How do I track my brand mentions in AI search engines like ChatGPT and Perplexity? Our approach

There is a habit we try to break early with every new client: opening ChatGPT, typing the brand name, and reading the answer as if it were a ranking report. It is a sample of one. The same question tomorrow may give a different answer, and Perplexity will probably say something else entirely.

Tracking brand mentions in AI search engines is a measurement exercise. Here is how we set it up.

Decide what a mention is

Before any tool, we write down the rules. Does a mention need the brand name exactly? What about product names, founders, or common misspellings? What if the engine names a competitor with a similar name?

In Promptwatch, which is the platform our client programs run on, this lives in the Brand Book, where analysis instructions define what counts, plus brand aliases and brand merging. It is dull work and it prevents bad data from the start.

Build the panel

We write prompts in three groups: category questions, comparison questions and direct brand questions. We organize them by topic and persona, and add country where the market is local. A panel of fewer than 30 prompts is too thin to trust, in our experience.

We run it daily on each engine the client's buyers actually use. For most of our clients that means ChatGPT and Perplexity at minimum, with Google AI Overviews and Gemini close behind. Promptwatch monitors the live-search platforms through their real interfaces, which matters because API answers can differ from what a person sees.

Separate mentions from citations

This is the distinction that most often gets lost.

A mention is the engine naming your brand in its answer.
A citation is the engine linking a source.

You can have either without the other. If Perplexity cites a review site that describes you but never names you in the answer, that is a different problem from being named with no link. We track both and read them side by side. Citation analytics give us page-level and domain-level views, and Page Tracker lets us watch particular URLs, ours or a competitor's.

Perplexity leans heavily on cited sources, so the offsite mentions report is valuable: it surfaces third-party pages that name the brand without a link.

Whether you appeared is a yes or no. Share of Voice, alongside Position in Answer, tells you how much of the conversation you hold against named competitors. We trend it monthly and use A/B period comparison, so a quiet week does not trigger a false alarm.

Check the plumbing

If a brand is missing from Perplexity, one possible cause is that PerplexityBot cannot read the site. Agent Analytics shows IP-verified crawler logs for PerplexityBot, Perplexity-User, GPTBot, OAI-SearchBot and ChatGPT-User, and error tracking with explanations. Crawler logs start on Professional ($245/mo), so on the Essential plan ($95/mo) this view is not available.

Tie it to traffic

Visitor analytics measures AI-referred sessions and conversions from ChatGPT, Perplexity, Gemini, Claude, Copilot and others. Volumes are often small, but we want to know whether those visits convert differently. We treat the numbers as directional, not precise.

What we do with the findings

Each month we pick the two or three highest-impact gaps from the Action Board and work them: a page to refresh, a third-party source to approach, a crawler block to remove. Then we measure again.

Honest limits

We cannot promise an engine will name a brand, and answers vary between runs. We also cannot see why a model chose a source, only which sources it cited. The work is reducing obstacles and strengthening evidence. If you want to try the approach cheaply, Promptwatch has a free Explore plan with 10 prompts on ChatGPT. Perplexity coverage needs a paid plan, starting at Essential, because every paid plan enables all active models.