ChatGPT knows my business but never recommends it. Why?

The three failure states, and which one this is

There are three states a business can be in with an AI engine, and the work is different for each. Being known but not recommended is the second, and it is the cheapest of the three to move.

Recognition versus recommendation, in the data

In our scan the two came apart badly. Of the 137 businesses with complete records from both engines, 97 were recognised when asked about directly. Only 81 of those same 137 were named at least once in answer to a buyer question. Thirty-five of the 137, which is 26%, were recognised and never volunteered once.

One caveat that matters, and we would rather state it than let you infer something our data cannot carry. The recognition flag is graded from the stored direct-answer transcript, and in this dataset it fired for Gemini on 97 records and for ChatGPT on only 1 of all 300. So the recognition figures on this page are a Gemini-attributed measurement. We are not claiming a per-engine split for ChatGPT, because our own file does not support one.

What corroboration actually means in practice

Corroboration means somebody other than you, on a page you do not own, saying you are a good choice for a specific thing. That is the material an engine composes a recommendation from, and it is the thing the recognised-but-never-named group is missing.

What counts, roughly in order of how much it seems to matter:

• Reviews on platforms that already rank for your buyer questions, in volume and at a steady rate.
• Being included in a roundup, a best-of list or a comparison page written by someone else.
• Association registers, regulator listings and professional bodies, which are unusually trusted because they are hard to fake.
• Local or trade press that names you and says what you do.
• Directories your category actually uses, as opposed to directories that exist to sell listings.
• Supplier, partner and client pages that name you and describe the work.

What does not count, however good it looks on your own site: your testimonials page, your own case studies, a paid press release distribution, and a schema block asserting an award nobody else mentions. All of those are you, restated.

The fastest way to move from known to named

Get named on one page you do not own that already ranks for the buyer question you want to win. That is the whole move, and it is short enough to do in a week.

1. Write the buyer question exactly as a customer would say it, in your city, with no business name in it.
2. Ask it of ChatGPT and Gemini. Write down who gets named, and where the answer shows citations, which pages it cites.
3. Open those cited pages. That list is your target list, because those are the sources the engines are demonstrably already reading for this question.
4. Get yourself onto them legitimately: a listing, a submission to the roundup, a review on that platform, an association entry, a correction if the page is out of date.
5. Re-ask the same question, unchanged, a month later.

The word legitimately is doing real work in step four. Buying placements in manufactured roundups produces a paragraph on a page nobody trusts, and it is the first thing that gets cleaned up when these systems tighten their source selection. The six underlying causes, ranked by how often we found each, are set out in why ChatGPT doesn't mention your business.

How to confirm which state you are in

Two questions, in a fresh chat, in this order. It takes about two minutes per engine.

1. The buyer question. Ask what a customer would ask, in your category and city, with your name nowhere in it. Note every business named.
2. The direct question. Ask what the engine knows about your business by name and city. Check the description for accuracy: services, suburb, hours, prices, and whether it has confused you with anyone.

Read the pair like this:

• Named in question one: you are in the third state. Move to holding the position and correcting the details.
• Accurate in question two but absent from question one: recognised, never recommended. This page is about you.
• Vague, wrong or confused in question two: unknown. Start with legibility and basic corroboration before anything else.

Repeat the buyer question two or three times before you trust it, because a single answer is one sample. We run the same two-question shape mechanically, across a fixed set, and publish the counting rules on the technology page.