the stakes are real, because a user who gets a direct answer doesn’t always click through to the links. GEO relies heavily on the basics of SEO (indexable pages, clear content) and adds work on third-party sources, reviews and consistent information.

AEO (Answer Engine Optimisation)

AEO (Answer Engine Optimisation) means optimising content so that an answer engine picks it up as a direct answer: a “position zero” featured snippet, a voice assistant or an AI assistant.

The term predates GEO: it emerged with Google’s featured snippets and voice assistants. Today, AEO and GEO are often used almost interchangeably. In practice, both rest on the same habits: answer a precise question in the first few lines, with verifiable facts and a clear structure. For a small business, a page that clearly answers “how much does an accountant charge to set up a limited company?” serves both goals.

LLM (large language model)

An LLM (large language model) is an artificial intelligence model trained on vast amounts of text to understand and generate language; it is what powers ChatGPT, Gemini and Claude.

What an LLM knows stops at the cut-off date of its training data. A new business, a change of address or a new price list may therefore be unknown to it, unless the assistant runs a web search when it answers. That’s why there are two levers: being well represented on the web in general, and being findable by the search tools these assistants use in real time.

AI web search (grounding, RAG)

AI web search is the mechanism by which an assistant consults sources in real time before answering, then bases its answer on those documents; this is known as grounding or RAG (retrieval-augmented generation).

ChatGPT decides for itself when to run a search and then shows its sources; Google offers grounding for Gemini on Google Search results. In practice, the assistant writes one or more queries, reads a few pages, then drafts its answer. To be recommended in this mode, your pages must be open to the relevant crawlers, indexed, and state in plain words the information people are looking for (town, services, prices, opening hours). Information that isn’t on the web can’t be picked up.

Citation (cited source)

A citation is the link or reference an AI assistant displays to show where a piece of information in its answer came from.

Being cited and being recommended are not the same thing: your site can be cited for an advice article without your business being recommended, and a competitor can be recommended with a review platform or a directory as its only source. Recording the sources cited on questions in your sector shows you where to be present: Google Business Profile, trade platform, local press, comparison site. It’s often more revealing than looking at your own website alone.

Brand mention

A brand mention is when a business’s name appears in an AI assistant’s answer, with or without a link to its website.

Not all mentions are equal: being named first as a recommendation doesn’t have the same effect as a place at the bottom of a list, or a mention alongside wrong information. Mentions of your brand elsewhere on the web (press, directories, forums, comparison articles) also shape what assistants know about you. For a small business, tracking these mentions shows up both the gaps and the inaccurate descriptions.

AI share of voice

AI share of voice is the proportion of AI assistant answers, across a defined set of questions, in which a brand is mentioned, compared with its competitors.

Fictional example: across 30 questions put to three assistants, an accountancy firm in Newcastle appears in 4 answers while a competitor gets 15. The metric only makes sense if the questions stay the same from one measurement to the next, because AI answers vary with the wording and from day to day. It’s mainly useful for tracking a trend and spotting the competitors who are taking the spot.

AI Overviews (Google)

AI Overviews are AI-generated summaries shown at the top of some Google results pages, with links to the pages that support them.

Google launched AI Overviews and AI Mode in France in July 2026. According to Google’s documentation, there are no technical requirements or special markup needed to appear in them: the page must be indexed and eligible to be shown with a snippet in the standard results. The usual controls (nosnippet, max-snippet, noindex) apply here too. For a small business, SEO fundamentals therefore remain the way in.

AI Mode (Google)

AI Mode is Google’s conversational search mode, which answers complex questions with a written response and links, and lets you ask follow-up questions.

Google says that AI Mode, like AI Overviews, may use a “query fan-out” technique: it runs several searches on sub-topics to put its answer together, which can surface a wider range of sites. Available in France since July 2026, it follows the same eligibility rules as standard search. A page that covers a sub-topic precisely (prices, area served, a specific case) therefore has a chance of being picked.

ChatGPT Search

ChatGPT Search is the web search feature built into ChatGPT, which lets it answer with recent information and links to its sources.

Launched by OpenAI on 31 October 2024 and opened to all users in early 2025, it relies on third-party search providers and partner content. ChatGPT decides on its own whether to search the web, and users can also ask it to. According to OpenAI, a site that blocks the OAI-SearchBot crawler won’t be shown in ChatGPT Search answers, except as a plain navigational link.

Perplexity

Perplexity is an AI-based answer engine that runs a web search for every question and shows its sources alongside the answer.

Perplexity runs two crawlers: PerplexityBot, which indexes sites so they appear in its results (and which, according to the company, isn’t used to train foundation models), and Perplexity-User, which visits a page at a user’s request. Because its sources are so visible, Perplexity is a good place to see which sites carry authority in your sector.

Hallucination

A hallucination is false information produced by an AI and stated confidently as fact.

For a business, that might be the wrong opening hours, a service you don’t offer, an old price or an out-of-date address. Hallucinations are more common when information is scarce, contradictory from one source to another, or missing from the web. Publishing clear, identical information on your website, your Google Business Profile and in directories reduces the risk without removing it; so you need to check regularly what AIs say about you.

Structured data (schema.org)

Structured data is standardised markup, usually in JSON-LD format and based on the schema.org vocabulary, that explicitly describes what a page contains: type of business, address, opening hours, products, prices.

