How to Track ChatGPT Brand Mentions

Why ChatGPT brand mentions are invisible to your analytics

A buyer opens ChatGPT and asks which vendor to use for a problem you solve. ChatGPT writes three paragraphs naming four companies. The buyer reads it, forms a shortlist, and never clicks anything.

Nothing about that exchange reaches you. There is no session, no referrer, no impression, no query in Search Console. If you were named, you gained consideration you cannot attribute. If you were not, you lost a deal you will never know existed. This is the structural reason brands end up invisible in AI answers without noticing for quarters at a time.

Referral traffic from chatgpt.com is a weak proxy for the same reason. It captures the minority of readers who click a citation, and only when a citation exists at all. A brand can be mentioned favourably in hundreds of answers a week and see almost no referral sessions.

So the measurement has to move to the generation side: ask the model the questions your buyers ask, and read what it says. That is what a ChatGPT mention tracker does, and it is the practice behind prompt monitoring.

What a ChatGPT mention tracker has to capture

A tracker that only returns "mentioned: yes/no" is close to useless — it tells you there is a problem but nothing about what to do. Four fields per answer make the data actionable.

The fourth field is the one teams skip and later regret. Mention rate tells you where you stand; the source list tells you what to change. If the same competitor comparison page is cited in eleven of your thirty answers, you have found your single highest-leverage piece of work.

The five-step method

1. Build a fixed prompt set

Write 30 to 100 prompts phrased the way a buyer phrases them — describing a problem, not naming a vendor. "Best tool for tracking brand mentions in AI assistants" is a valid prompt. "Tell me about [your brand]" is not: it guarantees a mention and measures nothing.

Cover four intent types: category discovery ("what tools do X"), comparison ("X vs Y for a 50-person team"), problem-first ("how do I solve Z"), and qualifier-led ("GDPR-compliant option for a European company"). That last type is where under-served positioning shows up fastest.

Then freeze the list. A prompt set you keep editing produces a trend line that means nothing.

2. Run in clean sessions

Personalisation is the most common source of false confidence. Run prompts logged out, or with memory and chat history disabled. Your own account has been asking about your own category for months, and ChatGPT has learned from it. A founder checking from their own account almost always sees a flattering answer no buyer would ever get.

3. Score every answer on the four fields

Mention, position, framing, sources. Framing is best scored on a five-level scale from strongly negative to strongly positive rather than a binary, because most mentions are neutral list entries and neutral is a distinct — and very common — outcome worth tracking on its own.

4. Log the citations

Save every URL, then aggregate across the whole run. Rank sources by how many answers they influenced. This turns a monitoring exercise into a content backlog: the top ten domains in that list are your actual competitive set inside AI answers, and they are frequently not the competitors you track in SEO. Our breakdown of how LLM citations work covers what makes a source retrievable in the first place, and the Reddit citation analysis explains why community threads punch so far above their weight here.

5. Re-run weekly

The same prompt asked twice in one afternoon can return a different set of brands. Generation is probabilistic and the browsing layer retrieves live results. One run is a sample, not a measurement. Weekly runs over a frozen prompt set are the minimum cadence at which a five-point move means something real.

Doing it by hand vs using a ChatGPT mention tracker

The manual version works, and it is the right way to start. Fifty prompts, a spreadsheet, an afternoon. You will learn more about your category's AI answers in that afternoon than from any vendor demo.

It stops working at the second week. Fifty prompts run weekly across five assistants is 250 answers to read, score and source-log every single week, and the value only appears once you have eight or ten weeks of trend. That is where the exercise reliably dies.

If you are comparing vendors, our comparison against Otterly and against Peec AI cover the monitoring tools most often shortlisted, and Rankio vs SEO tools explains why rank-tracking suites answer a different question entirely.

Tracking ChatGPT alone is not enough

ChatGPT is the assistant everyone measures first, and measuring only ChatGPT is the most common way to get a confident wrong answer. Each assistant retrieves from a different source mix, so mention rates diverge sharply between them.

Perplexity and Google AI Mode cite inline and heavily, which makes them the easiest place to start diagnosing sources. Gemini and Claude behave differently again on the same prompts.

For any brand selling in France or French-speaking Europe, Le Chat (Mistral) is the omission that matters most. Le Chat weights French-language sources far more heavily than ChatGPT does, which means a French company with strong trade-press coverage in French can be well-covered on Le Chat and near-absent from ChatGPT — or an English-first brand can be strong on ChatGPT and invisible to the assistant a growing share of French buyers now use. A ChatGPT-only tracker cannot see either case. Our guide on getting cited by Le Chat covers what changes.

GDPR and ChatGPT mention monitoring

Mention tracking stores raw generated text. For smaller companies, that text names people constantly — founders, executives, and sometimes customers. A monitoring database of AI answers about a 40-person company is, in practice, a database containing personal data.

Most teams never ask about this until procurement does. The three questions worth asking any vendor: where are raw model outputs stored, how long are they retained, and does the processing itself happen inside the EU. Our GDPR compliance page sets out how Rankio answers each of them.

ChatGPT mention tracking checklist

• 30-100 buyer-intent prompts, none of them naming your brand
• Prompt set frozen and versioned before the first run
• Logged-out or memory-disabled sessions on every run
• Four fields logged per answer: mention, position, framing, sources
• Cited URLs aggregated and ranked by influence across the run
• Weekly cadence, with at least eight weeks before drawing conclusions
• The same prompt set run on Perplexity, AI Mode, Gemini, Claude and Le Chat
• Vendor questions answered on storage location, retention and EU processing

Frequently asked questions