AI Search · Local Business
Does ChatGPT Recommend Your Business? How to Check
You can see every dollar a lead platform charges you. You cannot see the customer who asked an assistant for a recommendation and never heard your name. That is the one channel nobody is checking.
Key takeaways
- Assistants answer local hiring questions with named businesses, and those names come from retrieved third-party sources rather than from search rankings.
- Answers vary run to run. A single check is not evidence in either direction, which is the most common mistake people make here.
- The test is ten real customer questions, run twice on different days, scored as named, mentioned or absent.
- Ranking well in Google does not guarantee being named. The two systems select on different things.
- Nobody invoices you for being absent, which is exactly why this goes unmeasured while smaller, visible line items get scrutinised.
01The channel with no invoice
Every other marketing channel you use announces itself with a charge. This one is silent, and silence is why it goes unchecked.
A lead platform bills you per lead, so you notice. An agency invoices monthly, so you notice. When a homeowner opens an assistant, types “who is the best electrician near me for a panel upgrade”, gets three names back and calls one of them, nothing arrives in your inbox. If you were not one of the three, there is no record anywhere that it happened.
Contractors already apply hard scrutiny to the channels they can see. The question in every trade forum is some version of this one:
“I'd especially appreciate hearing what actually generates good electrical leads versus what is a waste of money.”
That scrutiny is right, and it has a blind spot. The channels being argued about are the ones that send a bill. The one that does not send a bill sits in front of all of them, because it increasingly happens before the customer ever opens a directory or a results page.
02The fifteen-minute test
Write ten questions a customer would genuinely ask before hiring someone in your trade, then run them properly. The quality of this test lives entirely in the quality of the questions.
Phrase them the way a person speaks, not the way a keyword tool writes. “My water heater is leaking, who should I call in Tucson” is a real question. “water heater repair Tucson” is a search term, and it tests the wrong system. Include the four shapes below.
| Question shape | Example | What it tests |
|---|---|---|
| Direct hire | Who is the best roofer in Boise for a full replacement? | Whether you are named when someone is ready to buy. |
| Problem first | My AC is blowing warm air, who do I call in Mesa? | Whether you are retrieved from a symptom rather than a service name. |
| Comparison | Compare the top-rated tile contractors in Sacramento. | Whether you make a shortlist your competitors are on. |
| Trust and vetting | How do I find a licensed electrician I can trust in Dayton? | Whether your credentials and reviews are legible to the model. |
Then follow four rules that make the result mean something:
| Rule | Why |
|---|---|
| Start a fresh chat for every question | Earlier messages steer later answers. A continuing conversation tests your chat history, not the channel. |
| Log out, or use a private window | Personalisation and memory can surface a business you have looked at before. That is your history, not visibility. |
| Record three outcomes, not two | Named first, mentioned among others, or absent. The middle case is the most common and the most actionable. |
| Run the whole list again on a different day | One pass is a sample of one. Two passes start to show you a rate. |
03Why one check proves nothing
Ask an assistant the same question twice and you can get two different lists of businesses. This is not a malfunction. It is how the systems work.
These models are probabilistic, and they retrieve material at the moment of the question. That makes any single check close to worthless as evidence. If you appear once, you have not established that you are visible. If you are missing once, you have not established that you are invisible.
What is meaningful is a rate: across ten questions and two runs, you have twenty observations, and the share in which you were named is a number you can compare against next month. This is the difference between checking and measuring, and it is why a serious answer here cannot be produced by asking ChatGPT about yourself once over coffee.
04Reading your result
Count the share of your twenty observations in which you were named at all.
| Named in | Read | What to do next |
|---|---|---|
| 0% | You are invisible in this channel. Common, and fixable. | Start with third-party corroboration. Reviews with real text, consistent listings, local mentions. |
| 10% to 30% | You exist but are not a default answer. | Look at which question shapes you win. Usually you appear on direct hire and vanish on problem-first. |
| 40% or more | You are an established option in your area. | Track it monthly and watch for competitors displacing you. |
Also record who does get named. That list is more useful than your own score, because it is the set of businesses the model currently treats as the credible answer in your market, and it tells you what a credible answer looks like from the outside.
