How to track brand mentions in ChatGPT, Perplexity and Gemini

What "tracking mentions" means for an AI assistant

Search rank tracking has one number per keyword: your position on the page. An AI answer has no positions. It names a handful of products, describes each in a line, and often leans toward one. So tracking a mention means answering four questions, for a fixed set of buyer prompts, on a fixed schedule:

• Inclusion. Does the answer name your brand at all?
• Company. Which other brands are named next to you, and in what order?
• Description. How does the engine describe you, and is it accurate?
• Sources. For engines that cite, which pages did the answer draw on, and is one of them yours?

Everything below is a way to answer those four questions consistently enough that the trend is meaningful.

Step 1: build a question set your buyers would actually type

The mistake most people make is asking the assistant about their brand by name. That tells you whether the engine has heard of you, which is useful, but it is not how buyers discover products. Buyers ask about the category. A good set mixes both, and it stays identical from month to month so the results are comparable. Six questions is enough:

Swap in your own category, audience and market leader. Write the questions the way a person types, not the way a marketer writes: "best", "recommend", "alternatives to", "which should I use".

Step 2: know how each engine decides, then ask it

ChatGPT

ChatGPT draws on two things: what its model learned from the web at training time, and, when search is on, live pages it retrieves. Both reward a broad, consistent footprint. Brands described the same way across many credible pages get recalled accurately. Brands that exist mostly on their own site get omitted or described vaguely.

To track it by hand, open a fresh chat for each question so earlier answers do not colour the next one, ask the six questions, and record which brands appear and how yours is described. Answers vary between runs even when nothing changed, so ask twice if a result surprises you. The ChatGPT mention tracker runs the six questions against the same model family and shows each answer in full.

Perplexity

Perplexity is retrieval-first. It searches, reads the most authoritative pages it finds, writes an answer and cites them with visible links that users click. That makes a Perplexity mention behave like a search ranking with a referral attached, and it makes the citation list the most valuable part of the answer: it is a list of the exact pages you need to be on.

By hand: ask the six questions, expand the sources under each answer, and write down the domains. If your site is not there and a directory or a comparison article is, that is your target list. The Perplexity mention tracker does this through Perplexity's API and tallies the cited domains for you, flagging whether yours appeared.

Gemini

Gemini sits inside Google Search, Android, Chrome and Workspace, and many of its answers are grounded on live Google results. So Gemini's picks move with the web and lean on the same authority signals Google uses. It also cites grounding sources when it uses them.

By hand: ask the six questions in Gemini and note the sources it shows. Because grounding changes with the index, expect more movement here than with ChatGPT. The Gemini mention tracker asks through the Gemini API with Google Search grounding switched on and lists the sources it grounded on.

Step 3: score it so the trend is visible

A spreadsheet with one row per question and one column per month is all you need. For each cell record: named (yes or no), position among the brands named, sentiment, and the description in one line. Then roll it up into one number so you can see movement at a glance. The formula our trackers use is simple on purpose:

Weighting category mentions heavily is deliberate. Recognition without recommendation is where most startups sit, and it is the gap that costs sales.

Step 4: change what the engines read

You cannot edit an answer, but you can change the evidence it is built from. These are the levers, roughly in order of impact for an early-stage brand:

1. Be on the pages the engines read. Directory listings, review platforms, "best tools" roundups and comparison articles are the most cited content types in recommendation answers. A launch listing with a clear one-line description is a citable source; your own homepage alone rarely is.
2. Say the same sentence everywhere. Models echo the most repeated description of a brand across the web. Pick one line in the form "X is a Y for Z" and use it verbatim on every directory, social bio, press mention and docs page.
3. Publish comparison pages. "X vs Y" and "X alternatives" pages teach assistants your position in exactly the frame buyers ask about. Write them against the competitors your tracker run named most often.
4. Keep facts fresh and public. Perplexity and Gemini retrieve live pages and quote them. Public pricing, current docs and a changelog give them concrete sentences to lift. Gated or stale pages are skipped, or worse, quoted when wrong.
5. Let the crawlers in. Check robots.txt. GPTBot and OAI-SearchBot feed OpenAI, PerplexityBot feeds Perplexity, and Google-Extended controls whether Google may use your pages for Gemini. Blocking them removes you from the answers you are trying to enter.
6. Join real conversations. Reddit threads, Hacker News comments and niche forums carry heavy weight in training data and in retrieval. A genuine recommendation from a user in a thread often outranks a paid placement.
7. Treat launches as visibility events. A launch creates a burst of fresh, consistent coverage: announcements, listings, discussion, reviews. That is exactly the signal that shifts how assistants describe you, so time your content push around it.

One more from our own data: when a tracker run names competitors you have never heard of, look them up. They are usually brands that got into two or three well-cited roundups. That is the whole trick, and it is repeatable.

Free checks versus paid monitoring

The free trackers here answer "where do I stand today" for one engine and one brand at a time, with the actual answers in front of you. Paid tools exist for the next stage: dozens of prompts, several markets, weekly history and team dashboards. If you are evaluating them, this is roughly how they differ. Pricing changes often, so check their sites.

Our honest advice: run the free checks monthly until a mention or a lost mention has a dollar value you can point to. That is the moment paid monitoring pays for itself.

A cadence that works

• Monthly: the six questions on each engine, recorded in the sheet. Fifteen minutes with the trackers.
• After a launch or press: re-run within two weeks. Fresh, consistent coverage is what shifts descriptions, and you want to see it land.
• Quarterly: review the competitor list and the cited domains, then pick two sources to get onto and one comparison page to write.

Frequently asked questions