How to check whether AI assistants recommend your business
Some of your customers have stopped typing ac repair near me into a search box. They ask an assistant instead, in a full sentence, and they act on the two or three businesses it names back. If yours is not one of them, nothing tells you. There is no notification, no rank drop, no line in an analytics dashboard. The call simply never comes.
You do not need to buy anything to find out where you stand. The first pass is a method, a list of questions and an afternoon. Here is how to run it, and — just as important — what the answer is worth once you have it.
This does not work like Google
Search rankings are stable enough to be a number. Position four is position four this morning and most likely this afternoon, for everyone, and a tool can watch it for you.
AI answers do not behave that way. Ask the same question twice and you can get two different sets of businesses. Change the city, the phrasing, or the time of day and the answer moves again. There is no position to record, because there is no list — there is a paragraph that names some companies and not others.
So the goal is not a score. The goal is a picture: across a realistic spread of questions, how often does your name come up, and when it does, what is the assistant reading to get there?
Write the questions a customer would actually ask
This is the part people get wrong. Searching your own company name tells you almost nothing — an assistant asked about you directly will find you. The questions that matter are the ones where the customer does not know you exist yet.
Aim for ten to fifteen, spread across:
- Your highest-value job, plus a real city you serve — not the metro, the city. "Who replaces residential AC systems in Pembroke Pines?"
- The emergency version, phrased the way someone types at 9pm. "My AC stopped working, who can come out tonight in Hollywood, Florida?"
- The comparison question. "Which AC companies in Fort Lauderdale have the best reputation for maintenance?"
- The qualifier questions your best customers care about — licensed, financing, a specific brand, weekend calls.
- One or two where you would expect a competitor to win, so the sheet is not built only from questions you like.
Ask them the same way every time
Consistency is what separates a test from an afternoon of curiosity. A few small habits do most of the work:
- 01Start a fresh session for every question.
An assistant carries context forward. If you have just been discussing your own company, the next answer is contaminated — and flattering.
- 02Sign out, or use a private window.
You want the answer a stranger gets, not the one tuned to your history.
- 03Use more than one surface.
ChatGPT, Gemini, Perplexity and Google's AI Overviews read different sources and answer differently. Being invisible on one and fine on another is a normal, useful result.
- 04Record the date and the exact wording.
You are going to want to re-run this in a quarter. Without the exact phrasing, the comparison is worthless.
- 05Ask each question at least twice, in separate sessions.
Answers vary. Two runs tells you whether a mention is solid or a coin flip.
Write it down as you go, one row per question per surface. The sheet is the deliverable — not your memory of how it felt.
| Question asked | Surface | Mentioned | Recommended | Cited source |
|---|---|---|---|---|
| AC repair · City A | ChatGPT | yes | no | third-party directory |
| AC repair · City A | Gemini | no | no | — |
| AC maintenance · City B | Perplexity | yes | yes | business website |
| AC replacement · City C | AI Overviews | not triggered | — | — |
Mentioned, recommended and cited are not the same
Most people collapse these into "did I show up?" and lose the signal. Keep them apart:
- Mentioned — your name appears somewhere in the answer. Better than nothing, and often just a list entry.
- Recommended — the assistant actually points the customer at you, usually with a reason attached. This is the one that turns into a phone call.
- Cited — the answer links a source. This is the most actionable column on the sheet, because it tells you what the assistant is reading.
That last column is where the work comes from. If the answers about your trade in your city keep citing a directory listing, a review aggregator or a competitor's service-area page, then those pages — not your homepage — are what the assistant has learned from. A mention you earn through someone else's stale listing is a mention built on information you do not control.
Fix the sources, not the answer
You cannot edit an AI answer. You can only improve what it has to work with — and most local service businesses have the same gaps, in the same order:
- 01Make the basics consistent everywhere.
Name, address, phone, hours and service area, matching exactly across your site, your Google Business Profile and the directories that showed up in your cited column. Contradictions are the cheapest problem you have and the easiest to fix.
- 02Put your service area in writing, in plain text.
The cities you actually serve, named on a page a machine can read. Not baked into a map widget, a graphic or a hero image.
- 03Answer the questions on your own site.
Take the list from step one and write a real, specific answer to each. The pages that get cited tend to be the ones that address the question directly rather than selling around it.
- 04Claim and correct the listings that got cited.
Your test named them. Those are the ones worth your afternoon — not every directory on the internet.
- 05Re-run the same questions in a quarter.
Same wording, same surfaces, fresh sessions. The comparison is the only way to know whether anything you changed mattered.
Notice what is not on that list: nothing about tricking a model, and nothing you would be embarrassed to explain to a customer. The businesses that come out well in AI answers are, for the most part, the ones whose public information is accurate, specific and easy to read. That is an unglamorous conclusion, and it is the honest one.
Run the pass yourself first. Even a rough sheet beats a guess, and it will tell you within an afternoon whether this is a live problem for your business or something to look at again next year.
We test a defined set of local questions across agreed AI surfaces, with repeated samples, competitor comparison and the evidence behind every row — then turn it into a prioritized list of fixes.
See how Byldr GEO works →