AI visibility: check if ChatGPT, Gemini and Perplexity mention your business

AI visibility is how often AI assistants name your business when someone asks them about what you sell. If a customer asks ChatGPT for a reliable office cleaning company in Manchester and your company is in the answer, you are visible. If three competitors are named instead, you have work to do.

There is no ranking to look up, so the most direct way to find out is to ask the assistants yourself, the way your customers do. This tutorial gives you a free method for any small business: ten prompts, a spreadsheet and a couple of hours each month. Then come the free reports from Microsoft and Google, the paid trackers, and what to change when the results disappoint.

It belongs to my guide to generative engine optimisation, which explains how assistants choose the businesses they mention.

What AI visibility means for a small business

An assistant can treat your business in four ways, and your spreadsheet will record each of them:

• Named: your business appears in the answer.
• Linked: the answer links to your website, not only to a directory that lists you.
• Recommended: you are named first, or presented as a good fit for the question.
• Described correctly: your services, area, prices and contact details are right.

Being named with the wrong details can be worse than not being named at all: a customer may call about a service you stopped offering last year.

AI visibility is also different from AI traffic. Plenty of people read a name in an answer, then call or search for the business without clicking anything. In my experience, people would rather call than write because it is quicker, so a business can be visible in ChatGPT while its analytics show almost nothing. My tutorial on how to track AI traffic in GA4 covers the visits; this one covers the mentions.

Step 1: write ten prompts your customers would type

A prompt is the question someone types into an assistant. People write longer, more conversational prompts than Google searches, and they often describe their situation. Take an office cleaning company in Manchester:

Ten prompts are enough to start. More prompts give you more rows to fill, not necessarily more insight.

To find the words your customers use, look in four places:

• Your phone calls. Questions my clients ask on a first call, such as “Do you have experience?” or “How long will it take?”, are the kind of questions people now put to an assistant before calling anyone.
• Search Console. The searches that already show your site on Google reveal the words people use for your trade.
• Keyword costs. I use Semrush mainly for keyword research and for what keywords cost in ads. Advertisers pay most for the searches that bring business, so an expensive keyword usually makes a good prompt.
• Deep research. I use deep research tools to find what people in a niche complain about most and which words keep coming back. My method is in how to use Deep Research for copywriting and SEO.

Step 2: set up your spreadsheet

Create a spreadsheet in Google Sheets or Excel, with one tab per month and these columns:

Each answer is one row: ten prompts in ChatGPT, Gemini, Perplexity, AI Overviews and AI Mode make fifty rows a month.

Step 3: ask each assistant without memory or history

Your own account would skew the test: an assistant that remembers your chats, or knows you run the business, may mention you more readily than it would for a stranger. Test as a new customer would, logged out where you can, otherwise in a chat without memory, and start a new chat for every prompt so one answer does not colour the next.

ChatGPT

Open chatgpt.com in a private window and do not sign in. ChatGPT search has been open to logged-out users since February 2025, so a logged-out chat can still search the web.

An answer without sources most likely comes from what the model learned in training, not from a search. Write “none” in the Sources column: new pages and mentions can change a search answer fairly quickly, while training data only changes with a new model.

Gemini

In Gemini, if you use it with a Google account, start a Temporary Chat. Google says these chats are not used to personalise Gemini or to train its models, and they are kept for up to 72 hours. Gemini’s memory of past chats is switched on by default, one more reason not to test in an ordinary chat.

Perplexity

Perplexity works without an account, so a private window is enough. It numbers its sources and lists them with each answer, which makes the Sources column quick to fill.

Google AI Overviews and AI Mode

Search each prompt on Google in a private window. AI Overviews appear above the results for some searches, not all of them: when there is none, write “no overview” in the What it said column and No in Named. Then ask the same prompt in AI Mode, available in the UK since July 2025. Google’s AI answers draw on Google Search, so they are the ones your usual SEO work affects most directly.

If your customers use Microsoft Copilot or Claude, add them under the same rule: no history, no memory.

Location matters. Assistants adapt local answers to where they think you are. ChatGPT search, for one, can pass a general location based on your IP address to the search providers it uses (OpenAI’s help article on ChatGPT search). Run the test from your own area, or keep the town in every prompt.

Step 4: record what each answer says

Fill in each row as soon as the answer appears: by the fifth answer about office cleaning, they all start to blur.

• Named: Yes only if your business name appears. A link to a directory that lists you, without your name, counts as No.
• What it said: copy the sentence about you exactly, mistakes included, then compare it with your current services, area, prices, hours and phone number.
• Sources: note the domains cited: directories, review sites, local press, forum threads, your own site.

