How to track whether ChatGPT, Gemini and Perplexity mention your brand

What you are actually measuring

"Are we in ChatGPT?" is not one question. AI visibility is a set of measurements, and each answers something different.

• Visibility (mention rate). Of the questions you track, in what share is your brand named at all? This is the headline number.
• Share of voice. Of all brand mentions across your questions, what portion are yours compared with each competitor's? It shows whether you are gaining ground, not only whether you appear.
• Position. When you are named, are you first, third or last in the list? Order matters because readers and assistants weight the first names more.
• Sentiment. Is the mention positive, neutral or negative, and what does the answer say about you? Being named as "good for small teams but limited" is different from being named as "the best option".
• Citations. Which pages did the engine cite? These are the sources that decide the answer and the places to work on.
• Who is named instead. For questions where you are missing, which brands are present? This is your real competition in AI answers, which may differ from your competition in search.

You do not need all of them on day one. Start with visibility and citations, and add the rest once you have a baseline.

Build the question list

Everything depends on the questions, so spend time here.

Write them as a customer would type them, not as a marketer would label them. "Best tool for making faceless YouTube videos" beats "AI video generation software".

Cover the whole buying journey. A good list mixes:

• Category questions: "best X for Y".
• Comparison questions: "X vs Y", "alternatives to Z".
• Problem questions: "how do I stop doing this by hand".
• Price and fit questions: "how much does X cost", "is X good for small teams".
• Brand questions: "what is [your brand]" and "is [your brand] legit", which show what the engines say when someone asks about you directly.

Aim for 30 to 100 questions to start. Fewer than about twenty is too noisy to read, and more than a few hundred is hard to act on.

Keep the wording fixed. Changing a question changes the retrieved sources and the answer, so you cannot compare across weeks. Add new questions as new rows rather than editing old ones.

Include your language and market. Answers differ by language, and sometimes by country. If you sell in more than one, track each.

Which engines to cover, and how they differ

Different engines retrieve and cite differently, so cover each that your customers use rather than assuming one stands in for all.

• ChatGPT answers from its training and, when it searches, from the web. It often cites a handful of sources.
• Google AI Overviews are tied to Google Search, so pages that rank there matter most.
• Gemini is Google's assistant and can draw on Google's index.
• Claude can search the web when enabled and answers from training otherwise.
• Perplexity is built around live search and shows its sources prominently, which makes it the easiest to analyse.
• Grok and Meta's Muse Spark add further audiences, particularly on social platforms.

Their source mixes differ. Profound's data on four billion citations found Reddit ranked first on Perplexity and second on ChatGPT, Google AI Overviews and Grok, and much lower on Microsoft Copilot, so the same effort can pay off differently by engine. AppearsIn checks ChatGPT, Google AI Overview, Gemini, Claude, Perplexity, Grok and Meta Muse Spark.

Running the checks by hand

You can start with a spreadsheet and an hour.

1. For each question and each engine, open a fresh session with no memory or custom instructions that could bias the answer, and use a logged-out or default setting where you can.
2. Ask the question exactly as written. Copy and paste it.
3. Record: the date, the engine, whether your brand is named, the order of brands, any wording about you, and the cited URLs.
4. If the engine shows sources, open the cited pages and note which are Reddit threads, review sites, lists, videos or your own.
5. Repeat the same list on the same schedule.

A simple layout is one row per question and engine per date, with columns for named (yes or no), position, sentiment, competitors named, and sources. After two or three runs you will see which answers are stable and which move.

By hand this is realistic up to a few dozen questions on two or three engines. Beyond that it becomes slow and error-prone, which is where a tool earns its cost.

Why answers vary, and how to handle it

AI answers are not fixed. The same question can give different answers minutes apart, for several reasons: the model generates text with some randomness, the retrieved sources change as pages and rankings change, and engines update their models and policies without notice.

Treat a single reading as a sample. To get a usable number:

• Run each question more than once if the stakes justify it, and count how often you appear instead of whether you appeared.
• Track trends, not points. A rise from 20% to 35% visibility over two months is a signal. A move from 30% to 28% in a week is noise.
• Compare like with like. Same questions, same engines, same settings, same time of day if you can.
• Watch for step changes. A sudden shift across all questions usually means an engine changed, not that your work took effect.
• Do not read too much into one answer that names you or leaves you out.

Being honest about the noise is what makes the data trustworthy to the people who have to act on it.

Turning the results into work

Measurement only matters if it changes what you do. Four patterns come up repeatedly.

You are missing, and the same sources are cited everywhere. Those pages decide the answer. List them, find out who controls them and whether your profile or mention is accurate, and look for ways to be included honestly: a correction, a profile update, a useful reply in a thread, a briefing for a list author.

You are named, but with poor wording or sentiment. Find the source of that wording. It is often an old review, a competitor's comparison page or an outdated post. Fix what you can, and publish accurate pages that answer the same question.

A competitor is named for a question you should win. Read what the sources say about them for that need. Either your own page lacks a direct answer, or the evidence that you serve it is missing.

You are named on one engine and not another. The difference usually traces to different sources. Look at what the second engine cites.

For Reddit specifically, the cited threads are the work list: threads that rank and are cited are the ones where an honest, helpful reply from someone who knows the product can matter. The guide to finding them covers the search steps.

How often to check, and how to report it

Weekly is a good cadence for a tracked list, because it is frequent enough to see trends and slow enough that noise averages out. Monthly is enough for slow-moving categories. Daily checks mostly measure noise.

Report three numbers to people who do not live in the data: visibility (share of questions where you are named), share of voice against two or three competitors, and the top five cited sources with what you did about each. Add a short note on what changed and why you think so. A simple trend line beats a dashboard of ten metrics.

Resist reporting rank as if it were a search ranking. AI answers have no stable positions, and promising a position invites disappointment.

Choosing a tool, and how AppearsIn tracks it

If you track more than a few dozen questions, a tool saves time. Look for: the engines you care about, fixed question sets with history, mention, position, sentiment and share of voice, a view of cited sources, and clear information on how often each check runs and what it costs.

AppearsIn is our product. Each question you track is checked on ChatGPT, Google AI Overview, Gemini, Claude, Perplexity, Grok and Meta Muse Spark: ChatGPT, Google AI Overview, Gemini and Perplexity every week, and Claude, Meta Muse Spark and Grok about once a month. We record whether the answer names your brand, what it says, and which pages it cites. The score is the share of your questions where any engine named you, shown with rankings against your competitors that give visibility, share of voice, average position and sentiment. A Cited view lists the domains the engines rely on. It is a measurement tool: it does not control what any engine says. The Answers docs explain each number, and the best Reddit marketing tools compare the alternatives.