> Site index: https://usehardal.com/llms.txt
> Every content route also serves markdown at <url>.md

# How to Measure ClaudeBot

ClaudeBot is Anthropic's training crawler. Measure crawl volume and pages in logs or Hardal AI Visibility, separate from Claude-User fetches.

Source: https://usehardal.com/blog/how-to-measure-claudebot-traffic
Published: 2026-08-17
Updated: 2026-08-17
Author: Ecem Bircan
Category: Marketing

---

**ClaudeBot** is Anthropic's training crawler. It reads your site so future Claude models can learn from public pages. There is no Claude ads network, and a live user fetch arrives under a different agent.

Anthropic has split the job the same way OpenAI did: training vs search vs user-initiated fetch. In logs you will also see **Claude-User** (someone asked Claude to open a URL) and, on newer stacks, **Claude-SearchBot** for in-product search indexing. Measuring "Claude traffic" as one number hides those three jobs.

Hardal's [AI Visibility](https://usehardal.com/ai-visibility) docs call out ClaudeBot by name and attribute its visits to Anthropic, on the trainer line. Treat fetch spikes on pricing the way we treat [ChatGPT-User](https://usehardal.com/blog/how-to-measure-chatgpt-user-traffic): live intent, not a training crawl.

## How ClaudeBot shows up

Look for `ClaudeBot` in the user-agent. Do not collapse `anthropic-ai`, `Claude-User`, and `ClaudeBot` into one regex unless you also split the chart. The playbook examples use `ClaudeBot` next to GPTBot and PerplexityBot because those are the high-volume names marketers actually see ([AI Visibility Playbook](https://usehardal.com/playbook/ai-visibility)).

Client-side analytics still miss it. ClaudeBot runs no JavaScript, so there is no GA4 session.

## What to report

- ClaudeBot hits vs Claude-User hits. If trainers dwarf fetchers, Anthropic has your pages but almost nobody is asking Claude to open them.
- Pages Anthropic reads most. Our playbook cites the ugly ratio problem: heavy crawl, almost no referred humans. That is a content and robots.txt decision, not a paid-media one.
- Whether `/docs` and comparison pages are in the crawl set. If Claude is going to describe Hardal-class products, those URLs should be current.

There is no Hardal "Claude Ads" destination. Do not copy the [ChatGPT Ads](https://usehardal.com/blog/how-to-measure-chatgpt-ads) setup onto ClaudeBot.

## robots.txt

Block training, keep fetches and search if you want to be cited:

```
User-agent: ClaudeBot
Disallow: /
```

If you also want to stay out of Claude search indexes, add the search token Anthropic documents at the time you ship the file. Tokens change. Re-read Anthropic's crawler page each quarter instead of copying a 2024 gist.

Blocking ClaudeBot does not change [Googlebot](https://usehardal.com/blog/how-to-measure-googlebot-traffic) rankings.

## How Hardal measures it

Same pipeline as GPTBot: first-party server-side identification, with Anthropic as the platform and trainer as the category. [isvisible.ai](https://isvisible.ai) checks whether Claude can access the homepage. Continuous volume and page lists are [AI Visibility](https://usehardal.com/ai-visibility) (beta / waitlist).

For the after-the-report actions, use [AI crawler traffic for marketers](https://usehardal.com/blog/ai-crawler-traffic-for-marketers).