AI Visibility Audit
See what ChatGPT, Perplexity, Gemini, Google AI Mode and Microsoft Copilot say about your brand when people ask about your field. The automation collects 150 answers straight from each assistant's own interface and shows how many of them mention you, who AI names instead, and which sources it draws on. No made-up score: every number links back to answers you can read.
How does the automation work?
You give Claude Code your site and brand. The automation picks five topics from your field, writes questions the way people actually ask them, and puts them to five AI assistants in their own interfaces. It then reads every answer, counts who appears and what gets cited, and builds the report. You own the strategy; the automation takes over the manual collecting and counting.
- Start
You have a site and a brand
In Claude Code, no coding.
-
Questions the way people ask them
Each of the five topics gets two kinds of question: generic ones that don't name your brand (does AI bring you up on its own?) and direct ones about your brand (what does AI say about you?). Every question is asked three times in different natural wording, because people phrase things differently.
-
Real answers, not the API
Answers come from the interfaces of ChatGPT, Perplexity, Gemini, Google AI Mode and Microsoft Copilot, collected through Bright Data. We don't use the vendors' APIs: they answer differently from what people see and often cite no sources. A standard run is 150 answers. Google AI Overviews is available as an optional beta if you have a SERP zone in Bright Data.
-
Honest counting
For each assistant separately it counts how many measured answers mention you, and never blends the numbers into one. Small samples are labelled "not enough data". An assistant that couldn't be measured is marked as not measured, with the reason, never as zero.
-
Material for your decisions
It shows who AI names instead of you, which sources it cites (your site, independent sites, competitors), which independent sites are worth getting into, and what the assistants claim about you. It ends with a prioritised list of next steps.
- Output
You know what AI says about you
A verdict per assistant, with the evidence underneath.
How to get started
-
Buy the automation
It's a one-off payment with no subscription. You'll get a licence key and a short setup guide by email straight away.
-
Add it to Claude Code
Paste the two lines from the email into Claude Code. The installer asks for your Bright Data API token and stores it on your machine. No coding.
-
Run the audit
Type something like "run an AI visibility audit for example.com". The report is ready in 20 to 40 minutes, most of which is waiting for the assistants to answer.
FAQ
What do I need to run it?
Claude Code, Python 3.11 or newer (the scripts use only the standard library, so there's nothing else to install) and a Bright Data account with an API token. The installer asks for the token and stores it locally. Bright Data charges pay-as-you-go at $1.50 per 1,000 answers, so a standard run of 150 answers comes to about $0.25. New Bright Data accounts get 5,000 free records a month. Claude usage is roughly that of a longer Claude Code session.
Why don't you just use each assistant's API?
Because API answers differ from what people see in the interface, and they often cite no sources. We want to see what an assistant tells someone asking about your field, links included. That's why answers are collected from the interfaces themselves through Bright Data.
Do I get an overall visibility score?
No, by design. For each assistant you get how many of the measured answers mention you, with its own denominator. We don't add up or average numbers across assistants, because each one answers differently and a single score would hide that. Under every number are the answers it came from.
What if an assistant can't be measured?
Sometimes an assistant is temporarily unavailable through the data provider. The report then marks it as not measured, with the reason, and never counts it as zero. You can re-run just that assistant later.
Can I repeat the audit and track changes?
Yes. The next run on the same brand in the same folder reuses the same questions, and the report adds a trend section comparing the counts by measurement date. It doesn't schedule itself, though; you start each run.
Does it work for other languages and markets?
Yes, the questions are written in the language and for the market you choose. We built and tested it on Czech first.
Can I measure only some of the assistants?
Yes. Before the run you pick which assistants to measure: all five, all but Copilot (a faster run), or any combination. The report then shows only the ones you picked. A repeat run reuses the same set so the numbers stay comparable.
What about Google AI Overviews?
It's available as an optional beta and needs a SERP zone in Bright Data. Without one, the audit runs on the five assistants.
What doesn't it do?
It doesn't check whether AI bots can read your site at all; that's our separate free automation, the AI Bot Access Check. It doesn't write content, and it doesn't track visibility on a schedule by itself; you start each run.