Measurement 9 min read
What Does AI Say About My Business? A 30-Minute Check Across ChatGPT, Gemini, Perplexity, Claude and Copilot
Before you try to change what AI says about you, write down what it says now. Twelve questions, five assistants, one sheet, and a way to tell a wrong fact from a missing one.
Founder, SEO·AEO·GEO
Builds SEO·AEO·GEO and writes the crawler, schema, and citation research published here.
Key takeaways
- Ask three kinds of question: who you are, who’s best in your category, and you versus a named competitor. Each one fails for a different reason.
- Use a logged-out or temporary chat and run every question at least three times; one answer is a sample, not a verdict.
- Score each answer against a one-page fact sheet: wrong, missing, outdated or right. Wrong facts and missing mentions have different fixes.
- Write down the sources each answer cites. They are the pages you need to fix or get onto, and they differ from one assistant to the next.
To find out what AI says about your business, ask each assistant three kinds of question (who you are, who is best in your category, and how you compare with a named competitor), run each one at least three times in a logged-out or temporary chat, and score every answer against a short sheet of facts you know to be true. Then write down the pages each answer cites. That list tells you what to fix, and where.
It takes about 30 minutes for five assistants. It’s worth doing before any “AI visibility” work, because the fix for a wrong fact is different from the fix for a missing mention, and you can’t tell which one you have from a single screenshot.
Before you start: write your fact sheet
You can only grade an answer against something. Put these on one page, exactly as you’d want an assistant to say them:
- One sentence that says what you do, for whom and where (“a family dental practice in Leeds for nervous patients”).
- Your legal and trading names, and any old names you’ve used.
- Address, service area, phone, hours, and the year you started.
- Your main services or products, and a starting price where you publish one.
- Who runs the business, and any licences or accreditations.
- Two or three competitors your customers compare you with.
The 12 questions to ask
Ask them the way a customer would, in plain words. Change the brackets to your details; keep everything else the same on every assistant so the answers are comparable.
| Type | Questions | What it reveals |
|---|---|---|
| Identity | “What is [business]?” · “Where is [business] and when is it open?” · “Who owns [business]?” · “How much does [business] charge for [service]?” | Whether the facts AI holds about you are right, and where they come from |
| Category | “Best [service] in [city]” · “Who should I hire for [job]?” · “[Service] for [customer type] near [area]” · “Affordable [service] in [city]” | Whether you’re named at all when nobody types your name: the question that brings new customers |
| Comparison | “[You] vs [competitor]” · “Alternatives to [competitor]” · “Is [business] any good?” · “What do reviews say about [business]?” | How you’re described next to others, and which reviews and opinions get repeated |
Where to ask, and what to look for in each assistant
| Assistant | Where its answer comes from | What to record |
|---|---|---|
| ChatGPT | Its training, or a web search when the question needs current information. OpenAI says search rewrites your question into queries for search providers. | Whether it searched (inline citations and a Sources button), and every cited page |
| Google AI Overviews and AI Mode | Google’s search index and ranking systems, plus Business Profiles for local questions | The links shown beside the overview; your Business Profile details if they appear |
| Gemini app | Its training, plus Google Search when it grounds an answer | The Sources button under the answer (not every answer has one) |
| Perplexity | Its own search index, built by PerplexityBot, with numbered citations on almost every answer | All numbered sources, including the ones it lists but doesn’t quote |
| Claude | Its training, or a web search when you ask for current information | Whether it searched, and the cited pages |
| Microsoft Copilot | Bing’s index | The cited links and the “show all” sources list |
The assistants rarely agree. When Custom Legal Marketing ran 3,200 prompts about law firms, ChatGPT, Claude and Google AI Mode seldom recommended the same firms (vendor study). So score each assistant separately, and don’t assume a good ChatGPT result carries over.
