AI bots Pro
See which AI bots read your WordPress site, page by page
Is ChatGPT reading my site? Is Claude? PressBot answers with Cloudflare’s own counts: which AI companies fetch your pages, how often, and for what. You see it in WordPress, every week, without reading server logs.
The reader report is part of PressBot Pro. Telling bots what they may do is free.
What you see
PressBot groups AI readers by what they are doing. The groups matter because they mean different things for your business.
-
Training
The bot collects pages to train a model. Your page may shape future answers, but this kind of read does not send visitors to your site.
-
Search
The bot builds an index that an AI search product uses to find sources. This is the group that can send visitors.
-
Agents
The bot fetched a page because a person asked an AI assistant a question. Someone wanted this page right now.
A real week: pressbot.io, one week in October 2026
Our own site, seven days ending 27 October 2026, as Cloudflare counted them and PressBot showed them.
- 42,813requests to pressbot.io in the week
- 1,121requests from AI crawlers and assistants
- 2.6%of all requests came from AI crawlers and assistants
| Operator | Purpose | Requests |
|---|---|---|
| ByteDance | Training (Bytespider) | 324 |
| OpenAI | 180 | |
| OpenAI, training | Training | 14 |
| OpenAI, search | Search | 106 |
| OpenAI, agent | Agent | 60 |
| Anthropic | 162 | |
| Anthropic, training | Training | 161 |
| Anthropic, agent | Agent | 1 |
| Amazon | Training | 146 |
| Apple | Search | 138 |
| Perplexity | 95 | |
| Perplexity, search | Search | 94 |
| Perplexity, agent | Agent | 1 |
| Meta | Training | 70 |
The other 6 requests came from other operators. Source: Cloudflare, read by PressBot, for pressbot.io only. This is one site in one week, so do not read it as a trend or as typical for your site.
What this means for us
Most of the reads were for training: ByteDance’s Bytespider made 324 requests, Anthropic 161, Amazon 146 and Meta 70. Two operators were mostly search: Apple made 138 requests and Perplexity 94. OpenAI made 180, and 106 of them were search. Only 62 requests came from agents, which fetch a page because a person asked an assistant a question. That split between training, search and agents is what you need to decide what to allow.
Nearly a quarter of the “AI” traffic was not AI
The numbers above leave something out, on purpose.
In the same week, 337 requests carried the name of an AI bot, for example a Claude user agent, but were scanners. They asked for files such as /.env and old copies of wp-config.php, which a real AI assistant has no reason to fetch. PressBot 1.26.0 leaves them out of the AI count.
Before that filter, the same week looked like 1,458 AI requests and 275 visits from Claude-User, the Anthropic agent. Almost all of those 275 visits were scanners pretending to be Claude. With them removed, Anthropic’s agent made 1 request.
A user agent is only a text label, and anyone can send any label. So PressBot drops requests for scanner addresses before it counts them. It still cannot prove that the rest are who they say they are.
Why it matters
People now ask AI assistants for a business, a product or an answer, and some of them are sent to a website. If an AI reads a page often and answers badly from it, that page is the one to fix first.
Fixing the page is a different job from counting the readers. For the audit of whether your content is ready for answer engines, see AI Visibility. This page is about traffic: who fetched what.
Agent Door: tell bots what they may do
Counting is only half of it. The Agent Door sets what bots are told when they arrive.
-
Content-Signal Free
Turn it on and PressBot adds a Content-Signal line to your
robots.txt. The defaults are: search yes, AI answers yes, AI training no. It is a request, not a lock. Well-behaved crawlers follow it. Others can ignore it. -
Markdown for AI agents Pro
An agent that asks for Markdown gets your page as clean Markdown instead of HTML. It is easier for the agent to read and carries less markup around your words.
Your robots.txt on pressbot.io, with the Content-Signal line:
User-agent: *
Content-Signal: ai-train=no, search=yes, ai-input=yes
Disallow: /wp-admin/
Allow: /wp-admin/admin-ajax.phpAsk for a page as Markdown, and the start of the reply:
$ curl -i -H "Accept: text/markdown" https://pressbot.io/changelog/
content-type: text/markdown; charset=utf-8
vary: Accept
content-signal: ai-train=no, search=yes, ai-input=yes
# Changelog
> PressBot is a free AI chatbot for WordPress. This changelog documents updates and improvements to the platform.What is honest about the numbers
-
Names come from user agents
Cloudflare and PressBot name a bot by the user agent it sends. Anyone can fake a user agent, so a name is a strong hint, not proof. PressBot drops requests for scanner addresses, but it cannot check the rest.
-
Cloudflare counts cached pages
A request served from Cloudflare’s cache still counts. The counts include reads that never reached WordPress.
-
Without Cloudflare, you see less
Without Cloudflare in front of your site, PressBot sees only the requests that reach PHP. Cached pages and blocked requests do not show up.
-
A read is not a citation
The counts show that a page was fetched. They do not show that an AI used it in an answer or sent a visitor to you.
How it is set up
- Your site is on Cloudflare. The counts come from there. Cloudflare’s free plan covers them, and PressBot does not change your DNS.
- You create one scoped Cloudflare token and paste it into PressBot. The Cloudflare page shows the setup in order.
- Blocking happens in Cloudflare. If you decide a bot should not read your site, use Cloudflare’s AI Crawl Control. PressBot links to it and does not copy it.
Questions
Does PressBot block AI bots?
No. PressBot shows you which AI bots read your site and lets you state what they may do. Blocking is done in Cloudflare's own AI Crawl Control, and PressBot links to it.
Is it free?
Part of it. The Content-Signal line in robots.txt is free. The AI readers report on the Analytics page and Markdown for AI agents are in PressBot Pro.
Do I need Cloudflare?
For the counts, yes. Your site has to be served through Cloudflare, and PressBot reads the numbers with one scoped Cloudflare token. Cloudflare's free plan covers these features, and PressBot does not change your DNS.
How does PressBot know a request came from an AI bot?
From the user agent that each request sends, as Cloudflare reports it. A user agent can be faked, and scanners often use AI names. PressBot leaves out requests for scanner addresses, but treat the names as a strong hint, not proof.
If an AI bot reads my page, will AI quote my site?
Not necessarily. A read means the page was fetched. It does not mean the page was used in an answer or that anyone was sent to your site.
Get PressBot
The free plugin sets your Content-Signal. Pro adds the AI readers report and Markdown for agents.
Keep reading
Cloudflare is a trademark of Cloudflare, Inc. PressBot is not affiliated with Cloudflare.