Every term you need to understand how ChatGPT, Claude, Gemini and Perplexity decide which businesses to recommend, and how yours becomes one of them. Our own concepts alongside the standard language of the field, defined plainly.
A recommendation made to a buyer by an AI assistant such as ChatGPT, Claude, Gemini or Perplexity, as opposed to a link shown by a search engine. The unit of visibility in AI search: being named, not just ranked.
When a buyer asks an AI assistant who to use, the assistant does not hand back ten blue links. It names a short list of specific businesses. Being on that list is an AI referral. It is a different outcome from a search ranking, decided by a different system, and it is increasingly where the choice is made.
Everything else in this glossary is about how AI referrals are won and lost.
The distance between being findable (ranking in a search engine) and being recommended (named by an AI assistant when a buyer asks who to use). Most businesses close the first and never notice the second.
Ranking on Google gets you found. It does not get you recommended. A page can sit on the first page of search and still never be named by ChatGPT, Claude or Gemini when a buyer asks who to use in your category.
The gap between the two is where a growing share of customers now make their decision, and most businesses have not noticed it opening.
Read the full piece →When a buyer asks an AI assistant who to use, receives a short list of businesses, and yours is not on it. The enquiry is decided before anyone reaches your site, so it never appears in your analytics.
A customer was shortlisting suppliers in your category, in your town, at that moment, and you were quietly left off. There is no missed call, no bounced visit, no line in your analytics, because the decision happened inside the assistant before anyone reached you.
It is the quietest way businesses are losing work to AI, and the first time most owners see it is when they run a scan and watch competitors get named while their own business gets a blank.
The compounding advantage a business builds once AI assistants learn to cite it as a source for its category. Each citation makes the next one more likely, so early movers become harder to displace.
AI referrals are not a fixed leaderboard. They feed on themselves. The more the wider web associates you with your category, the more the engines name you, and the more they name you, the more that association hardens.
That is why closing the Recommendation Gap early matters. The businesses that become the cited source first are the ones competitors then have to dig out from underneath.
The three checks an AI assistant runs before it will name a business: Retrieval (can it find and read you), Resolution (can it work out exactly who you are), and Corroboration (does the wider web vouch for you). Fail any one and you are left off the answer.
Most small businesses pass the first gate and fail the third. That is why more schema rarely moves the answer on its own, and why the work that changes who gets recommended usually happens off your own site.
Read the full piece →The first gate: whether an AI assistant can find and read your content at all. Served as real text, crawlable, structured. If the engine cannot retrieve you, nothing else matters.
Retrieval is the technical layer. Content that only appears after JavaScript runs, a robots file that blocks the AI crawlers, or a page an engine cannot parse all fail here. Schema, clean structure and an llms.txt file help. This is the gate most AI SEO advice stops at, and it is necessary, but rarely the reason a business is missing.
The second gate: whether an engine can work out exactly who you are and trust it has the right business. This is entity clarity, a single consistent identity across your site and the profiles you link to.
An assistant that cannot resolve you into a confident, single answer will leave you out rather than risk naming the wrong business. Stating plainly what you do, where and for whom, and keeping that consistent everywhere, is how you pass.
The third gate, and the one most small businesses fail: whether the wider web agrees you are a credible answer. Reviews, mentions, coverage and listings that back up your own claims.
An assistant recommending you is staking its answer on you. It wants the rest of the web to back it up before it does. No amount of on-site work closes this gate; it is earned off your own site, and it is where the real movement in AI visibility comes from.
A generative AI tool that answers a question directly instead of returning a list of links. ChatGPT, Claude, Gemini and Perplexity are the main ones. When asked who to use, they name specific businesses.
Each engine reaches its answer slightly differently, and they do not always agree, which is why AI visibility is measured across several of them rather than one.
When an AI assistant runs a live web search to support its answer, rather than relying only on what it learned in training. Grounding is why fresh third-party signals can change what a model recommends.
A business that did not exist when a model was trained can still be recommended if grounding surfaces it. It also means your recent reviews and mentions can move the answer without waiting for the next model release.
