Manual
Product site My Page

Visits from AI

Last updated · Sorabun Editorial Team

Count, from this site's own records, which articles AI crawlers fetched and which articles people landed on from links in ChatGPT, Perplexity and other AI answers

Is AI reading your site, and are people arriving from AI?

AI Recommendation (LLMO) asks the AI and measures whether you are named as a recommendation. This screen looks from the other side and counts what actually came to this site (since 2.133.0; Pro, administrators only).

  1. Which articles AI crawlers fetched
  2. Which articles people landed on from links in AI answers

Open it from "Sorabun > Leave it to AI > Visits from AI." You can choose the last 30 days or the last 90 days.

What it does not show

It does not show a number such as your "rank" inside AI. An AI's answer changes every time, so a single number would suggest something that does not exist. This screen holds only counts of what actually arrived.

Times AI crawlers came

The number of times an AI crawler fetched a single article page. Crawlers are told apart by their user agent.

GroupCrawlers
Search and answers (to list the page in AI search, or to read it on a user's request)OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, DuckAssistBot, MistralAI-User, MistralAI-Index, Meta-WebIndexer, Meta-ExternalFetcher, Amzn-SearchBot, Amzn-User
TrainingGPTBot, ClaudeBot, CCBot, Bytespider, Meta-ExternalAgent, Amazonbot, MistralAI-Training

Anyone can fake a user agent. So when the company publishes its IP address ranges, Sorabun checks whether the request came from those ranges and counts it in one of four groups.

ResultMeaning
VerifiedCame from the IP ranges the company publishes. Safe to treat as genuine
Name onlyA crawler from a company that does not publish IP ranges (Meta, Amazon, ByteDance and so on). Whether it is genuine cannot be told
UnverifiedThe IP ranges could not be fetched, or a CDN or proxy in front of the site hides the real IP (on sites behind Cloudflare, and sites that set "Trusted proxies" in "Security / Speed", the real IP passed on is used)
Out of rangeCame from an IP outside the published ranges. The user agent may be faked, so it is left out of the totals

OpenAI, Anthropic, Perplexity, DuckDuckGo, Mistral AI, and Common Crawl publish their IP ranges. Sorabun fetches each list at most once a day (it only reads a public list, and does not send this site's URL). When a fetch fails, the last list that was fetched is used for up to 7 days. "Are the IP ranges being fetched?" on the screen shows the current state.

What cannot be counted (more visits happen than are shown here)
  • Pages served by a page cache. A visit can be counted only when WordPress (PHP) runs. Pages served by WP Super Cache, WP Rocket, the Cloudflare cache and the like never reach WordPress
  • List pages such as the home page and categories. The count exists to show which articles were fetched, so only single article pages are counted
  • Google-Extended and Applebot-Extended. Both are names used only inside robots.txt. When fetching, they identify themselves as Googlebot and Applebot, so they cannot be told apart

If there are no records at all, besides having only just started counting, a page cache may be serving the pages, or robots.txt, a CDN or a WAF may be blocking the crawlers. You can check for blocks with "AI search crawlers" and "Fetching as an AI search crawler" in the technical audit.

Times people arrived from AI answers

The number of times someone clicked a link inside an AI answer and landed on this site. They are told apart in two ways.

  • The referrer (where they came from): chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai, grok.com, chat.deepseek.com, chat.mistral.ai, meta.ai, felo.ai, genspark.ai, duck.ai
  • utm_source on the link: ChatGPT adds utm_source=chatgpt.com to its links. When there is no referrer (for example when the link is opened from an app), this is used instead

The check runs inside the visitor's browser, and it notifies this site once only when the visitor came from AI. For ordinary visitors it adds no requests at all. It works even when a page cache is active.

How it counts
  • When the same person reopens the same article within 30 minutes, it counts as one
  • Reloads, and views by people who can write articles (logged-in editors and administrators), are not counted
  • To make fake numbers harder to inject, at most 20 per hour are counted from the same IP, and at most 5,000 per day for the whole site (the screen tells you on days when the limit was reached)
What cannot be counted
  • Google's AI Overviews and AI Mode, and Copilot inside Bing's search results. Their referrer is the same as ordinary search, so they cannot be told apart
  • People arriving from AI apps that add neither a referrer nor utm_source, and people with JavaScript turned off
  • The count told apart by utm_source alone also includes links from AI answers that were pasted into an email or a note and opened later (the screen shows this count in a separate column)

Top articles

The top 20 articles, ordered by people arriving from AI or by crawler visits. It helps you spot articles AI reads but nobody arrives at, and articles people arrive at with no crawler records (a cache is serving them).

What is stored and what is not

Only the date, the kind (crawler or arrival), the name of the source, the verification result, the article, and the count are stored.

  • IP addresses, full user agents, and referrer URLs are not stored
  • The limit on arrival counts uses the IP, but the IP itself is not stored. A hash mixed with the site's secret is kept for a short time only
  • Records older than 400 days are deleted every day. When you permanently delete an article, its rows are removed and its counts move to "pages other than articles, and deleted articles"
  • Deleting the plugin deletes the records

Turning it off

Under "Settings" at the bottom of the screen, clear "Count AI crawler visits and arrivals from AI" (aivisits_enabled). Counting stops, and the IP ranges are no longer fetched. Existing records stay.

On the Cloud edition's single and pack plans, counting also stops once the full-feature period has passed (the records can still be viewed).