05What actually moves it
Local recommendations lean on what other sources say about you, not on what your own website claims. A business whose only presence is its own site gives the model one unverified source and little reason to name it over a competitor with corroboration in five places.
| Lever | Why it matters here | Effort |
|---|---|---|
| Reviews with substantive text | A review that describes the job done is retrievable evidence. A bare five-star rating is not. | Ongoing. Ask customers to say what the work was. |
| Consistent listings across directories | Conflicting names, addresses and service areas make you harder to state confidently. | One-off cleanup, then occasional checks. |
| Plain statements of service and area on your site | Assistants lift self-contained claims. Vague brand copy cannot be quoted. | An afternoon of rewriting service pages. |
| Mentions on sites you do not own | Local press, supplier pages, community sites and forum threads all count as corroboration. | Slow, and the most durable of the four. |
Notice that none of these is a generator. Drafting the copy is something you or any assistant can already do well. The part that is hard, and the part worth paying attention to, is knowing what is actually missing and whether a change moved the number. That is measurement, and it is the same discipline that decides whether a lead platform is earning its place and whether an SEO retainer is doing anything. Same question, three channels: what did I pay for, and what did it produce?
Run this properly instead of by hand
Linkeddit runs your buying questions across the major AI assistants on a schedule, scores every answer as named, mentioned or absent, records the sources each answer drew on, and tells you whether a change you made actually moved the rate. The manual test above works. This is the version that keeps working every month without you sitting there.
06FAQ
Frequently asked questions
Does ChatGPT recommend local businesses?+
Yes. When someone asks for a plumber, a roofer or a dentist in a given area, assistants routinely return named businesses rather than a list of links. The names come from what the model can retrieve about your area at that moment: directory listings, review pages, local publications, forum threads and your own website. That is a different mechanism from ranking in Google, which is why a business can sit at the top of local search results and still never be named.
How do I check whether ChatGPT recommends my business?+
Write down the ten questions a customer would actually ask before hiring you, phrased the way they would speak them rather than as keywords. Run each one in a fresh chat with no prior history, ideally while logged out, and record whether you were named, named among others, or absent. Then run the same list again on a different day. The second pass matters, because assistant answers vary between runs and a single check tells you almost nothing.
Why do I get a different answer every time I ask?+
Because these systems are probabilistic by design and retrieve fresh material at the time of the question. Two identical prompts minutes apart can return different businesses. This is the single most misunderstood thing about the channel, and it has a practical consequence: one lucky check where you appear is not evidence you are visible, and one unlucky check where you are missing is not evidence you are invisible. Only a repeated sample across many questions and runs gives you a real rate.
Why does my business not appear in AI answers?+
Usually because the assistant cannot find enough corroborating evidence about you across the wider web. Local recommendations lean heavily on third-party sources: reviews with substantive text, consistent listings across directories, mentions on local sites, and pages that state plainly what you do and where you do it. A business whose only real presence is its own website gives the model a single unverified source and very little to recommend on.
Does traditional SEO make me appear in ChatGPT?+
It helps but it is not the same job, and assuming they are identical is the common mistake. Traditional SEO optimises a page to win a position on a results page. Assistants build an answer out of retrieved material and need claims they can lift cleanly: a plain statement of what you do, where you serve, what you charge and what proof exists, ideally corroborated somewhere other than your own site. A page can rank well and still be nearly unquotable.
How often should a local business check AI visibility?+
Monthly is a sensible floor. Models are updated, retrieval changes, competitors publish, reviews accumulate, and your rate moves without anything happening on your own website. Checking once and assuming the result holds is the same error as checking your Google ranking once in 2019 and never looking again.
Is this worth a local business paying attention to yet?+
It deserves measurement before it deserves budget. The honest position is that nobody can currently tell you what share of your customers arrived because an assistant named you, because the influence usually happens with no click to trace. What you can establish cheaply is whether you are present at all. Being absent from every answer to the ten questions that precede a purchase in your trade is a fact worth knowing, whatever you decide to do about it.