Tracking

The Tracking page shows what AI platforms do on your site: people who arrive from an AI platform, the conversions they complete, and the AI crawlers that fetch your pages. Open it from the dashboard sidebar under Analytics > Tracking.

Tracking measures something different from your Visibility Score. The score comes from SurfacedBy's checks of what AI platforms say when asked. Tracking comes from a tracker installed on your site and records what reaches it. The two move independently: you can be visible with little traffic, or the reverse.

What you need

• A paid plan or an active trial. On other plans the page shows a locked preview.
• A tracker installed on your site. Until one reports in, the Tracking page shows the install options instead of data.

Each install method records a different set of signals:

The browser snippet cannot see crawlers, because crawlers do not run JavaScript. To record crawler activity, use one of the server-side methods. Crawler requests answered entirely from a page cache or CDN never reach the WordPress plugin, so it does not record them; its browser script still records AI visitors on cached pages. The Cloudflare edge tracker records requests before they reach your server. See WordPress plugin for the plugin, and Conversion Tracking for the install steps of every method and the conversion side.

Setting up

Setting up, rotating, or removing a tracker requires the Admin role or higher. Viewing the Tracking page is open to every workspace member.

Every member who can open the domain, Viewers included, can reveal a collector secret or the conversions webhook secret once it exists, so a developer you add to the workspace can finish an install. Creating the first secret, which happens on its first reveal, needs an admin on a paid plan or an active trial. Rotating a secret needs an admin on a paid plan or an active trial, and disconnecting an installation needs an admin.

1. Open Tracking for the domain. Before any tracker reports in, the page lists the install methods and marks the one that matches your site when it can detect it.
2. Follow the steps for your method. Each one uses the domain's Site ID shown on the page.
3. Once the tracker reports in, the page switches to the live view. Use Manage installation in the page header to see each tracker's status and last report, rotate or revoke the Site ID, rotate collector secrets, or disconnect the installation.

The install chooser orders methods using the edge, hosting and DNS evidence it can read. Cloudflare nameservers do not prove you control the Cloudflare account. Each recommendation explains the access it needs; if the site blocks detection, retry or choose a method whose prerequisites you can meet. A Vercel project needs Drains access, and its log drain needs a browser tracker alongside it for visitor sessions and conversions.

If the WordPress plugin stops reporting, the page shows a banner with a reconnect action instead of presenting the silence as zero traffic.

What the page shows

The time window control offers the last 24 hours, the last 7 days, or the last 30 days. From top to bottom:

• Headline tiles. AI Visitors, AI Conversions, Attributed Revenue, and AI Bot Crawls for the selected window.
• AI share. Lower-bound shares for the selected window: AI-referred human landings as a share of all human landings, and verified retrieval/agent document loads as a share of those loads plus human document loads. Training and indexing are excluded from the second share. Each share keeps its own sample and collector coverage; a small sample or unavailable capability shows an explanation instead of a percentage.
• Purpose cards. Verified crawler requests grouped by the agent's registered purpose:
  • Retrieval - fetching a page in response to a user request. A fetch does not prove that an answer cited it.
  • Search indexing - crawling so your pages can appear in AI answer engines.
  • Training - collecting content to train future AI models.
  • Agent and other-purpose activity appears when observed.
• Topics. Your prompt groups connect checked questions to cited or semantically linked site pages. Demand shows the selected window's AI landings and verified retrieval or agent fetches on those pages. Citation share uses the current completed check. Country and language filter the prompts; anonymous page demand has no country or language label. A page counts once within each topic, and topics can overlap.
• Pages. One table combines AI visits, verified reads and pages needing investigation. Use All, From AI visits, Read by AI or Needs attention to narrow it. Needs attention includes cited pages without a recorded AI landing in the selected window and crawler access changes. Missing attribution is a possibility, not a measured hidden visit. Crawl-silence findings use a separate nine-day activity check to find regular page/crawler pairs that went silent for 48 hours. Open a page for its scoped visits, reads, conversions and linked prompts.
• AI referrals. Real visits attributed to an AI platform, by platform, with the conversions each one produced. Open a row to see the pages those visitors landed on. Visits from google.com are hidden by default under Hide ambiguous, because the same address serves ordinary search and Google's AI features; they never count as AI visits.
• Conversions and Trials. Purchases, signups, and free-trial starts from AI-referred visitors. Amounts retain their reported currency; mixed currencies are shown separately. See Conversion Tracking.
• Agent responses. Observed Markdown requests and responses, average time until response headers by collector, and failed verified retrieval/agent requests. This is response evidence, not a complete-download latency or a readiness score.
• Supporting details. Expand this section for crawler activity, the When AI is active grid and industry data. Open a crawler for its pages and blocked responses. The grid uses its own fixed four-week window in your local time. Industry figures come from Cloudflare Radar and provide context rather than measuring your site; unavailable panels stay hidden.

The export menu in the top bar offers Export tracking (CSV) for the selected window.

Page detail

Clicking any page path opens its detail view: requests over the selected window, the crawlers behind them, whether each request succeeded or hit an access error, and a chart that puts crawler activity next to recorded AI referral visits for that page.

Verified crawlers

A request that uses an AI crawler's name is not proof of who sent it. A signed collector can establish identity using the operator's supported check: published addresses, supported network evidence, or a valid request-bound Web Bot Auth signature. Only positively verified requests enter headline crawl counts.

Rejected identities, identities that could not be checked, and agents with no published verification method remain visible as excluded observations. None of them count as verified crawls. A missing client address or a proxy address does not prove a failed identity check.

Collector coverage explains whether measurement is complete, partial, unavailable, or requires an update. Browser blocking, page caches, missing health evidence, collector losses and older versions can limit coverage. Two collectors observing the same hour do not count the same traffic twice. Counts from before the verified-only measurement change are not compared with the new formula as growth or decline.

Suggestions from crawler activity

Verified retrieval activity and anonymous page demand help select public pages for weekly prompt suggestions. Questions are grounded in the page and your category, then reviewed for buyer intent before qualified suggestions appear under Suggested by AI on Tracked Prompts. Review a suggestion before tracking it. Public page content can establish its language, but does not establish a buyer's country.

Repeated verified retrieval fetches on a page that your checks have not cited can also produce an "AI reads pages you are not measuring" action. This diagnostic uses a 30-day observation window. A server-side collector supplies crawler evidence; browser activity alone cannot establish a crawler fetch.

Limits of what this measures

• A crawler fetching a page does not prove an AI answer used or cited it.
• Some AI platforms send visitors without an identifiable referrer, so AI Visitors undercounts real AI traffic. The Needs attention view in Pages helps investigate cited pages without recorded AI visits.
• Crawler requests and human visits are counted separately and never combined.