GitHub ↗

How to check whether ChatGPT, Perplexity and Claude mention your brand


Short answer: pick a fixed set of prompts a real buyer would ask, run them on each assistant on a schedule, and log four things every time: were you named, in what position, which URL was cited, and who got recommended instead. The prompt set has to stay frozen — otherwise you're measuring your own wording, not your visibility. Most of it can be automated in your own logged-in browser.

Why "just search for your name" doesn't work

Asking an assistant "what is [your brand]" almost always produces a flattering answer — it's reading your own site. That measures nothing. What matters is whether you appear when someone describes the problem you solve without naming you. That's the query that decides whether an AI sends you a customer.

The method

1. Build a frozen prompt panel (~16 prompts), in three layers:

2. Run every prompt on each assistant — ChatGPT, Perplexity, Claude, Copilot, Google's AI answers. Use a clean session so prior chat history doesn't contaminate the result.

3. Log the same four fields every time:

FieldWhy
Mentioned?The headline metric
PositionBeing named third is not being named first
Cited URLThe one people skip — and the most important
Who was recommended insteadTells you who owns the answer today

4. Repeat monthly. A single run is a snapshot; the trend is the signal.

The trap: verify which URL was cited

If your brand name is shared with another project, a mention is ambiguous — the assistant may be citing your competitor. Read the actual cited hostname, not the chip label.

We hit this ourselves: on a prompt describing our core use case, ChatGPT named "Browser MCP" second — and only by checking the cited hosts could we confirm it pointed at our domain and not the similarly-named project. Same run, Perplexity didn't mention us at all, and stated that a competitor was actively maintained when its repository had been silent for over a year. Both facts were only visible because we logged the citation, not just the mention.

Automating the run

Doing this by hand across five assistants monthly gets old fast. The awkward part is that the assistants worth measuring are the ones you're logged into — so a headless scraper is the wrong tool: it isn't signed in, and a fresh browser profile gets a different (or blocked) experience.

Running it in your own browser sidesteps that. With Browser MCP, an agent drives the Chrome you're already signed into: submit the prompt, wait for the answer to finish rendering, read the response and the cited links, and write the row. Two practical notes from doing it: read the cited link hostnames programmatically rather than trusting the visible label, and take a screenshot as evidence — some assistants render answers in ways that plain text extraction misses.

What to do with the result

FAQ

How often should I run it? Monthly. More often mostly measures model noise.

Do I need paid accounts? Not necessarily — several assistants answer without login, though logged-in sessions better reflect what your buyers see.

Why not use an API? The API and the consumer product often use different retrieval and different system prompts. Measure what your customers actually use.

Is this the same as SEO rank tracking? Related but not the same. Classic tracking measures a ranked list; this measures whether you're named inside a synthesised answer — and which source it credits.