> ## Documentation Index
> Fetch the complete documentation index at: https://docs.growthmention.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Analytics

> See which AI crawlers read your website, which pages they fetch or fail to fetch, and who arrives from an AI answer.

Prompt tracking shows what the AI assistants *say*. Agent Analytics shows the other half: what they **read** on your own site, and the people who **arrive** from their answers.

It's measured from your own server, so it sees things no other tool can — including the pages an AI crawler tried to read and couldn't.

## What it shows

| View            | What's there                                                                                                |
| --------------- | ----------------------------------------------------------------------------------------------------------- |
| **Overview**    | AI crawler visits over time, the share of your traffic they make up, and the errors they hit.               |
| **Pages**       | Each page of your site: how often AI crawlers read it, and how many people arrived on it from an assistant. |
| **Bot Visits**  | Each AI platform, how much of your site it reads, and what for.                                             |
| **Referrals**   | People who read an answer in an assistant and clicked through to your site.                                 |
| **Logs**        | Every individual crawler visit and referral.                                                                |
| **Integration** | Your ingest keys and install instructions.                                                                  |

## Three kinds of crawler

Not every AI crawler is doing the same job, and it matters which is which:

| Purpose        | What it's doing                                                 | If you block it                                       |
| -------------- | --------------------------------------------------------------- | ----------------------------------------------------- |
| **Search**     | Building the index an assistant answers from.                   | You become invisible to that assistant's answers.     |
| **User fetch** | Fetching a page right now, because someone just asked about it. | The assistant can't read the page it was asked about. |
| **Training**   | Collecting pages to train future models on.                     | Nothing, for your visibility today.                   |

So a search crawler getting **403 Forbidden** from your firewall or CDN is a problem worth fixing, and a training crawler getting it may be exactly what you intended. Agent Analytics shows the status code each crawler got, per page. See [Recognised crawlers](/agent-analytics/crawlers) for the full list.

## Only visits we can prove

Anyone can send a request claiming to be *GPTBot*. So every visit is checked against the addresses the AI company says its crawlers use, and **only visits that pass are counted** — in the totals, the charts and the logs alike.

That has two consequences worth knowing:

* **Impostors are left out.** A request pretending to be a known crawler from the wrong address never appears.
* **Crawlers that can't be checked are left out too.** Some companies don't publish their crawlers' addresses, so there's no way to prove a visit came from them. That covers mostly smaller crawlers — every crawler from OpenAI, Perplexity, Apple and Meta can be verified, as can the main ones from Anthropic, Google, Amazon and Mistral. xAI and Microsoft publish nothing, so their crawlers can't. See [Recognised crawlers](/agent-analytics/crawlers).

To confirm your integration is working even on a quiet day, the **Integration** page shows when each key last reported.

<Note>
  Your Agent Analytics numbers will be lower than a raw count from your server logs. That's deliberate: every visit shown is one we could prove.
</Note>

## The visit after the answer

People who arrive from an AI assistant usually land in your web analytics as *direct* traffic, because assistants often don't pass on where the visitor came from. Agent Analytics recognises them as they arrive and attributes each one to the assistant that sent them — the last step of a journey whose earlier steps you can see on the rest of the dashboard.

## Privacy

The integration runs on your server, and it decides what's sent before anything leaves it.

* **People's visits aren't recorded** unless you choose to. By default only requests from recognised AI crawlers are stored. The WordPress plugin has an opt-in setting to report every request, which is personal data under GDPR — turn it on only if your privacy policy covers it.
* **Arrivals from an assistant** are recorded as: which assistant, which page, when. Nothing that identifies the visitor.
* **Query strings are dropped** from every path — they routinely carry session IDs and email addresses.
* **Crawler IP addresses are deleted after 30 days.** They're only needed to verify the visit.

## Monthly allowance

Each plan includes a number of crawler visits per calendar month:

| Plan   | Visits per month |
| ------ | ---------------- |
| Watch  | 100,000          |
| Growth | 500,000          |
| Scale  | 1,000,000        |

If you reach it, collection pauses until the next month starts, and the plugin shows the reason. Visits already collected stay.

## Get started

<CardGroup cols={2}>
  <Card title="Install on WordPress" icon="wordpress" href="/agent-analytics/wordpress">
    Download the plugin and paste in your key. About two minutes.
  </Card>

  <Card title="Any other site" icon="code" href="/agent-analytics/http-api">
    Send visits from your server, CDN or edge worker over HTTP.
  </Card>
</CardGroup>