Machine readers
See which AI assistants and search crawlers read your pages, and why.
Why this needs your server
A crawler asks for your HTML and leaves. It does not execute scripts, so no browser tracker on earth can see one. Machine tracking is a separate call you make from your own server or middleware, where the request actually arrives.
Add the call
Call this on incoming requests and forward the visitor's user agent. Termind decides what is worth recording, so you do not need to filter anything yourself.
export function middleware(request: Request) {
const ua = request.headers.get("user-agent") ?? "";
const url = new URL(request.url);
// Fire and forget. Never block the response on analytics.
fetch(
`https://www.termind.tech/v1/crawl?key=YOUR_KEY` +
`&ua=${encodeURIComponent(ua)}&path=${encodeURIComponent(url.pathname)}`,
).catch(() => {});
}Do not await the call. If Termind is slow or down, your page should still render instantly.
The three purposes
Most tools report one number for bot traffic. That number adds together three things that mean opposite things for your business, so Termind splits them.
| Purpose | What it means | Example |
|---|---|---|
| Answering | A person asked a question right now and was handed your page | ChatGPT-User, Claude-User, Perplexity-User |
| Indexing | You are being catalogued so you can be cited later | OAI-SearchBot, Googlebot, Bingbot |
| Training | Your text goes into a corpus. No reader at the far end | GPTBot, ClaudeBot, Google-Extended |
Nine hundred training fetches with zero answering fetches means you are being consumed and never quoted. Forty answering fetches means you are being recommended to real people. The totals look similar. The situations are opposite.