ChatGPT, Claude, Perplexity and other assistants read websites with their own crawlers. In Matomo you usually see none of this. That is not a wrong setting, it is how the measurement works.
Why Matomo does not see the crawlers
Matomo’s tracking runs as JavaScript in the browser. Crawlers mostly load only the HTML and do not run scripts. Matomo also filters out known bots. The crawlers’ requests therefore only appear in the access logs of the web server or of a service in front of it, such as Cloudflare.
Three kinds of AI access
- Training: crawlers collect content for training language models, for example GPTBot from OpenAI, ClaudeBot from Anthropic or CCBot from Common Crawl.
- Search: crawlers build the index for search inside the assistant, for example OAI-SearchBot from OpenAI or PerplexityBot.
- Fetching on request: when someone asks the assistant about a specific page, a fetcher retrieves it live, for example ChatGPT-User, Claude-User or Perplexity-User.
The distinction helps with deciding what to allow. Many exclude training but still want to appear in the assistants’ answers.
Google-Extended and Applebot-Extended are not crawlers of their own but tokens for robots.txt. They control whether content may be used for AI models. Crawling continues with the normal Googlebot and Applebot, so they do not appear separately in the logs.
A crawler’s identifier can be faked. Several providers publish the IP ranges of their crawlers, including OpenAI, Perplexity and Google. That lets you check whether a request is genuine.
The AI reports in Matomo
Matomo has dedicated reports for this: under AI Assistants, the AI Chatbots Overview shows which chatbots fetch your content. Matomo counts these fetches as requests, not as visits, so your regular reports, goals and campaigns stay clean. On a self-hosted Matomo, the BotTracking plugin has to be active.
Among other things, the overview shows:
- requests per chatbot, over time
- which pages and documents are fetched
- how many visits come from links in the answers, and their ratio to the requests
- requests that hit missing pages (404) or server errors (5xx)
Which bots Matomo recognises there depends on the version. Currently these are mainly the assistants’ fetchers, such as ChatGPT-User, Claude-User and Perplexity-User.
How the fetches get into Matomo
Because chatbots do not run JavaScript, their fetches have to be reported server-side. Matomo offers integrations for Cloudflare, Amazon CloudFront and WordPress as well as the HTTP Tracking API. For your own servers, a pipeline reads the access logs continuously and reports the fetches via the Tracking API. I use my own pipelines based on Filebeat or Vector for this, which run for many clients.
Server logs contain IP addresses. The bot, the page and the time are enough for the analysis. Agree with your data protection team what is transferred and how long it is kept.
Visitors from AI assistants
The second question is who comes to your website via a link in an AI answer. Matomo measures these visits as usual. Current Matomo versions recognise referrals from ChatGPT, Claude, Copilot, Gemini, Perplexity and other assistants and show them in a report of their own. If you want to compare them with search, social and newsletter, including goals and revenue, group them into a channel of their own with the Channel Grouping plugin from feinwerk.io.
Two limitations apply: visits from apps often arrive without a referrer and appear as direct traffic. And not every mention in an answer leads to a click. The numbers are therefore a lower bound.
More under AI crawlers & AI search.
Questions about your setup?
In an intro call we'll look at your situation together.
Book an intro call