You can now see every AI crawler that reads your website, which company operates it, and whether it came to train a model, index you for search, or answer a reader's question. Analytics tools that run as a page script can't show you this, because crawlers never run page scripts. We record it at the edge instead, so the visit is counted as the page is served.

• Two kinds of traffic, kept apart: People who found you and bots that read you are different things, so they're counted separately rather than blended into one "referral" number.

• Who's reading: 21 AI crawlers from 12 companies, including OpenAI, Anthropic, Google, Perplexity, Meta, Apple, and Amazon. Each hit is named, not bucketed as "bot traffic".

• Why they came: Every crawler is labeled by intent. Training collects data for a model, search indexes you for later, and assistant means a reader asked something and the bot fetched your page to answer it.

• Which posts they take: A most-crawled list shows which pieces the bots keep coming back to, which is rarely the ones you'd guess.

• Where readers come from: Visits from 11 AI assistants, including ChatGPT, Perplexity, Claude, and Gemini, are separated from 11 search engines, including Google, Bing, DuckDuckGo, and Kagi.

• Down to one post: Narrow any of it from your whole publication to a single piece to see how that one is being found.

Crawler tracking needs a live website on Paragraph, since the crawl has to reach a page we serve. The search and AI visitor split works on any publication, with no website required. Neither needs a paid plan.

Some AI assistants strip the referrer, so the real number of readers they send you is higher than what's shown. And crawler visits usually take a few days to appear after you publish, because the bots have to come back.