Does ChatGPT recommend your business? Here is how to check, and what to change

Someone in your town opened an AI assistant this morning and typed something close to “best [what you do] near me”. They got back a short list of names, maybe three or four, with a sentence about each. Then they picked one.

You have no idea whether you were on that list.

That is the uncomfortable part. A search engine at least leaves a trail: impressions, clicks, a position you can look up. An AI answer leaves nothing. No referral, no ranking report, no notification that a competitor got named and you did not.

This post is the manual version of the check. It takes about half an hour, costs nothing, and gives you a real baseline. At the end there is a short note about doing it on a schedule instead, which is what we build, but the method below works whether or not you ever use our app.

First, a caveat worth more than the rest of the post

AI answers are not stable the way search results are. The same question can produce different names on different days, in different accounts, in different countries. Assistants personalize, they use memory of past chats, and they change models without telling anyone.

So treat any single answer as one sample, not a verdict. What you are looking for is a pattern across several questions, repeated over time. One answer that leaves you out means nothing. Eight out of ten leaving you out is a finding.

Part 1: Check it yourself in about 30 minutes

Step 1: Write ten questions a buyer would actually type

Not your brand name. Almost nobody who has not heard of you types your brand name. The useful questions are the ones a stranger asks when they have a problem and no shortlist yet.

A good set mixes four shapes:

1. Category plus place. “Best gluten free bakery in Austin.”
2. Job to be done. “Where can I get a wedding cake that is dairy free in Austin?”
3. Comparison. “Which bakeries in Austin do custom tiered cakes?”
4. Constraint. “Bakery in Austin open Sundays that takes last minute orders.”

Write ten. Use your real category, real city or service area, and the words your customers use, not your industry’s words. If your customers say “sourdough”, do not write “artisanal fermented bread”.

Log out or use a private window first. If you are signed in, the assistant may already know you and quietly bias the answer toward your own business. That feels great and tells you nothing.

Step 2: Ask each question in each assistant

The three that most people use for recommendations today are ChatGPT, Perplexity and Gemini. Ask the same ten questions in each. Yes, that is thirty questions. It goes faster than it sounds, and the differences between them are the interesting part.

Step 3: Record four things per answer

A spreadsheet with five columns is enough:

Also note, in a sixth column, how you were described when you were named. Wrong hours, a service you stopped offering, or an outdated price is a different problem from not being named, and a more fixable one.

Step 4: Turn it into three numbers

• Mention rate. Answers that named you, divided by total answers. Nine out of thirty is 30%.
• Prominence. When you were named, how high in the list. First and second mentions carry far more weight than a name buried at the end of a paragraph.
• Share of voice. Your mentions divided by all business mentions across the whole set. This is the one that tells you whether a specific competitor is quietly owning your category.

Write the three numbers down with today’s date. That single row is the most valuable output of the exercise, because it is the thing you will compare against in a month.

Step 5: Note which competitors keep winning

By the end you will notice two or three names that appear in almost every answer. Those are not necessarily the biggest businesses in your category. They are the ones whose public information is easiest for a machine to read and trust. Open their sites and look at what is different. It is usually boring: clear service pages, consistent business details, real content answering the exact questions you just asked.

Part 2: What the answer actually depends on

Assistants do not have a secret list of good businesses. When a question sounds local or commercial, they generally search the live web, read a handful of pages, and summarize. Which means the levers are the same public signals that have always mattered, plus a few that are new.

Ranked roughly by how much control you have over them:

1. Whether the assistants’ crawlers can read your site at all.

   Each AI company publishes named crawlers, and they are not one thing. Search crawlers that fetch pages to cite in answers are separate from training crawlers. OpenAI publishes GPTBot for training and OAI-SearchBot for its search features. Perplexity publishes PerplexityBot. Google uses Google-Extended as the control for Gemini training. If your robots.txt blocks the search crawlers, you have opted out of being cited. Blocking training crawlers while allowing search crawlers is a legitimate choice, and a different one.

   Two honest caveats: this landscape changes often, and independent reporting has documented some crawlers ignoring robots.txt entirely. Check the current published list from each provider rather than copying a blog post, including this one.

2. Whether the facts about you are stated plainly, in one place.

   Assistants summarize. Summarizing is easier when the source is unambiguous. Name, what you do, where you do it, hours, service area, price range, and what makes you different, all stated in text, on a page, in sentences. Not in an image. Not only in a PDF. Not implied by your photography.

