How do I track whether ChatGPT and other assistants mention my brand?
Pick a fixed set of 12 questions your customers ask without your brand name, plus two that name you, and ask them the same way every 14 days. Do it by hand in the apps, through the models’ APIs with a panel run, or with a hosted tracker. Log named, recommended and cited separately, count across many answers rather than trusting one, and add the citation reports Bing and Google publish for their own assistants.
Why every method is a sample
ChatGPT, Claude, Gemini, Perplexity and Copilot don’t publish which brands they name. So every tracker, paid or not, does the same thing: it asks questions and counts what comes back. Answers change between runs, between accounts and between the API and the app. Two rules follow:
- Fix the question set and ask it the same way each time, or the trend means nothing.
- Count a rate across many answers, such as “named in 9 of 42”, never a position in one answer.
Three ways to track, compared
| Method | What it sees | Cost | Main weakness |
|---|---|---|---|
| Manual question set | The real apps, logged out | Your time, about an hour a round | Slow; easy to drift from the protocol |
| API panel run | The models behind the apps, with web search | Pay per run for model usage | API answers differ from app answers |
| Hosted tracker | Usually daily runs across many prompts, with a dashboard | Monthly subscription, priced by prompts | You see data, rarely what to change |
| Bing and Google reports | Real citations in Copilot and in Google’s AI features | Included with the webmaster tools | Their own assistants only |
These aren’t exclusive. A small site can start by hand, add an API run when the manual round gets tedious, and leave the hosted dashboards to teams that need seats and client reports.
Method 1: a fixed question set, by hand
The full protocol is on how to check if ChatGPT recommends your site. In short:
- Write 12 questions a customer would ask without your brand, plus “What is [brand]?” and “What is [brand] ([domain])?”.
- Ask each in a new chat, logged out, with web search on, in every assistant your customers use.
- Log one row per question and assistant.
- Repeat every 14 days with the same questions.
Stay logged out. On 4 September 2026 a signed-in ChatGPT account placed one of our sites second for a buyer question, while a logged-out chat didn’t name it at all (own measurement).
Example: the log
date,assistant,q,named,recommended_first,cited_url,also_named
2026-10-22,chatgpt,q03,yes,no,none,"BrandA, BrandB"
2026-10-22,perplexity,q03,no,no,none,"BrandA"
2026-10-22,claude,t1,yes,n/a,https://yoursite.com/about,""
Keep named, recommended and cited apart. The “also named” column becomes your competitor list, and the cited URLs show which pages assistants read for your customers’ questions.
Method 2: an API panel
Sending the same questions to the models through their APIs, with web search on, turns an hour of copying into one command and makes repeats cheap. Ask each question more than once per run if you can afford it, and report a range. Read the results with care: in Surfer’s test of 1,000 prompts across five AI products, the brands named through the APIs overlapped with the scraped app answers by only 15.5% to 23.8%, rising to 21.3% to 31.6% after merging name variants (Surfer, updated 4 Oct 2026, read 15 Oct 2026; vendor study from a company that sells app-based tracking). Treat “were we named at all” as fairly reliable and “what position” as noise, and confirm your most important questions by hand.
Method 3: hosted trackers
Profound, Peec AI, Otterly.AI, Semrush, Ahrefs Brand Radar and others run prompts on a schedule and chart mentions, citations and competitors. They suit teams and agencies that need a dashboard and reports. Our comparison, with each vendor’s price as checked on its own page, is on Which AI visibility tool fits your situation?. Before you pay, ask whether the tool measures through the apps or the APIs, how many times each prompt runs, and whether you can export the raw answers.
Add the official reports
- Google: Search Console’s Generative AI performance report shows how your content performs in Google’s AI features (Google AI optimization guide, read 15 Oct 2026). Google also says no third-party tool has access to its internal ranking or AI systems.
- Bing: the AI Performance report in Bing Webmaster Tools shows total citations, cited pages and a sample of grounding queries for Copilot, Bing’s AI summaries and some partners (Bing, 10 Feb 2026).
- Your analytics: clicks from assistants, counted by referrer and
utm_source=chatgpt.com; see track ChatGPT traffic in GA4.
What Katman does here
Katman is an MCP server that runs inside your AI coding tool. katman_visibility_run asks your 12 + 2 questions to the models behind ChatGPT, Claude, Gemini, Perplexity and Grok through OpenRouter, on your own key: about $2.40 a run on the default panel, shown before anything is spent and capped at $3 by default. It logs named, recommended first and cited, plus who else was named. katman_manual_check records the answers you paste from the apps, with no API cost, and katman_setup_monitoring schedules runs in GitHub Actions. Katman Pro is $99 a month for up to 3 sites, with fix prompts and page briefs uncapped. Or have us do it: Growth writes 30 pages a month for you.
It isn’t a hosted dashboard with seats or client reports, and it never drives the consumer apps itself. Nobody can promise an assistant will recommend you; tracking tells you whether your fixes are showing up. How runs work is on visibility runs.
Written by the Katman team, Rnify Limited. Surfer, Google and Bing pages checked 15 October 2026.
Who this is not for
- Social listening for posts on Reddit, X or news sites: that’s a different kind of tool.
- Anyone who wants one AI visibility score: answers vary too much for one number to mean much.
- Sites AI bots can’t read yet: fix access first, because you’ll measure zero either way.
Frequently asked questions
Is it possible to track brand mentions in AI search at all?
Yes, as a sample. Assistants don’t publish who they mention, so every method asks questions and counts the answers. Track a rate across a fixed question set, not one answer.
What’s the difference between a mention and a citation?
A mention is your name in the answer text. A citation is one of your pages in the sources. You can have either without the other, so log them in separate columns.
Are API results the same as what ChatGPT users see?
No. In Surfer’s test of 1,000 prompts, brands named through the APIs overlapped with the apps’ answers by 15.5% to 23.8%, or 21.3% to 31.6% after merging name variants (updated 4 October 2026).
How many questions do I need?
Twelve unbranded plus two branded is enough for a small site. More questions help only if they are ones your customers really ask; padding the set with easy branded questions inflates the rate.
Do Google and Bing show AI citations themselves?
Yes. Search Console has a Generative AI performance report for Google’s AI features, and Bing Webmaster Tools has an AI Performance report for Copilot and Bing’s AI summaries.
What should I do when competitors are named and I’m not?
Read the pages the answers cite. They show what assistants rely on for that question, and usually which page you’re missing.