Watch the Accurate column closely for ChatGPT: in my experience it makes things up a bit too often, which is one reason I use Claude for my own work. When the same wrong detail keeps coming back, such as an old phone number or a service you dropped, it can usually be traced to a source, and a source can be corrected.

Then ask your five most important prompts a second time, in a new chat and on a new row. A second run shows whether a mention is stable or a fluke.

Step 5: work out your mention rate each month

Your mention rate is the share of answers that name you. Work it out for each assistant separately, because each one draws on different sources. With the assistant in column B and Yes or No in column D, this formula gives ChatGPT’s rate:

=COUNTIFS(B:B,"ChatGPT",D:D,"Yes")/COUNTIF(B:B,"ChatGPT")

Format the cell as a percentage and copy the formula for each assistant, changing the name. A rate of 30% means you were named in three answers out of ten.

Keep a summary tab with one line per month: the mention rate for each assistant, the number of inaccurate answers and the competitor named most often. Read it over three months rather than one. The same competitor named month after month, or the same mistake coming back, is a pattern worth acting on; a single odd answer is not.

Free reports from Microsoft and Google

Microsoft and Google now publish some AI data about your site for free. It covers their own AI features, so it says little about ChatGPT, the Gemini app or Perplexity, but it adds real figures to your manual test:

Grounding queries are the searches an assistant runs before it answers. They show which words led Copilot to your pages, and they make good prompts for your own test. Clarity’s Citations report is free for every Clarity user (Microsoft’s announcement).

Paid trackers: Semrush and Squarespace

Paid trackers run your prompts on a schedule and chart how often you are named. They save time and handle more prompts than you would by hand. Run the manual test for three months before paying for one, though: it is easy for a small team to fall for every new tool, and the channels that already bring you clients come first.

Semrush AI Visibility Toolkit

Semrush, owned by Adobe since April 2026, sells an AI Visibility Toolkit on its own and inside Semrush One, its bundle with the SEO Toolkit. According to Semrush’s knowledge base, the toolkit costs $99 (about £75) a month for one domain, with 25 prompts in Prompt Tracking, 50 more prompts for $60 (about £45) a month and no free trial. Semrush’s help names ChatGPT and Google’s AI Mode for prompt tracking, but I could not confirm Gemini or Perplexity: if those two matter to you, check the list before paying.

These are the prices at the time of writing (October 2026), taken from Semrush’s documentation. Semrush lists them in US dollars only, so the pound figures are my conversions at early October 2026 rates, not official prices: check the current offer, the amount you will actually be charged and whether $99 is the monthly or the annual-billing rate on Semrush’s pricing page. My review of whether Semrush is worth it covers the rest of the product.

Squarespace AI Visibility

If your site runs on Squarespace, as the sites I build for clients who edit their own pages usually do, look at its AI Visibility tool, released in July 2026. It sits in the SEO & AIO section of the dashboard, next to the SEO report. Squarespace drafts up to 12 example prompts from your site’s content and industry, which you can edit or add to. It then shows when your site appears in answers, for which prompts, and who appears when you do not.

For now it covers ChatGPT and Gemini, on English-language sites built on version 7.1 (Squarespace’s announcement). Squarespace says it comes with your plan at no extra cost, but its use is capped by the AI credits your plan includes, and that number varies from plan to plan: check what yours allows in your dashboard or in Squarespace’s help centre.

What to do with your results

The first two rows are the groundwork in my guide on how to get recommended by ChatGPT: crawler access, a clear page for each service and each town you really cover, and identical details on your site, your Google Business Profile and your directory listings. For the third row, my tutorial on how to write content for AI search shows how to write pages that assistants quote.

The Competitors column is the most useful one. On social media, I like Metricool because it shows what competitors do, especially the ones that are succeeding; this column does the same for AI answers. Take the competitor named most often and open the sources cited next to it: a directory, a review site, a local news article, a forum thread. Those are the places to be next.

Mentions on other sites take the longest to earn and count the most. Start with your town’s paper, then the regional press, then national titles. My guide on how to get press coverage explains how to pitch a journalist.

Your monthly AI visibility check

1. Duplicate last month’s tab and clear the answers, keeping the ten prompts.
2. Ask each prompt in ChatGPT, Gemini, Perplexity, AI Overviews and AI Mode, logged out or without memory, in a new chat each time.
3. Fill in every row, then ask your five main prompts a second time.
4. Work out the mention rate for each assistant and note the competitor named most often.
5. Look at Bing’s AI Performance report, Clarity’s Citations report and Search Console’s generative AI report.
6. Choose one fix for the month: a page, a correction, a directory listing or a pitch to a journalist.

You do not need me or an agency to run this check. If the results show you are missing from most answers and you would rather hand over the fixes, that groundwork is what my SEO service is built on.