How to get ChatGPT to recommend your business
You cannot make an assistant recommend you. You can change what it reads about you, which is the only lever that has ever existed. Here is the work, in the order it pays.
Start with the uncomfortable part. There is no submission form, no listing to buy, and no setting inside ChatGPT that puts your business into an answer. Anyone selling you a guaranteed placement is selling you something they do not control.
What you can control is the raw material. An assistant answering “who is a good electrician near me” is assembling that answer from things other people wrote: your Google Business Profile, your website, the text of your reviews, directories, and any mention of you anywhere else. Improve the raw material and you change the odds. That is the whole game, and it is more winnable than it sounds, because most of your competitors have not touched it.
45% of consumers now use AI tools for local business recommendations, up from 6% a year earlier, and 82% read AI generated review summaries. Both figures come from BrightLocal’s Local Consumer Review Survey 2026, cited at the bottom of this page. That is a step change inside twelve months, not a trend line.
Step 1. Find out where you actually stand
Almost nobody has done this. Before you change anything, spend twenty minutes finding out what assistants currently say about you, because otherwise you will have no idea whether any of the work below did anything.
Open ChatGPT, Gemini and Google’s AI mode. Ask each one the same five questions. Start a fresh chat for every question so an earlier answer cannot influence the next one.
- Who is a good [your trade] in [your town]?
- I need [your most urgent service] in [your town] today. Who should I call?
- What should I look for when choosing a [your trade] in [your town], and who is worth contacting?
- Roughly what does [your most common job] cost in [your town], and who does it?
- Tell me about [your exact business name]. What do they do, what do people say about them, and what do they not do?
Paste every answer into a document, whole, not summarised. Write the date next to it. Five questions across three assistants is fifteen answers, and that is your starting position.
The fifth question is the one that pays. It shows you what the internet currently believes about your business, in one paragraph, written by something with no reason to flatter you. Most owners find at least one fact in it that is wrong, out of date, or lifted from a directory they had forgotten about.
Step 2. Understand why it names some businesses and not others
An assistant needs three things before it will put your name in an answer. Almost every failure traces back to one of them.
It has to know you exist, from facts that agree
Your name, address or service area, phone, hours and services have to match everywhere they appear. When two sources disagree, an assistant hedges, and hedging means it names somebody else instead.
It has to be able to match you to the need
The assistant is matching a described problem to a described service. If your site says “bespoke property solutions” and the customer says “blocked drain”, there is nothing to connect.
Somebody other than you has to say it too
Your own website claiming you are the best in town is marketing. Reviews, directories and third party mentions are evidence that survives a second opinion. This is usually the thing separating the businesses that get named from the ones that do not.
Step 3. How to optimize for ChatGPT search
This is where most of the winnable work sits, and it is unglamorous. Four things, roughly in order of return.
Let the crawler in
ChatGPT surfaces websites through OpenAI’s own crawler, OAI-SearchBot. OpenAI states that sites opted out of it will not appear in ChatGPT search answers. Check your robots.txt allows it, and check your host or CDN is not blocking OpenAI’s published address ranges either. This takes five minutes and it is the one item on this page that can silently cancel everything else.
Answer first, explain second
Web copy traditionally warms up. Assistants extract. Put the direct answer in the first one or two sentences under each heading, then explain underneath.
Here is the test. Delete everything on the page except the first two sentences of each section. Does it still answer the question? If not, restructure until it does.
Say the specific things nobody else will say
Vague claims cannot be quoted, because there is nothing to lift. “We respond fast” is unquotable. “We answer within five minutes in working hours and text back automatically outside them” is a sentence a model can lift word for word.
Most service websites leave out exactly the facts an assistant needs in order to recommend them. Put these in plain language on the page.
| Fact | Why it gets used |
|---|---|
| What you charge, or how you charge | Callers ask it first. If the number is not on the page, nothing can quote it |
| How fast you respond | Directly comparable, so easy to rank you against someone else |
| Exactly which towns you serve | Turns a general recommendation into a local one |
| What you do not do | Filters out the wrong callers, and gives an assistant a boundary it can state for you |
| Insurance, accreditations, registration numbers | Facts a stranger can verify |
| How long you have operated | Cheap credibility, trivially quotable |
Publish what you do not do, and who you are not for. It costs nothing, it removes ambiguity about your boundaries, and it is reversible in an afternoon. We rate it highly from our own work rather than from a published study, so treat it as an experiment you can score: write it, then ask three assistants what your business does not do, and see whether they get it right.
