Blog · November 9, 2026 · Punchline Analytics
What your analytics calls bot traffic
"Bot" lumps together crawlers, scripts, impostors and AI agents reading your site for a real buyer. Filtering them all out made sense once. It doesn't now.
Most analytics tools have one word for every visitor that isn't a person in a browser: bot. And one thing to do with them: filter them out.
That made sense when the non-human traffic on a site was search crawlers, uptime monitors and spam. It makes less sense now that a growing share of it is software reading your site for a person: an assistant answering "which of these three tools fits a 200-person company?", or an agent comparing your pricing page with a competitor's.
Who's actually in there
What gets counted as bot traffic is at least five very different kinds of visitor:
- Crawlers collect pages to train models or build search indexes. GPTBot, ClaudeBot and Googlebot are examples. They read a lot and buy nothing, and their operators publish who they are.
- Agents reading for someone right now. When a person asks ChatGPT or Claude about you, the assistant may fetch your pages on the spot (ChatGPT-User, Claude-User, Perplexity-User). This is the research step of a purchase, done by software.
- Browser agents drive a real browser on a person's behalf: they click, scroll and fill in forms. Some sign their requests so a site can verify them.
- Fetchers and scripts: curl, Python, monitoring tools, scrapers.
- Impostors claim a well-known crawler's name from an address that doesn't belong to its operator. OpenAI, for example, publishes the IP ranges its crawlers use, so a "GPTBot" from anywhere else is lying.
Only one of those is noise in the old sense. Two are your buyers' research, one is a training pipeline you may want to allow or block on purpose, and one is someone pretending.
What analytics does with them
Google Analytics 4 excludes traffic from known bots and spiders automatically. According to Google, you can't turn this off or see how much was excluded.
Privacy-first tools filter too, often more aggressively. Plausible filters known AI crawlers and other bots by user agent, referrer spam lists, data-center addresses and traffic patterns. Fathom blocks known bot user agents, including AI scrapers, and data-center traffic, and shows a count of what it blocked.
There's a second blind spot under the first. Many of these visitors never run JavaScript, so a tag-based tool never sees them at all. Matomo's own documentation makes the point when explaining why its AI chatbot report needs a server-side integration.
Filtering keeps your people numbers clean, and that's worth keeping. But it throws away the setup of the story: the research that happened before the visit you can see.
What to look at instead
Treat agents as an audience, not a filter. For each one, four questions matter:
- Who is it, and how sure are we? A verified signature or a published address is strong evidence; a cloud IP and a steady request rate is a hint. Labels should show which.
- Why did it come? Training, building a search index, and answering a person's question right now are different businesses with different value to you.
- Did it get the page? An agent that gets a redirect, an error or an empty JavaScript shell can't tell anyone about you.
- What happened next? Did an assistant then send people? Did the company whose agent read your security page book a demo?
That's the analytics Punchline is built for: every visit labelled with its evidence, people's numbers kept clean, and agents kept, explained and connected to what they led to.
Start free to see your whole funnel on your own site.