What is the difference between an AI recognising my business and recommending it?

This page is the mechanism, not the fix list

It explains why an engine can hold a true fact about you and still not put you in a recommendation, and it applies to ChatGPT and Gemini alike. If what you want is the sequence of moves that gets you out of that position, that is a different question and it has its own page: ChatGPT knows my business but never recommends it sets out the three states, what corroboration actually means and the week of work that shifts it.

Recognition and recommendation are not two grades of the same thing

They are two different jobs, and the difference is decided by what is in the question, not by how good you are. This is the part most owners have backwards, and getting it right changes what you do next.

When you ask an engine about your business by name, your name is the anchor. Everything the engine does afterwards is scoped to you: any retrieval it runs is aimed at pages about you, and any recall it draws on is recall about the entity you just named. The engine does not have to decide whether you are worth mentioning. You have already been mentioned, by the person asking.

When a buyer asks who to use for a job in a city, your name is nowhere in the question. The engine now has to produce a set of candidates from a category and a place, then compose an answer from that set. Nothing about that operation starts from you. It starts from the question and works outwards to whoever the available material puts forward, and if you are not in the material it assembles, there is no later step at which the engine consults its stored knowledge of you and adds you in.

That is the whole mechanism, and it explains the thing that feels like a contradiction. A stored fact about you is retrievable by your name. It is not indexed under "best plumber in Adelaide". The two operations never meet.

Where the candidate set comes from

From material about the category, not material about you. There are two ways a name gets into a buyer answer, and neither of them reads a stored profile of your business.

• Retrieval for the buyer question. The engine searches on the question it was given, gets back pages that answer that question, and composes from those. The pages that come back are roundups, comparison pages, directories, review platforms, local press and association listings, because those are the pages written to answer that question. Your own website is not usually among them, because your website answers "who is this business", not "who should I use".
• Unprompted recall. The engine names businesses it associates with the category and place strongly enough to volunteer them without searching. That association is built from how the corpus talked about the category, which again is third-party text.

Both routes run on pages you do not own. That is why a business can be perfectly described when asked about directly and completely absent from every buyer answer: the first draws on material about you, and the second draws on material about the job.

Which of the two routes produced any particular answer is usually visible, because retrieval leaves citations and recall does not. The check is on how often ChatGPT updates what it knows about your business, and the sources worth chasing once you have the list are in where ChatGPT gets its information about you.

An accurate description is evidence about the wrong thing

It proves your website did its job, which is exactly why it is such a poor predictor of whether you get recommended. Owners read a correct, fluent description of their business and reasonably conclude the engines are on top of them, so the absence from buyer answers must be a website problem waiting to be found.

Our data says it is not. Across the 300 businesses scanned on 30 July 2026, on-site AI readiness averaged 74.9 out of 100 while the businesses behind those sites averaged 38.3 for visibility, and readiness correlates with the per-engine visibility scores at Pearson r = 0.116 across the 137 with complete two-engine records. A readable site is close to uninformative about who gets named.

How we measured recognition, and where that measurement stops

Recognition is graded from the stored transcript of a direct question about each business, and in this dataset that flag is effectively a Gemini measurement. It fired for Gemini on 97 of the records and for ChatGPT on 1 of all 300. So when this page says 26% of the 137 businesses with complete two-engine records were recognised and never volunteered, that is 35 businesses, and the recognition half of it is Gemini-attributed. We are not claiming a per-engine split for ChatGPT, because our own file does not support one.

Two further limits worth stating in the same breath. Each question was asked once on 30 July 2026, with no repetition runs, so a business one answer short of a mention on the day is counted as never named. And a single direct answer is one sample of a system that is not deterministic, which is why we report rates rather than screenshots. The counting rules are published in full on the technology page, and every headline figure with its denominator is on the statistics page.

Three things people believe about this that the machine does not do

Each of these is a reasonable inference from how the answers read, and each one sends people to work on the wrong thing.

1. There is no stored ranking of you being withheld. The engine is not holding an opinion about whether you are good and declining to share it. There is no score attached to your business inside the model. Absence from a recommendation is not a judgement that was made about you; it is a set you were not in.
2. Recognition is not a stage on the way to recommendation. Nothing promotes a recognised business into the candidate pool over time. You can stay recognised and unrecommended indefinitely, because the two run on different material and neither feeds the other.
3. There is no submission queue, toggle or verification step. No form marks you as available for recommendation, and no amount of asserting your own quality in your own markup does it either. A schema block claiming an award nobody else mentions is still you, restated.

The productive version of the question is not "why is the engine ignoring what it knows about me", because it is not ignoring anything. It is "what would put me in the material this question retrieves". The six causes we found most often, ranked, are in why ChatGPT doesn't mention your business, and the head-to-head version, when a competitor is named and you are not, is in why ChatGPT recommends your competitor.