Write a real question and answer section
Cover the ten questions you get on every call, written as plain headings in the words customers use. Skip FAQ schema markup. Google stopped showing FAQ rich results on 7 May 2026 and removed the documentation the following month, so the markup now buys you nothing. The plain text still works.
Step 4. How to show up in AI search results beyond ChatGPT
ChatGPT is one assistant. The corroboration work below is what moves all of them at once, and it is the part most businesses skip because it happens off their own website.
- Keep reviews current and specific. Google builds its review summary from reviews in the last twelve months, so it is the recent text being read, not the star average. A review that names the job and the outcome can be summarised. “Great service, thanks” cannot.
- Reply to reviews with specifics rather than thanks. Your replies are part of what gets read, and they are in your own voice.
- Claim Bing Places. It is free, it relaunched in October 2025, and Bing is a separate index rather than a copy of Google’s, so it is genuinely a second source.
- Claim Foursquare. Its places data is now an open dataset under an Apache 2.0 licence, distributed well beyond Foursquare itself, so an accurate record there travels further than you would expect.
- List on the two or three directories that actually rank for your trade plus your town. Search it and see which ones appear. Ignore every other one.
- Get one genuine third party mention: a supplier’s installer list, a trade register, a chamber page, local press.
- Add LocalBusiness schema to your site carrying the same facts as everywhere else, then confirm it with Google’s Rich Results Test.
What does not work
- Buying a bundle of a hundred directory listings. Two or three that rank do the whole job, and bulk submissions look like exactly what they are.
- Writing your customers’ reviews for them, or tying any reward to a review being positive. Ask for the fact, not the verdict.
- Seeding questions in your Google Business Profile from second accounts. Google prohibits content posted from multiple accounts at one person’s request. Put the answer in your business description instead, where you control the wording anyway.
- Asking the model which page it used. A system asked after the fact will produce a plausible answer whether or not it read one. Use the sources it actually displays, and record no citation when it displays none.
- Reacting to a single bad answer. These systems are non deterministic. The same question can return a different answer an hour later.
How to know it worked
Re-run the same fifteen questions once a month and score each answer. Two points if it names you and describes you accurately, one if it names you but vaguely or with a wrong fact, zero if it does not name you at all. Fifteen answers puts the ceiling at thirty.
Nobody publishes a benchmark score for this, so anyone quoting you one has invented it. The number to beat is your own from last month, and the direction across three months matters far more than any single run.
Expect things to move in this order. The description of your business gets accurate first, because you control that material directly. Then you start getting named on the questions you should own. Then, eventually, enquiries arrive mentioning things you never advertised, and nothing in your analytics will ever explain where they came from. That last one is what success actually looks like here, and it is exactly why the score sheet matters.
- VerifiedBrightLocal, Local Consumer Review Survey 2026 1,002 US adults. 45% of consumers use AI tools for local business recommendations, up from 6% in the 2025 survey, and 82% read AI generated review summaries.
- VerifiedOpenAI, crawler documentation OAI-SearchBot is used to surface websites in ChatGPT’s search features, and sites opted out of it will not be shown in those answers. This is the basis for the robots.txt check.
- VerifiedGoogle Search Central, search updates FAQ rich results stopped appearing on 7 May 2026 and the documentation was removed on 15 June 2026. This is why the guide says to write a plain question and answer section and skip the markup.
- VerifiedMicrosoft, Bing Places for Business Free listing platform for Bing search and Bing Maps, relaunched October 2025. Microsoft publishes no setup time and no adoption rate, so this page claims neither.
- VerifiedFoursquare, FSQ OS Places Open dataset of over 100 million places, released under an Apache 2.0 licence and distributed via Snowflake and Hugging Face.
- On weighting, and what we do not know Google publishes what it uses for local ranking, and OpenAI publishes which crawler feeds ChatGPT search. Neither publishes how an assistant weighs one source against another. Where this page says corroboration is what decides who gets named, that is our read from doing the work, not a documented rule. We flag it once here rather than hedging every line.
Run it monthly instead of once
Doing this once tells you very little. Doing it every month tells you whether anything you changed actually worked. The AI Visibility System is the instrument: the five prompts written out, the scoring scale, a failure map that turns each kind of bad answer into the specific fix it calls for, and a twelve month score sheet.
The thinking behind it is free and always will be: The GEO checklist · The local AI visibility checklist · The entity checklist · The AI-proof review guide
Get the next one
We publish a new field guide most weeks. Plain method, sources cited, no pitch. Leave your address and we will send the next one.
