How do I track if my brand is showing up in Perplexity and Gemini answers?

The way we track Perplexity and Gemini for clients is simple to describe. We pick a fixed list of the questions buyers actually ask, run them on a schedule in both engines, and keep a record of what comes back and which sources are cited. A single check proves nothing, because assistants can word an answer differently each time. A pattern across many prompts and many days does.

Two things to settle before you start

Gemini is not Google AI Overviews or AI Mode. They share a parent company and are different surfaces, with different behavior. We track them as separate engines so a client can see exactly where they are strong or weak.

Perplexity shows its sources prominently. That makes it a good place to learn what gets cited. It also means you can be mentioned without being linked, so we look at both.

Building the prompt list

We build roughly 20 to 30 prompts per client to start, in three groups.

Category questions. "Best tools for X for a team of Y."
Comparisons. "[Brand] vs [competitor]," "alternatives to [competitor]."
Brand questions. "What is [brand]," "is [brand] good for Z," "how much does [brand] cost."

The sales team's call notes and the support inbox are the best raw material, because they use the buyer's real wording. Prompt Explorer in Promptwatch can suggest more from Search Console queries, ranking keywords and People Also Ask, and personas with persona stacking let us phrase the same question as different kinds of buyer. We add a topic tag and a prompt type to each, so reports can be filtered later.

The manual method, and why we outgrow it

For a first look, you can ask each prompt in Perplexity and Gemini yourself, in a clean session, and record the result in a sheet: mentioned or not, position, description, sources. We do this on the first day of an engagement because it builds intuition.

It stops being workable fast. Thirty prompts across two engines is sixty checks a week, with no way to run them per country or language and no consistent history. After the first month we move to automation.

What we run it on

We use Promptwatch for client programs. It monitors the real interfaces of ChatGPT (Search), Google AI Overviews, Google AI Mode, Copilot, Gemini, Perplexity, Amazon Alexa and Meta AI, and runs prompt panels daily per model, country and language. Paid plans include all active models, which matters if you were hoping to pay extra only for Perplexity or Gemini.

What we use most for this question:

Visibility Score, Share of Voice, Sentiment and Position in Answer, reported per model, so "strong in Perplexity, weak in Gemini" shows up as a clear difference.
Citation analytics at page and domain level, so we see which pages each engine cites and which competitors replace us.
Heatmaps comparing any two dimensions, such as brands against models or topics against countries.
Prompt trends over time, and A/B period comparison to check whether a change helped.
Offsite mentions, to find where the brand is named on third-party pages without a link.

Beyond the mention: access and traffic

A mention is not the end of the story, and two checks follow it.

Can the bots reach the site? Agent Analytics provides real-time, IP-verified AI crawler logs, including PerplexityBot and Perplexity-User, and Google-Agent and GoogleOther on the Google side. A crawl-to-citation path shows a citation rate per page, and error tracking explains failures. Crawler logs start on the Professional plan, so we flag that in scoping.

Does it send people? Visitor analytics tracks AI-referred traffic and conversions from Perplexity, Gemini, ChatGPT and others, using a lightweight script, a GTM template or the WordPress plugin. We install it early so we have a baseline before any change.

What we report

For each client, a short monthly readout covers:

Share of voice per engine, with the trend.
Prompts where a competitor is named and we are not.
The pages and domains cited in those answers.
Wrong or outdated descriptions found.
Visits and conversions from AI referrers.
Three recommended actions, tracked on the Action Board.

Honest limits

We tell clients up front that tracking does not guarantee a mention. Answers vary between askings, and personalization can change what an individual user sees. The tracker shows an aggregate picture across many runs, which is useful, but it is not a replica of any single person's session.

If you are starting today

Write 25 prompts, run them by hand in both engines this week, and look at which sources get cited. Then decide whether the pattern is worth automating. Most teams find that the cited sources teach them more than the mention counts do.