How to track brand mentions in ChatGPT

Why one check tells you almost nothing

SparkToro and Gumshoe.ai asked 600 volunteers to run 12 prompts through ChatGPT, Claude and Google’s AI features. That came to 2,961 runs.

The lists almost never repeated. The same list came back less than once in a hundred runs, and the same order about once in a thousand.

Brands still showed a pattern. In narrow categories the top brands turned up in roughly 55 to 77% of runs, and in broad ones nearer 30 to 40%.

So don’t judge a brand on one run. Count how many answers name it, out of how many you checked.

What counts as a mention?

A mention is your brand name showing up in the answer. It’s the easiest thing to count, though it isn’t the only thing worth counting.

• NamedYour brand appears anywhere in the answer, even in passing.
• RecommendedThe answer presents your brand as a good fit for the person asking.
• CitedYour website shows up as a source link under the answer.

You can be cited without being named, and named without being cited. Log them separately, since each one points to a different fix.

How to check if ChatGPT mentions your brand for free

You need a spreadsheet and some patience. It’s slow, but it’s a fair first look and it costs nothing.

1. Write eight to twelve questions in your buyers’ words. Mine sales calls and support tickets for them, not your homepage.
2. Open a fresh chat for every run, so an earlier answer can’t steer the next one.
3. Ask each question at least three times, on different days if you can.
4. Log the brands in the order they appear, whether you’re one of them, and any label ChatGPT gave you.
5. Copy the cited links into the sheet as well.
6. Count a question as a win only when you show up in most of its runs.

Keep one row per run, not one per question. Thirty runs means thirty rows, and that’s normal.

Which questions should you ask?

Ask what buyers type, not what your marketing answers. Sales calls, support tickets and search queries beat your own homepage as a source.

• Shortlist“Best invoicing tool for a five-person agency?”
• Comparison“Tool A vs Tool B for remote teams”
• Switching“Alternatives to Tool A that cost less”
• Problem“How do we stop chasing late payments?”

Add a few problem questions that never name your category. Buyers often describe the pain first, and the answer decides which category they hear about.

How do you read what you found?

Look at the pattern across runs. Named in four of five is a different position from named in one of five.

• Named in most runsProtect it. Read what the answer says about you and check it’s accurate.
• Named sometimesCompare the runs where you appear with the ones where you don’t. The wording or the cited pages often differ.
• Never namedOpen the cited pages and see who’s on them. You’re probably missing from the sources the answer draws on.

None of this proves why ChatGPT wrote what it did. It shows you where to look next.

When is a paid tool worth it?

Tracking by hand works for a first look. It drags once you pass a handful of questions, and it only shows change over time if you repeat it every month.

Crawlbird does the tedious part. You add your site, edit the suggested questions, and it asks each one on the AI platforms you choose (ChatGPT, Gemini and Google AI Mode) and saves the answer. Recheck later to see what changed.

For every answer, the report shows whether you were mentioned, whether you were recommended, and which pages were cited. Each number opens the saved answer.

Every plan checks ChatGPT, and paid plans add Gemini and Google AI Mode. New accounts get a one-time free allowance of 20 ChatGPT answers, enough to try a few questions.

Common questions