How to Track Brand Mentions in ChatGPT, Gemini and Perplexity

Keep the prompt set small and meaningful

Build prompts from real sales calls, support questions, search queries and competitor comparisons. Group them by intent, then choose a manageable sample that you can run consistently. Avoid dozens of near-identical prompts: they make the sample look larger without adding many independent buyer questions.

Keep branded questions in their own group. Asking whether a provider knows your company tests recall; asking for the best option for a problem tests discovery. Report those separately so a good branded answer does not disguise absence from category recommendations.

• Use one intent label per question.
• Preserve exact wording and language.
• Record date, interface and location when observable.
• Review the sample when the product or buyer changes.

Measure outcomes and referrals separately

For every response, capture mention, position in the answer if meaningful, recommendation language, cited URL, accuracy, and competitor set. Keep a citation even when it points to a third-party page about you. The page that influences an answer may not be on your own domain, which can reveal a proof or reputation gap.

Track referral sessions as a different signal. OpenAI documents a utm_source=chatgpt.com parameter on ChatGPT search referral links; analytics still depends on consent, attribution settings and the visitor reaching your site. A lack of recorded sessions does not prove that the assistant never mentioned you.

• Archive the raw answer and URLs.
• Separate cited-page traffic from answer visibility.
• Compare repeated runs before attributing a change to one edit.

A record that makes next month comparable

A practical record can be a table with one row per run: prompt ID, exact wording, intent, provider, interface, language, date, raw answer, mentioned brands, cited URLs, accuracy notes and the page hypothesis. Keep a link to the full answer or archive. Do not store a percentage without the numerator and the set of prompts it describes.

At review time, compare the same prompt IDs first, then describe any changes in the sample. If the tool changed its model or collection process, show a break in the series rather than joining incompatible results into one smooth chart.

• Keep a stable prompt ID and revision history.
• Archive raw outputs under an explicit retention policy.
• Document changes to provider or collection method.

Report coverage, not just a visibility percentage

If a report says the brand appeared in a share of runs, show how many prompts and runs produced that fraction, which providers were included and how missing answers were treated. A change in that sample can move the percentage even when the underlying answer pattern has not changed.

When summarizing for a founder or client, include one representative raw answer and one counterexample. This keeps the result grounded and shows where an aggregate hides differences between category discovery, brand recall and accurate recommendation.

• Show numerator, denominator and provider coverage.
• Include an example and an exception.
• Annotate sample changes over time.

Build a prompt set

Create prompts for your category, use cases, alternatives, competitors and buyer questions. Run them weekly in the same tools and record whether you appear.

Use Search Console to seed the prompt list. Queries like brand mentions in Gemini, track ChatGPT mentions, track mentions in Perplexity and how can I see if AI search engines mention my brand are strong signals because they describe the exact job users want done.

• Best tools for [category].
• [Competitor] alternatives.
• How to solve [pain].
• What is [category].
• Which brands does Gemini recommend for [job]?

Track the right fields

Record brand mentioned, position in answer, cited URL, competitors mentioned, and whether the answer describes your product correctly. Perplexity citations reveal source URLs clearly. Gemini is useful for conservative brand understanding. ChatGPT is useful for broad buyer-language prompts.

Fix missing mentions at the source

If your brand is missing, do not optimize the prompt. Improve the source pages: clearer category definition, stronger comparison pages, FAQ blocks, proof points, schema and internal links from pages that already have impressions.