Google recommends the JSON-LD format and points out that valid markup makes a page eligible for rich results without ever guaranteeing them. For a small business, useful types include Organization, LocalBusiness and its variants (Restaurant, Hotel, LegalService…), Product or Service. The markup must match exactly what is visible on the page. How each AI assistant uses this data isn’t documented in detail: it clarifies the facts, but it doesn’t replace clearly written text.

FAQPage

FAQPage is a schema.org type used to mark up a page made up of questions and their answers.

Google stopped showing FAQ rich results in its search engine in May 2026; the markup is still valid schema.org and doesn’t penalise a page, but it no longer triggers any special display. The main value of an FAQ is now its content: questions worded the way your customers ask them, and short, factual answers that assistants can understand and reuse. Ten real questions answered well beat forty generic ones.

llms.txt

llms.txt is a Markdown file placed at the root of a website (/llms.txt) that summarises the site and points language models to its key pages.

It’s a community proposal published in September 2024 by Jeremy Howard, not an official standard. Its effect on whether a business appears in assistants’ answers hasn’t been publicly demonstrated to date. For a small business, it’s a cheap extra that shouldn’t come before the priorities: clear pages, crawlers allowed, Google Business Profile and reviews.

robots.txt and AI crawlers (GPTBot, ClaudeBot…)

The robots.txt file, placed at the root of a website, tells crawlers what they may visit; AI companies run several crawlers with different roles, which you can allow or block separately.

At OpenAI, GPTBot collects content for training, OAI-SearchBot powers ChatGPT Search and ChatGPT-User acts at a user’s request (OpenAI notes that robots.txt may not apply in that case). At Anthropic, ClaudeBot is used for training, Claude-SearchBot for search and Claude-User for user requests. Perplexity uses PerplexityBot and Perplexity-User. A common mistake is to block every “AI bot” in one go, which can make the site vanish from answers that use web search; you can refuse training while still letting search crawlers through.

Google-Extended

Google-Extended is a token you can use in robots.txt to refuse that content crawled by Google is used to train Gemini models and to ground the answers of Gemini apps.

According to Google, this setting has no effect on whether a site appears in Google Search or how it ranks; AI Overviews and AI Mode follow the usual Googlebot rules and snippet controls. However, the documentation states that it also covers grounding in Gemini apps. Blocking Google-Extended is therefore a choice that may reduce how much of your content Gemini reuses: make it knowingly.

E-E-A-T

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness: the qualities Google’s systems try to recognise in content, with trustworthiness the most important.

Google makes clear that E-E-A-T is not a direct ranking factor, but a set of qualities that several signals help assess, with extra weight on health, finance and safety topics. For a small business it comes down to simple things: a named author, verifiable qualifications, real contact details, concrete examples, mentions in independent sources. AI assistant providers don’t publish an equivalent framework, but the same signs of credibility make a business easier to recommend.

Entity and Knowledge Graph

An entity is something that can be identified unambiguously (a business, person, place or product); the Knowledge Graph is Google’s database that links these entities and their attributes, visible for instance in knowledge panels.

Thinking in entities means making sure search engines and assistants understand that the local bakery mentioned on its website, its Google Business Profile and in the local paper is one and the same business. An identical name, an identical address and links between your official profiles (the sameAs property in structured data) all help. Google lets an entity’s official representatives claim its knowledge panel after verification and suggest changes.

Wikidata

Wikidata is a free, collaborative knowledge base run by the Wikimedia Foundation that describes entities (organisations, people, places…) as reusable structured data.

Its data is reused by many services, which makes it a reference point for identifying an entity unambiguously. Wikidata applies notability criteria: an entry must describe a clearly identifiable entity, backed by serious, public sources. For a small business, creating a promotional entry without independent sources is a no-go; a business already covered by the press can, however, check that the information about it is accurate.

Customer reviews

Customer reviews are ratings posted by customers on Google, on sector platforms or on your own website; AI assistants often use them to describe a business’s reputation.

Volume, recency and content all count: a review that describes a specific service gives the AIs something concrete to work with. In France, publishing or commissioning fake reviews is a misleading commercial practice, and a business that displays reviews must say whether and how they are checked. Google also forbids offering anything in exchange for a review, and asking only happy customers. The right approach: ask every customer for a review, simply, and reply to each one.

Prompt and typical question

A prompt is the text you submit to an AI assistant; a typical question is one that represents what your potential customers really ask AIs before choosing a provider.

Example: “which accountant in [town] for a newly formed one-person limited company?” rather than the name of your practice. Answers vary with the wording, the location and the user’s context; a serious measurement therefore relies on a stable set of questions, worded like a real customer would, and asked again at regular intervals. Building that list is also an excellent way to find out which pages are missing from your website.

AI Score

The AI Score is Saykora’s score out of 100 summarising how visible a business is in the answers of ChatGPT, Gemini and Claude, asked with web search on.

It adds up the business’s presence in the answers (up to 60 points) and its website’s technical signals (up to 40 points). The questions are generated from the website (trade, town, country), asked the way a real customer would, then frozen for 90 days so that two measurements remain comparable. The full calculation is set out on our method page.

Sources consulted (18)

To go further: the GEO guide for small businesses.