Track Brand Mentions in AI: ChatGPT, Gemini, Perplexity

To track brand mentions in AI, ask the same fixed set of buyer questions in each assistant (ChatGPT, Gemini, Perplexity, Claude and others) on a regular schedule. For every answer, record whether your brand appears, where it sits relative to competitors, and which sources were cited. Trends across repeated runs, not single answers, show where you really stand.

Checking whether ChatGPT mentions your brand once is easy. Knowing whether it mentions you consistently, how that compares with Gemini, Perplexity and Claude, and whether things are getting better or worse is a different job. That is a measurement problem, and it needs a repeatable method rather than a few screenshots shared in Slack.

This guide walks through an AI brand monitoring process that holds up when your CMO asks what changed and why.

Key takeaways

• Each assistant has its own retrieval behavior, citation style and answer format. A single “AI visibility” number hides real differences.
• Track the same prompts on every assistant, on a regular schedule, so results are comparable.
• Record three things per answer: whether your brand appears, where it appears relative to competitors, and which sources were cited.
• Look at trends over weeks, not individual answers. AI responses vary from run to run.
• Use the cited sources to decide where to invest, not just the mention count.

Why you need to track brand mentions in AI across several assistants

Buyers do not all use the same tool. Some live in ChatGPT, some default to Gemini because it is built into their Google account, some prefer Perplexity for research, and developers often reach for Claude. Google AI Overviews reach people who never open a chatbot at all.

These assistants also behave differently in practice. Some lean heavily on live web search and show numbered citations. Others tend to answer more from what the model already knows. A brand can be the default recommendation in one and absent in another for the same question. For background on why answers diverge, see how AI assistants choose sources.

What to record when you track brand mentions in AI

For every prompt and every assistant, capture the same fields:

1. Mentioned or not. Did the answer name your brand at all?
2. Position. Were you the first brand recommended, one of five in a list, or a footnote?
3. Competitors named. Which other brands appeared in the same answer?
4. Cited sources, in order. Which URLs did the assistant link to, and in what rank?
5. Framing. Were you described accurately, or is the pricing or positioning out of date?
6. Searches run. Where the assistant exposes it, which web queries did it run before answering?

The first three give you share of voice. The last three tell you why the numbers look the way they do.

A step-by-step AI brand monitoring process

1. Fix your prompt set

Pick 10 to 30 questions your buyers actually ask, such as “best CRM for a 10-person sales team” or “HubSpot vs Salesforce for startups”. Keep the wording stable. If you change a prompt, treat it as a new series. Our guide on choosing the prompts to track for AI visibility goes deeper on this.

2. Run every prompt on every assistant

Ask each prompt in ChatGPT, Gemini, Perplexity, Claude and any other assistant that matters to your audience, such as DeepSeek or Grok. Check Google AI Overviews for the search-style versions of the same questions.

Use the real consumer apps where possible. API responses can differ from what a user sees in the app, because the app may add web search, personalization or a different setup.

3. Keep conditions consistent

• Use the same account type each time (signed in, same plan).
• Start a fresh chat for each prompt so earlier context does not leak in.
• Note the date, the assistant and, if visible, the model or mode used.
• Avoid leading the model. Ask the question the way a buyer would, without your brand name unless the prompt is a comparison.

4. Repeat on a schedule

Weekly works for most teams. Daily can be useful during a launch or after publishing major content. Because answers vary between runs, a single sample tells you little. A series of samples over several weeks shows a real trend.

5. Store answers, not just scores

Keep the full answer text and the source list. When your share of voice drops, you will want to see exactly what the assistant said instead and which page it pulled from.

6. Review by assistant, then by prompt

Start with a per-assistant view: where are you strong and where are you missing? Then drill into the prompts where a competitor dominates and read the cited sources. That is usually where the action plan comes from.

How to track brand mentions in AI with Citegram

A spreadsheet is enough for a first audit. The problem is scale: twenty prompts across six assistants is 120 answers per run, and done every week, consistency tends to slip. Automating the collection keeps the method identical from run to run.

Here is how the workflow looks in Citegram, a Chrome extension:

1. Sign in to your assistants. Make sure you are logged in to ChatGPT, Gemini, Perplexity, Claude, DeepSeek and Grok in Chrome. The extension uses those existing sessions, so no API key is needed. Questions count toward your own plan usage on each assistant.
2. Add your prompts. Enter the buyer questions from your fixed prompt set.
3. Start a run. The extension asks each prompt in the real assistant apps in background tabs, and checks Google to see whether an AI Overview appears and which sources it cites. You can stop and resume a run, and nothing is lost if Chrome closes.
4. Read each answer. For every response you get the answer text, whether and where your brand is mentioned, the cited sources in rank order and, for ChatGPT, the web searches it ran.
5. Open the reports. Compare share of voice per brand and per assistant, and see the most cited domains across your prompts.
6. Re-run regularly. Repeating runs builds the trend lines you need to judge whether a change is real.

You can see the screens in the product tour.

Turning cross-assistant data into actions

Once you have a few weeks of data, patterns tend to show up quickly.

• You appear in one assistant but not others. Compare the sources each one cites. The assistant that mentions you is probably citing a page or review site where you are well represented. The others are drawing from places where you are not.
• You are mentioned but described wrongly. Look for the outdated source. It is often an old comparison article, a stale review profile or your own legacy pricing page.
• A competitor wins comparison prompts everywhere. Check whether they publish honest comparison pages, have more recent reviews or appear in more third-party “best of” lists. Our guide on how to get cited by AI through review sites covers how to close that gap.
• Results swing widely week to week. That often means the assistant has no strong source for the topic. A clear, well-structured page that answers the question directly can become that source.

To roll these observations into one comparable metric, read our explainer on AI share of voice.

Quick weekly review checklist

• Share of voice per assistant, compared with last week
• Prompts where you dropped out of the answer
• New competitors appearing in answers
• New or lost cited sources
• Any inaccurate statements about your product
• One action item per finding, with an owner

FAQ

How do I track brand mentions in AI assistants for free?

You can do it manually: ask a fixed list of buyer questions in each assistant, in fresh chats, and log mentions, positions and cited sources in a spreadsheet. It works for a first audit but gets hard to keep consistent. Citegram’s free plan also lets you automate one prompt across assistants.

How many assistants should I track?

Start with the ones your buyers use. For many brands that means ChatGPT, Gemini, Perplexity and Google AI Overviews at minimum, with Claude, DeepSeek and Grok added if your audience uses them. It is better to track four assistants consistently than eight sporadically.

Why do I get a different answer every time I ask?

Language models generate answers with some randomness, and assistants that browse the web may retrieve different pages on each run. That is why single checks are unreliable. Tracking the same prompts over time is the useful approach.

Can I compare my visibility across ChatGPT, Gemini and Perplexity directly?

Yes, if you use the same prompts and the same method on each one. Share of voice per assistant is comparable as long as the prompt set is identical. The gaps between assistants and the direction of change matter more than absolute numbers.

Check your own brand

If you want to see where your brand stands today, you can track your first prompt across assistants for free, with no API key required. The free plan covers 1 prompt, and Pro adds unlimited prompts if you need broader coverage. See the options on the pricing page.