Is AI reading my site?
More and more people ask an assistant before they ask a search engine, and the assistant goes and reads websites to answer them. So "is AI reading my site?" has become a fair business question, and until now there was no honest way to answer it.
Open Reports and press AI traffic. You get one sentence with the number in it, then the figures behind it.
The three numbers, and why they are never added up
- Read to answer someone's question. A person asked an assistant something, and it came and fetched your page to answer them. This is the headline, because there was a real human on the other end of it.
- Read to keep AI search up to date. An assistant refreshing what it knows about the web. This is the step that comes before being quoted.
- Read to learn from. Bulk reading, for training. It says nothing at all about whether you get mentioned in an answer.
They mean three different things, so one combined "AI visits" figure would be worse than useless: a thousand training reads and one real question would look identical to a thousand real questions. The report keeps them apart, always.
Underneath, you get a line per AI company in the names you would recognise - ChatGPT, Claude, Perplexity - rather than the internal names their systems use. If you run more than one site, each site gets its own row.
How the counting works
AI companies' systems introduce themselves by name when they visit a website. bernard notices those visits as your pages are served. There is no tracking of any kind, and your human visitors are never involved.
Anyone can claim a name, so bernard double-checks every visit against the caller lists the AI companies themselves publish, and only counts the ones that check out. Anything that claimed a name and could not be confirmed gets one quiet line at the bottom, and is never counted in your figures.
What it deliberately does not show
- Google's AI answers. They are gathered by the same system that runs Google's ordinary search, and the two cannot be told apart, so they are not in these figures at all. What people searched, and how often Google showed you, lives in the Traffic report.
- A site behind your own Cloudflare account may read a little low, because some visits are answered there before they ever reach us. Treat those figures as "at least".
- Being read is not being recommended. Reading comes first, and nothing here promises a mention.
If nothing has read your site yet
That is an ordinary state, not a fault, and the report says so calmly. Counting began on 30 July 2026, so a window reaching back further than that is shorter in truth than it looks, and the card tells you.
What tends to change it is having pages that answer a real question in plain words - what you make, who it is for, where you are - and having them published long enough to be found. The same work that helps a person understand your site is the work that gets it read.
The prompt
“Run the AI traffic report for the last month. Tell me who has been reading my site, whether any of it was to answer somebody's question, and what it does not cover.”
Press Copy, then paste it into your AI and fill in the [bracketed] parts. First time? Connect your AI to bernard first - a one-time setup in bernard → your site → Use your own AI.
Questions people ask
- Is ChatGPT reading my website?
- Open Reports and press AI traffic. You get one line per AI company - ChatGPT, Claude, Perplexity and so on - and how many times each one read your pages over the window you picked. If the answer is nobody yet, it says so plainly rather than showing you an empty chart. It only reads your figures, so there is nothing to approve and nothing can change.
- How do you know it was really ChatGPT and not somebody pretending?
- Anyone can put a name on a visit, so bernard does not take the name on trust. Every visit is checked against the caller lists the AI companies publish themselves, and only the ones that check out are counted. Anything that claimed a name and could not be confirmed is mentioned separately, at the bottom, and is never added to your figures.
- What is the difference between the three numbers?
- They answer different questions, which is why they are never added together. "Read to answer someone's question" means a person asked an assistant something and it came and fetched your page to answer them - that is the closest thing to a customer. "Read to keep AI search up to date" is an assistant refreshing what it knows, which is the step before being quoted. "Read to learn from" is bulk reading for training, and says nothing about whether you get mentioned.
- Does this track my visitors?
- No. Nothing is tracked, and your human visitors are never involved in these figures at all. AI companies' systems introduce themselves by name when they visit, and bernard simply notices those visits as your pages are served. There is no tag, no cookie and no third party.
- Why does Google not appear for its AI answers?
- Google's AI answers are gathered by the same system that runs its ordinary search, and the two cannot be told apart, so they are honestly left out rather than guessed at. What people searched on Google, and how often it showed you, is in the Traffic report instead.
- Does being read mean I am being recommended?
- No, and the report is careful not to imply it. Reading comes first: an assistant has to fetch your pages before it can ever mention you. So a healthy set of figures means your pages are reachable and worth fetching. It is not a promise of a mention in anybody's answer.