Is ChatGPT sending you visitors?

How is a visit from an AI assistant recognised?

When somebody follows a link, their browser usually tells your site which site the link was on. That is the referrer. An answer in ChatGPT that links to your page sends the visitor along with chatgpt.com as the referrer, and the visit is counted under AI assistants rather than search or social. Only the host is kept, never the page it came from.

These are the addresses counted as AI assistants today:

An address nobody has listed yet counts as elsewhere. It moves to the right row the day it is added, for every month already counted, because where a visit came from is worked out when the figures are read rather than when the visit is stored.

Why is the number lower than the truth?

Because a referrer is a courtesy, not a guarantee. A visit that started in an assistant ends up in one of three places, and only the first one can be counted as one.

• AI assistants, when the link was followed in a browser and the referrer came with it.
• Nowhere in particular, when the link opened from an app that does not say where it came from, or was copied and pasted. The visit counts, with no referrer at all.
• Search, when somebody read your name in an answer and looked you up. The visit is real, and it counts where it landed.

So read the AI assistants row as the least it could be, and watch which way it moves from month to month. The trend is reliable even where the total is not.

Being read is not being visited

Before an assistant can send anybody, something has to read your page. OpenAI, Anthropic and Perplexity each run separate crawlers for separate jobs, and robots.txt can name each one:

Which AI crawlers should you let in? goes through the other companies' crawlers too, and what refusing each one does.

Most crawlers never run JavaScript, so a script in the page cannot see them. BackendFree counts them through the fallback image in the snippet or a report from your own server, which makes its crawler figures a floor too. It names 23 AI crawlers one by one, and none of them spends the events allowance.

What should you do about it?

1. Write pages that answer one question each. Put the question in the heading and the answer in the first two sentences, with the real numbers. That is the part an assistant quotes.
2. Let the answer crawlers in. If OAI-SearchBot, Claude-SearchBot or PerplexityBot cannot read a page, those assistants cannot cite it.
3. Decide about training on its own. Refusing GPTBot or ClaudeBot is a choice about future models, and on its own it does not take you out of answers.
4. Watch the trend, month by month. Compare the AI assistants row with itself rather than with search.

One way to write that robots.txt, letting answers in and keeping training out:

# Answer engines: let them read, so they can cite you
User-agent: OAI-SearchBot
User-agent: Claude-SearchBot
User-agent: PerplexityBot
Allow: /

# Training: your choice, made separately
User-agent: GPTBot
User-agent: ClaudeBot
Disallow: /

Sitemap: https://salon.example/sitemap.xml

Where to see it in BackendFree

On the traffic screen, AI assistants is its own row among the places visitors came from, beside search, social, maps, directories and email. The crawler panel underneath names each crawler and how often it came. Web analytics has the rest of what the measurement does, and what it refuses to collect.