How to track your brand's mentions in ChatGPT

Why track brand mentions at all

When a prospect asks ChatGPT "what's a good self-hosted analytics tool?" or "which of these two vendors is better for a small team?", the model writes one synthesized answer. There is no list of ten links to scroll. Either your brand is named in that paragraph or it isn't — and if it isn't, you're invisible at the precise moment the buyer is forming a shortlist.

That influence happens before any click. A prospect can read a recommendation, narrow their options, and walk into a sales conversation already leaning toward a competitor, all without ever visiting your site. You never see the impression in your analytics, so the absence is silent. Tracking mentions is how you make that silence visible.

Be honest about the frame, though: this is not a promise of a traffic surge. AI referral traffic is still a small share of the web today, and anyone quoting you a specific uplift is guessing. The real near-term value is intelligence — knowing whether ChatGPT speaks well of you, ignores you, or recommends someone else — which is cheap to measure and expensive to keep ignoring.

Mention vs citation

Two outcomes look similar but are not the same, and conflating them produces noisy data. A mention is ChatGPT naming your brand in the body of its answer. A citation is your domain appearing as a linked source the model drew from. They move independently.

ChatGPT can describe your product accurately from what it learned during training without ever linking to you — a mention with no citation. It can also lean on a third-party review or comparison that talks about you, citing that source while never naming you in the prose — a citation with no mention. Both signals carry information: mentions tell you how the model talks about your category; citations tell you which sources it trusts to build that answer.

Measure them separately. A visibility rate counts how often you're mentioned; a citation rate counts how often your domain is cited. When the two diverge — say you're mentioned often but rarely cited — that itself is a finding, usually pointing at where you need stronger first-party or third-party presence.

Build a prompt set that mirrors real questions

Your measurement is only as good as the questions you ask. A prompt set should mirror how buyers actually talk to an assistant in your category, not the flattering phrasing you wish they used. Asking "why is [your brand] the best?" is a vanity prompt: it puts your name in the question, so a mention in the answer proves nothing. The useful prompts are the ones where your name is not given, and you're watching to see whether the model brings it up on its own.

Cover the categories of questions that lead to a decision:

• Comparison — "[competitor A] vs [competitor B] for a small team" — does the model add you to the comparison unprompted?
• Alternatives — "alternatives to [well-known tool]" — are you on the list of substitutes?
• Best-of — "best [category] tool for [use case]" — do you make the recommended shortlist?
• How-to — "how do I [job your product does]?" — are you named as a way to get it done?

Keep the set fixed and representative — a few dozen prompts spanning these categories is far more useful than hundreds of near-duplicates. Stability matters more than volume, because you'll be re-running the same prompts over time and comparing like with like.

Probe on a schedule, not once

Asking ChatGPT a question once tells you almost nothing. The model samples its responses, so the same prompt can name you in one run and skip you in the next. Retrieval and the underlying model also change over weeks and months. A single snapshot is a coin flip dressed up as a measurement.

The fix is repetition. Run each prompt several times to average out run-to-run variance, and re-run the whole set on a cadence — daily or weekly — so you can watch the trend rather than react to one lucky or unlucky answer. The trend is the signal; the single answer is noise.

Do this through the official API, sending your prompts and recording the answers and any citations returned. Never scrape the ChatGPT web interface: it's brittle, it breaks the terms of service, and it gives you data you can't stand behind. This is exactly the loop runs — a fixed prompt-set, sampled repeatedly on a schedule through official engine APIs, turned into metrics over time.

Read the signal

Once you're collecting answers consistently, a few plain metrics do most of the work. Visibility rate is the share of probes where you're mentioned at all — your baseline presence. Watch its trend, not its absolute value on any one day.

Then look at position within the answer. Being named first in a recommended shortlist is very different from being a footnote in the last sentence. ChatGPT's ordering tends to mirror how confidently it associates you with the question, so a low but consistent position is a different problem than not appearing at all.

Finally, track who is named instead of you. Every prompt where a competitor appears and you don't is a concrete gap. Aggregated across your set, the competitors that keep surfacing reveal your real share-of-voice in AI answers — and a shortlist of exactly where you're losing the conversation.

Act on the gaps

A gap is only useful if it points to an action. When ChatGPT names a competitor instead of you, the question to ask is why the model reached for them — and the answer usually lives in the sources it leaned on. Where citations are available, look at which domains recur in the answers that exclude you: comparison sites, directories, documentation, independent reviews. Those are the places the model trusts, and the places where you're either absent or poorly represented.

From there the work is concrete: be accurately and clearly represented on the sources that keep getting cited, and make your own pages the cleanest available answer to the prompts you're losing — direct, well-structured, current. Then re-run the set and check whether visibility, position, and share-of-voice actually moved. A few prompts you can put to work:

• "For the prompt where [competitor] is recommended and we aren't, list the sources you used to form that answer."
• "What would a page need to say to be a credible answer to [buyer question] in this category?"
• "Which independent sources discuss [our category], and how are we represented there versus [competitor]?"

None of this is a trick. Tracking mentions in ChatGPT is the discipline of watching the answers, finding the gaps, earning presence on the sources that fill them, and measuring again — turning scattered, invisible impressions into a trend you can act on.