3. Structured data.

   Schema markup, mainly LocalBusiness for a service business and Product for a store, is a machine readable version of those same facts. It is not magic and it does not force anyone to cite you, but it removes ambiguity, and ambiguity is what gets you left out of a shortlist.

4. llms.txt.

   A plain text file at the root of your site summarizing what your site is and where the important pages are. It is a proposed convention rather than a settled standard, and no assistant promises to read it. It is also cheap, and Wix generates one for Wix sites automatically, which is a decent signal about where the platform thinks this is going.

5. Content that answers the actual question.

   If people ask “do you do dairy free wedding cakes” and your site never says those words in that order, you are relying on an assistant to infer it. Assistants are not great at inferring, and they are cautious about recommending things they cannot verify. A page that answers the question in the question’s own words is the single most reliable content change you can make.

6. What other sites say about you.

   Directories, local listings, review platforms, press, roundups. Assistants weigh corroboration. A business described consistently in five independent places is easier to recommend than one that only describes itself.

Part 3: What to change on a Wix site specifically

Wix has quietly shipped a lot of the plumbing here. Most owners have not looked.

Check your AI visibility overview.

Wix Analytics includes an AI Visibility Overview that tracks how often your site is mentioned across ChatGPT, Gemini and Perplexity, compares you to competitors, and shows which sources are cited. Per Wix’s Help Center it needs at least 50 unique visitors in the last 30 days before it will display data, keeps a limited history of past tests, and lets you change your questions once a day. If your site has traffic, look at it before you do anything else. It is included with your site and it is a real data source.

Look at your llms.txt.

Wix generates and maintains this file automatically, and per Wix’s Help Center it is now available on all sites, not just upgraded ones. You can find it under SEO & GEO in your site’s dashboard, in tools and settings, where you can view it, edit it, or opt out. One thing to know before you edit: once you edit the file it stops updating automatically, so your changes are preserved but new pages will not be added for you. You can reset it back to automatic.

Read what it says about your business. If the description is generic, that is what a machine reading your site concluded about you.

Confirm you have not blocked the crawlers.

Wix has a setting for blocking AI crawlers. It exists for good reasons, and some site owners deliberately turn it on. If yours is on and you are wondering why you are never mentioned, that is your answer.

Fill in the SEO fields you skipped.

Product SEO fields, page titles, meta descriptions, alt text. Empty fields are common on Wix stores because the product upload flow does not force them. Filling them is unglamorous and it is the closest thing to free signal available.

Add structured data for your business type.

LocalBusiness with real hours and address for a service business. Product markup for a store. Wix supports custom structured data, and getting the basics right beats getting exotic types wrong.

Answer your three lost questions.

Go back to the spreadsheet from Part 1. Take the three questions where you were never named and a competitor always was. Write one clear page or section that answers each one directly, in the words the question used. This is the change most likely to move the needle, and the one most people skip because it is work rather than a setting.

Part 4: Then do it again

Here is the part that separates a useful exercise from a one off scare.

Change one thing. Wait a week or two. Run the same ten questions again. Compare the three numbers.

If nothing moved, that is information too, and it is worth more than a one time audit that told you a number with no way to check whether it meant anything. The tools that measure and stop leave you exactly where you started, just better informed about it.

A reasonable rhythm for a small business:

1. Baseline today, with the same ten questions.
2. One change per week from Part 3, starting with the crawler check and the lost questions.
3. Re measure every two weeks, with the same questions, logged out.
4. Keep a single row per measurement. After three months you will have something no competitor of yours has: a trend line for something most business owners cannot see at all.

Doing this automatically

The manual method works. It is also thirty questions across three assistants, every two weeks, plus the bookkeeping, and most owners run it once and never again.

That gap is why we built Patesco, an app in the Wix App Market. It reads your published site, writes the buyer questions for you, asks the assistants on a weekly schedule, scores mention rate, prominence and share of voice, names the competitors that keep beating you, and pairs the score with a prioritized fix list. Several of those fixes, including llms.txt, structured data and product SEO, can be applied to your site from the dashboard with a preview and one click undo. A weekly email tells you what moved.

The free plan runs one scan and shows the real score, your top competitors by name, and one complete fix you can apply, with nothing blurred out. If the manual method above is enough for you, use the manual method. It is the same idea.

FAQ

Patesco is an independent tool for Wix site owners. ChatGPT, Perplexity and Gemini are trademarks of their respective owners and are referenced here only to describe which assistants are checked. Patesco is not affiliated with, endorsed by, or partnered with any AI provider or with Wix.