How to Check What AI Says About Your Business

A prospect tells you on a first call that ChatGPT suggested you. The next one says an assistant named two of your competitors and never mentioned you. Some buyers now ask an assistant before they search. You have no record of what it tells them.

Most founders check once. They type their company name into ChatGPT, read a flattering paragraph and move on. That answer came on one day, from a session that already knows who they are. It says little about what a stranger hears.

You can measure this properly with a browser and a spreadsheet.

Quick answer

To check what ChatGPT says about your business, write a fixed set of brand, category, comparison, shortlist ("best X for Y") and problem questions. Run each one in a fresh session with memory and history off, in ChatGPT, Perplexity, Claude, Gemini and Google's AI Overviews. Record the answer, the sources cited, whether you are named and which competitors are. Repeat monthly, because each answer is one sample from a system that varies.

Build a fixed query set

The query set is your measuring instrument. If it changes every month, you cannot compare one month with the next. Write it once and add to it only on purpose.

Cover five kinds of question. Brand queries ask about you directly, such as "What does [company] do?" or "Is [company] any good?" Category queries describe what you sell without naming you, such as "Who builds booking platforms for small hotel groups?" Comparison queries put you next to a named competitor.

"Best X for Y" queries are the shortlist questions buyers type when they are close to deciding. Think "best CRM for a small property management firm". Problem queries describe the pain with no category word at all, such as "How do I stop losing leads that come in over WhatsApp?"

Write them the way your buyers talk. Borrow phrasing from sales calls and enquiry forms. Fifteen to twenty-five queries is enough for a first pass. Weight them toward category, shortlist and problem queries. That is where buyers who have never heard of you decide who to consider.

Run every query in a clean session

Assistants personalise. ChatGPT and Claude can carry memory and past chats into a new answer. Gemini and Google can draw on your account and your location. If you run the audit logged in as yourself, you are partly measuring your own history.

Use a temporary or private chat where the product offers one, or a logged-out browser window. Start a new conversation for every query, so one answer does not shape the next. Note the date, the product, the model if it is shown, and the country you ran it from.

Google needs one extra step. Google's guide to AI features in Search says AI Overviews appear only when its systems judge them useful, and often do not appear at all. The same guide notes that AI Mode and AI Overviews may use different models, so their answers and links will vary. Record an empty result as "no overview shown", then run the same query in AI Mode.

Record the same five things every time

For each query, on each assistant, capture the same five things in the spreadsheet.

1. The answer itself, pasted in full.
2. The sources cited, with URLs. Perplexity, for example, includes numbered citations linking to the original sources with each answer.
3. Whether your business is named, and where in the answer it appears.
4. Whether what it says about you is accurate. Check services, location, pricing language and anything with a date on it.
5. Which competitors are named, and in what order.

Keep a free-text column for notes. The same third-party pages will start turning up in citations. That is often where the fix comes from.

Read the pattern

After one full pass, most queries fall into one of four buckets.

Named and accurate is the result you want. Note the cited sources. Those pages are holding your position up.

Named but wrong is the urgent bucket. The assistant knows you exist but describes an old service, a former location or a price you dropped. Check the cited source. Look for an outdated page on your own site or an old directory profile.

In the not-named bucket, competitors appear and you do not. Look at what was cited for them. If a competitor has a clear page answering exactly that question and you have nothing comparable, that is a content gap. If the assistant cited a list article or review site that includes them, that is a third-party gap.

The last bucket is answers that name you from a third-party page. That can be fine. It also means your description is only as current as that page, and you do not control it.

For Google, inclusion follows search eligibility. The same guide says a page must be indexed and eligible to show with a snippet to appear as a supporting link. If Google's answers miss you, check indexing first.

Why one check tells you very little

Every answer is generated fresh. Ask the same question twice and the sources can change, and sometimes so do the companies named. Products that search the web live also shift as new pages are indexed. One run is one sample.

So run each query more than once. We treat three runs per query on each assistant as the minimum for a first pass. Across six surfaces that is several hundred runs. Budget a few days. For the monthly repeat, one run per query is usually enough to spot movement, with three runs on any query whose result changed. Count how often you were named across all runs and use that share as your score. Being named in two of three runs is a different position from being named once in nine.

Then repeat the whole set monthly, with the same queries. The pages behind the answers change slowly, so monthly catches most movement. Also run it after a relaunch or a significant press mention.

What to fix first

We would fix wrong facts before anything else. A buyer told the wrong thing about you is worse off than one who never heard your name. Start with the cited source behind each wrong claim. Update or remove outdated pages on your own site, then correct the listings and profiles you control.

Next, close the content gaps where competitors were named and you were not. Write a page that answers the buyer's question directly in its first few lines, with specifics a model can quote. Schema and topic clusters are covered in The AEO Playbook: How to Get Cited by AI Answer Engines.

Third-party gaps come last, because they take the longest. List the articles and review sites that keep appearing in citations. Then work out where you can reasonably be included. This work belongs with the rest of your acquisition system, which we describe in The AI-Native Growth Stack.

If you would prefer to hand this over, our Answer Engine Optimisation assessment is the done-for-you version. Engagements in this shape typically run one week and $2,500 to $3,750. It ends with a written report and a three-month roadmap, then an optional monthly programme that tracks the score and ships the changes. It is part of our wider Digital Marketing practice.

Where to start

1. Write fifteen to twenty-five queries across the five types, using your buyers' own phrasing.
2. Set up a spreadsheet with one row per run and a column for each of the five things above.
3. Run every query three times on each assistant, each in a fresh private session.
4. Sort the results into the four buckets and list every wrong fact with its cited source.
5. Put the next run in your calendar for the same date next month, and keep the queries unchanged.

Frequently asked questions

How do I check what ChatGPT says about my business?

Open a fresh session with memory and history off, ask a fixed set of brand, category, comparison, shortlist ("best X for Y") and problem questions, and record each answer, the sources it cites and the competitors it names. Repeat the same set across several assistants and over several weeks, because single answers vary.

Why does ChatGPT give a different answer about my company each time?

Assistants generate each answer fresh, and many search the web live, so the wording, the sources and the companies named can change between runs. Treat each answer as one sample and look for the pattern across several runs.

How often should I check what AI assistants say about my brand?

Run the full query set once a month, and again after any significant change to your site or a major press mention. Monthly checks with the same questions turn scattered answers into a trend you can track.

What should I fix first if an AI assistant describes my business wrongly?

Find the source the assistant cited for the wrong claim and correct that page first, whether it is on your own site or a listing you control. Then make sure your own site states the correct facts plainly on a current page.