How to score the answers
Give every fact in every answer one of four marks. A simple spreadsheet works: one row per answer, one column per fact on your sheet.
| Mark | Example | Usual cause | Usual fix |
|---|---|---|---|
| Right | Correct hours, correct description | Your facts are consistent where it reads | Keep them current |
| Outdated | Old address, old prices, a former owner | A stale page that still ranks, or old training data | Update or redirect the stale page; correct the directory listing |
| Wrong | Wrong services, mixed up with a similarly named business | A cited page is wrong, or your name is ambiguous | Correct the source; make your name and description unmistakable |
| Missing | Not named for “best [service] in [city]” | You aren’t on the lists and directories it reads, or it can’t reach your site | Get onto the cited pages; check crawler access |
For category questions, also count how often you appear across the three runs, and in what place. Being named third in two runs out of three is a different position from being named first once.
The part most people skip: the sources
Every cited page is a lead. Sort them into five piles: your own pages, directories and review sites, “best X” lists and comparison articles, news and blogs, and forums. The pile that keeps coming up for category questions is where your next month of work goes. If your own site never appears even for identity questions, that points to access or readability: an assistant can’t cite a page its crawler was refused or couldn’t read. Our [free audit](/) checks that for each AI crawler.
If an answer has no sources at all, the assistant answered from training. You can’t edit that directly; it changes as the web changes and as models are retrained. That’s covered in how to fix wrong information in ChatGPT.
Add the data the engines report themselves
- Google Search Console’s generative AI report shows impressions in AI Overviews and AI Mode (impressions only, no clicks). See how the report works.
- Bing Webmaster Tools’ AI Performance report shows how often your pages are cited in Copilot and Bing’s AI summaries, and the grounding queries behind those citations. See reading Bing's AI Performance report.
- GA4’s AI Assistant channel counts visits from ChatGPT, Perplexity, Gemini and others (it leaves out Google’s own AI Overviews). See the GA4 guide.
Make it a habit, not a one-off
Repeat the category questions every month on the same days, and the identity questions whenever you change something important: a move, a price change, a new service. Our AI visibility checker runs buyer questions for you and shows who is recommended instead; the dashboard tracks the same questions weekly. The full measurement method, with what to report to a boss or client, is in how to measure AI visibility.
If the check shows a competitor named again and again where you aren’t, the next read is why ChatGPT recommends your competitor, which turns the source list into an outreach plan.
FAQ
How do I find out what ChatGPT says about my business?+
Open a temporary chat (or a logged-out window), ask “What is [business name] and what does it do?”, then ask the category question your customers ask, such as “best [service] in [city]”. Repeat each question three times, open the Sources list, and note every fact and every cited page. Do the same in Gemini, Perplexity, Claude and Copilot.
Why does AI give a different answer every time I ask?+
The models sample their wording, and when they search, the pages they retrieve change too. In SparkToro and Gumshoe’s test of 2,961 runs, there was less than a 1-in-100 chance of getting the same list of brands back twice. That is why the check uses repeated runs and counts how often you appear.
Is there a tool that shows what AI says about my business?+
Several trackers run your questions on a schedule. Our free AI visibility checker asks buyer-style questions and shows whether you are named and in what place (and, on engines that search the web, which sites are cited), and the dashboard repeats the same questions weekly. A manual check like the one in this guide is still worth doing once, because you will read the answers in full.
What should I do if AI says something wrong about my business?+
Find where the wrong fact comes from first. If the answer cites a page, fix or correct that page. If it cites nothing, the model is answering from what it learned in training, and the fix is to make the correct fact consistent and prominent across your own site and the profiles that describe you. Our guide to fixing wrong AI information covers each case.
Sources
- AI recommendations change with nearly every query (SparkToro and Gumshoe study) — Search Engine Journal (2026-01-30)
- ChatGPT search (Help Center; read via search snippets) — OpenAI
- Sources in Gemini Apps responses — Google Gemini Help
- AI Overviews on Google Search: feedback and reporting — Google Search Help
- Introducing AI Performance in Bing Webmaster Tools (Public Preview) — Microsoft Bing Webmaster Blog (2026-02-10)
- Search Console generative AI performance report — Google Search Console Help
- ChatGPT, Claude and Google AI Mode rarely recommend the same law firms (3,200 prompts) — vendor study — Custom Legal Marketing (CLM Sequoia) (2026-07-29)
Run it on your domain
See this lens as a finding, not a lecture.
Free technical GEO audit. Evidence-traced scores and an LLM-ready export. See your score with no login.
Audit your website