The technique behind grounded answers: the assistant retrieves relevant documents from the web or a database, then writes its answer from them. If your content is not retrieved, it cannot be cited.
RAG is why retrieval is the first gate. The best content in the world does nothing if it is not among the documents the engine pulls in to answer the question.
When an AI assistant states something confidently that is not true, including inventing a business or getting your details wrong. Strong entity clarity and corroboration reduce the chance an engine misrepresents you.
For a business, the risk is not only being left out but being described incorrectly. The clearer and better corroborated your information, the less room an engine has to get it wrong.
A named source an AI assistant credits in its answer. Being cited is stronger than being mentioned: it is the engine pointing the buyer to you as the authority.
Citations are also how the effect compounds. Once an engine treats you as a source for your category, it tends to keep doing so. See the Citation Moat.
The practice of getting a business recommended by AI assistants, as distinct from ranking in traditional search. What this whole glossary is about.
GEO is to AI assistants what SEO was to search engines, with one difference: the goal is to be named in the answer, not to appear in a list of links the user still has to choose from.
A near-synonym for GEO, framed around answer engines. The goal is the same: being the answer, not just a link.
Different people in the field prefer different labels. The work underneath, retrieval, entity clarity and corroboration, is the same whichever term is used.
A search resolved without the user clicking through to any website, because the assistant answered directly. Zero-click is why being named in the answer now matters more than a ranking that never gets clicked.
As more answers are settled inside the assistant, the value of a high ranking that the buyer never clicks falls, and the value of being the business the assistant names rises.
Machine-readable data, usually JSON-LD, that states what a page is about: the business, its services, prices, reviews and FAQs. Schema helps engines retrieve and resolve you. It amplifies authority; it does not create it.
Good schema makes you easier to read and harder to misunderstand. But a technically perfect site with weak corroboration still loses. Add schema on top of authority, not instead of it.
Structured data that marks up questions and answers so an engine can lift them directly. Answering the exact questions buyers ask, in FAQ schema, is one of the most direct ways to be quoted.
Phrase the questions the way people actually ask them, and keep each answer self-contained, so it makes sense when quoted on its own.
A plain-text file at /llms.txt that gives AI tools a concise, machine-readable map of your most important pages and what your site is about. An emerging standard for making a site easy for assistants to read.
Think of it as a summary written for machines: what you do, and where the substance lives, so an assistant does not have to guess.
The bots AI companies use to read the web, such as GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), Google-Extended and PerplexityBot. If your robots.txt blocks them, you opt out of being retrieved.
Some sites block these bots by accident, or on old advice about protecting content. For a business that wants to be recommended, blocking the AI crawlers is closing the first gate on yourself.
How unambiguously an engine can identify who you are: name, location, services and credentials, stated consistently across your site and the profiles you link to. Weak entity clarity is a common reason a business is read but not named.
Linking to authoritative profiles you own, and describing yourself the same way everywhere, gives the engine the confidence to name you rather than hedge.
The on-site signals an engine can read directly: schema, llms.txt, FAQ markup, crawlability and clean structure. Measured as part of the AI Referral Score. Necessary, but on its own rarely enough.
A high technical-readiness score means the door is open. Whether the engine walks through it and names you still depends on resolution and corroboration.
A measure of how often ChatGPT, Claude, Gemini and Perplexity recommend a business for the questions its customers actually ask, scored against a day-0 baseline so change is visible over time.
One check tells you almost nothing, because the same question asked twice can return different businesses. A score is only useful against a baseline you measure again and again.
It combines top-pick rate, mention rate and technical readiness, so you can tell whether the work you are doing is actually moving your position.
How often an AI assistant names your business as its first or primary recommendation, not just a passing mention. The strongest form of AI visibility.
Being mentioned somewhere in an answer is worth little if a competitor is the one the buyer is told to use. Top-pick rate is the metric that tracks the referral you actually want.
How often your business is named at all in an assistant's answer, in any position. Useful as a floor, but a mention without being the top pick still tends to lose the enquiry.
Rising mention rate is an early sign the engines are learning who you are. The aim is to convert those mentions into top picks.