How to track brand mentions in ChatGPT and Perplexity

To track brand mentions in ChatGPT and Perplexity, build a fixed set of 10–20 real buyer prompts, run them consistently across both engines, and log three separate outcomes for each response. Note when your brand was mentioned, whether it was cited with a linked source, and if it was actively recommended. Do that on a weekly or monthly cadence against 2–3 named competitors, and you have the start of a real LLM brand-visibility practice, not a vanity screenshot.

Tools matter once you’ve proven your tracking method works and you’re ready to scale past a spreadsheet. Start with the method.

Why “mention” isn’t the metric you think it is

Most teams start the brand mention exercise by typing a question into ChatGPT, seeing their brand name appear, and calling it a win. That’s a start, but it collapses three different outcomes into one:

• Mention — your brand name appears in the response text, with or without a source.
• Citation — a specific URL of yours is named as a source the model drew from.
• Recommendation — the model doesn’t just name you; it actively suggests you as the answer to the buyer’s question.

These matter differently. A citation is one that you can trace back to a page and improve. A recommendation is the one that actually moves a buyer. Tracking only “mention rate” flattens a meaningful signal into a shallow one.

It also matters which engine you’re checking. ChatGPT and Perplexity don’t retrieve information the same way. Perplexity runs live, source-grounded retrieval and shows citations on essentially every answer. It’s the more useful diagnostic engine precisely because it shows its work. 

ChatGPT’s search-enabled mode draws on a different retrieval path and links sources less consistently, which means a clean mention in ChatGPT often carries no visible citation at all. If you only check one engine, or check ChatGPT without search enabled, you’re not getting a comparable picture. Track both as genuinely different systems, not two versions of the same test.

Step 1: Build a buyer-prompt set, not a keyword list

The single most common mistake here is testing your brand name instead of testing buyer language. “Is [Your Brand] good?” tells you almost nothing. What you want are the prompts a real prospect would type before they’ve decided who to buy from:

• Category prompts: “What’s the best [category] tool for a mid-size sales team?”
• Comparison prompts: “[Competitor] vs [Competitor] vs [Your Brand]”
• Alternative prompts: “Alternatives to [top competitor]”
• Use-case prompts: “How do I solve [problem your product addresses]?”

Ten to twenty prompts are enough for a real baseline. Keep the exact wording fixed once you’ve settled on it. The value comes from comparing the same prompt over time, not from novelty.

Step 2: Run the prompt set consistently across both engines

Open ChatGPT and Perplexity with search enabled and run every prompt in both. Log the raw response, not just a yes/no. Take a screenshot or save the full text. You’ll want to see exactly how the model phrases things when you circle back to fix something.

Run each prompt more than once if you can. AI-generated answers vary between sessions, and a single run can make a real pattern look like noise or noise look like a pattern. A handful of repeats per prompt gets you a directional read; if you’re reporting this to a board, you’ll eventually want more volume than is reasonable when creating a manual spreadsheet. This is the point where automated tracking earns its cost.

Step 3: Log four outcomes, not one

For every prompt × engine combination, record:

Prompt	Engine	Mentioned?	Cited (URL)?	Recommended?	Competitors named
“Best [category] for remote teams”	ChatGPT	Yes	No	No	2
“Best [category] for remote teams”	Perplexity	Yes	Yes — /blog/remote-team-guide	Yes	1

This is the format that turns a screenshot habit into a dataset. It also immediately tells you if you’re mentioned but never cited or if the model has heard of you but has nothing specific to point to. This is a diagnostic revelation that’s usually a content-depth problem, not an awareness problem.

Step 4: Benchmark against named competitors, not just yourself

A visibility number in isolation doesn’t tell you much. What tells you something is the gap. For example, if you appear in 3 of 10 prompt runs and your closest competitor appears in 8, that gap is the thing worth acting on. Track the same 2–3 competitors across the same prompt set every time, and watch whether the gap is closing or widening as you publish new content.

Step 5: Decide when manual tracking stops working

Manual spot checks are legitimate for an initial audit. An hour with a spreadsheet gives you a real baseline. However, they break down once you’re tracking more than a couple dozen prompts across multiple engines on a recurring cadence; consistency and volume both suffer. That’s the point when you move to automated tracking, whether that’s a dedicated AI visibility platform or an AI Search Toolkit layered onto the SEO stack you already run — we use SE Ranking’s AI Results Tracker for exactly this reason.

Automating the check doesn’t automate the judgment call, though. A dashboard tells you what changed. Someone still has to decide what it means and what to fix. This is the step most tools stop short of and most teams underestimate.

Turning the log into content action

The log is only useful if it changes what you publish. A prompt where you’re mentioned but never cited should prompt you to strengthen the specific page that should be earning that citation. A prompt where a competitor is consistently recommended and you’re not tells you exactly which buyer question your content roadmap hasn’t answered yet. Treat the tracking log as a content backlog, reviewed on the same cadence you run the prompts, and the exercise stops being a report and starts being a program.

The takeaway

Tracking brand mentions in ChatGPT and Perplexity isn’t about catching a lucky screenshot. It’s a repeatable measurement discipline: fixed prompts, three separate outcomes, a real competitive benchmark, and a cadence you actually keep. Get that right by hand first. The tooling decision gets a lot easier once you know exactly what you’re asking it to scale.

Not sure where you currently stand? Check your AI brand reputation with our free GEO Grader — it’s a fast way to see how you’re showing up before you build the full tracking log.