AI visibility
How to track whether AI assistants recommend you
Write down the questions a buyer would ask an assistant before choosing a product like yours. Ask each assistant the same questions on a schedule. Record whether you were named, where in the answer, who else was named, and which pages were cited. One check tells you nothing, because answers vary from one run to the next. The trend over a few weeks is the measurement.
By hand, with a spreadsheet
This works for up to about ten questions and is the right way to start, because writing the questions teaches you more than any dashboard.
- Write 10 questions. Mix three kinds: a category question ("best invoicing software for freelancers"), a problem question ("how do I chase a late invoice without annoying the client"), and a comparison ("Xero or FreeAgent for a sole trader"). Do not put your own name in them. You are testing whether you come up unprompted.
- Pick your assistants. ChatGPT, Gemini, Perplexity and Claude cover most use. Turn web search on where it is a setting.
- Ask in a fresh chat, signed out or in a private window. A signed-in assistant bends its answers to your history, and yours is full of your own company.
- Record five things per answer. Were you named? In what position among the businesses named? Who else was named? Which pages were cited? Was anything said about you wrong?
- Repeat weekly, same questions, same wording. Change the list only by adding to it.
Or run three of them right now:
Run three questions now
After four weeks you have a mention rate (answers that named you, out of answers asked), a list of the competitors that keep appearing, and a list of pages the assistants keep citing. That last list is your to-do list: those are the pages to get mentioned on.
Free data you already have
Bing Webmaster Tools. In February 2026 Bing added an AI Performance report, in public preview. It shows how often your pages are cited in Copilot, Bing's AI summaries and some partner products, and a sample of the queries behind them. As far as we know it is the only citation data any of these companies gives site owners for free. It counts citations and says nothing about where in the answer you appeared.
Google Search Console. Clicks and impressions from AI Overviews and AI Mode are included in the Performance report, under the Web search type, according to Google's guidance. So AI traffic from Google is in your numbers already, mixed in with ordinary search.
Your server logs. They show which AI crawlers visit and which pages they read. If the search crawlers never fetch your most important pages, start there.
The four numbers
- Mention rate. The share of answers that name you. The headline number.
- Position. Where you fall among the businesses named. First is worth far more than fifth.
- Share of voice. Your mentions as a share of all brand mentions across the same answers. It tells you whether a competitor is growing faster than you are.
- Citations. How often one of your own pages is linked as a source. You can be mentioned without being cited (the assistant read about you elsewhere) and cited without being mentioned.
What makes the numbers lie
Too few runs. Ask the same question twice and you can get two different lists. Judge a question on several runs over a few weeks, not on one.
Personalisation. Location and chat history change answers. Check from a clean session each time, and expect customers in other countries to see something else.
Your own name in the question. "Is Acme any good" will always mention Acme. It measures what assistants say about you, which is worth knowing, and not whether they recommend you.
The tool is not the app. This one applies to us. Our tracker asks through each company's API with web search switched on. That is repeatable and affordable, and it is not exactly what a person sees in the app, where the answer is shaped by their history and location and sometimes by a different model. Trackers that drive the real apps get closer to that and cost more to run. Neither is the truth. Both are consistent enough to show a trend, which is the thing you are buying.
How often to check
Weekly is enough for most businesses, and it is our default. Daily checks mostly measure noise unless you have dozens of questions and enough answers to average. Spend the effort on choosing better questions.
When a paid tracker is worth it
When the spreadsheet takes more than an hour a week, usually past 10 to 15 questions on four assistants. Prices on 14 November 2026, from each company's own pricing page: Otterly starts at €29 a month for 16 questions, and Promptwatch at $95 a month for 50. Both are dedicated trackers with more depth on measurement than we have.
We sell a tracker too, so weigh this accordingly. Ours is part of a product that also writes and publishes the articles, so a question where you are missing can be turned into a planned article. If you only want measurement, a dedicated tracker is the simpler buy.
Questions
- How many questions should I track?
- Start with 10. Enough to cover your main product and your two or three most common buyer problems. Add more when the first ten stop teaching you anything.
- Can I see which questions people really ask AI about my business?
- Not directly. None of the assistants publishes query data the way Google does. Bing's AI Performance report shows a sample of queries behind your citations, which is the closest thing available.
- My mention rate dropped this week. Should I worry?
- Not on one week. Look at whether the same competitors gained and whether the cited pages changed. A model update can reshuffle every answer in a category at once, and that is not something you did.
Sources
Each one was opened and read on 14 November 2026. Figures in the text say what was counted, how many and when.
Content Autopilot team
We make Content Autopilot, an SEO and AI visibility product
Our guides are drafted with AI from the sources listed above, and every figure is checked against its source before it is published. Where we sell something a guide discusses, the guide says so.