Skip to main content
The Coverage control on Home groups your sensors by capability, and Settings › Integrations is where you connect them. This page explains which sensors you need and where the alternatives are genuinely interchangeable.

Capabilities and their sensors

The crawler providers are alternatives, so DevTune never requires both for the same hostname. The tracking snippet and website analytics provide different measurements: the snippet records first-party referral page views, while GA4 or PostHog supplies referral-session reports across AI Traffic. Search Console and GA4 are complementary for Search performance: each adds different Google ground truth, so connecting only one reports Partial.

What to connect, in order

  1. One crawler sensor per hostname. Cloudflare pull suits hostnames already proxied through Cloudflare, including docs and static hosts. Edge middleware suits Vercel, Next.js, and other app hosts where request middleware can run. On Vercel, edge middleware can run without an ingest key once you link the Vercel project, and a Vercel log drain needs no code changes. Compare the three.
  2. Connect GA4 or PostHog. Select the events you count as conversions. PostHog also supports existing actions. Use one analytics provider per site.
  3. Optionally add the snippet for first-party referral visits - a record you own end to end, reportable over the API and independent of your analytics provider.
The minimum viable setup is one crawler sensor plus website analytics. Together they cover both halves of AI Traffic - what machines fetch, and what people do afterwards. The snippet is an addition to that pair, not a replacement for either.

Compare the crawler sensors

Each crawler sensor has its own setup guide: Cloudflare pull, edge middleware and the Vercel log drain.

Where each number comes from

Referrals reads the selected analytics provider on both sides: AI-referred sessions and their selected conversion events come from the same GA4 property or PostHog project, matched on landing page and the session source DevTune classifies as an AI engine. That keeps both halves of the events-per-session ratio on the same measurement basis. It links AI referrals to outcomes directionally; it does not follow individual visitors. Where a sensor is not connected, the number reads as no data and links you to Settings › Integrations. You only ever see a zero where the sensor was watching for the whole period, so an empty stretch is reported as no data rather than as none.

AI referral visits

Which sensors record them

Cloudflare records crawls and answer fetches only. The Vercel log drain can’t see where a visitor came from, only a utm_source in the landing URL. A host either reports still needs the snippet or edge middleware for its referral visits to be measured.

Which count a host reads

DevTune reads one source for each host and UTC day:
  1. Edge middleware, on a day it measures the host. A day with no referral visits reads 0.
  2. Otherwise the snippet, when it reported page views from the host that day. A day with no referral visits reads 0.
  3. Otherwise the day reads Unmeasured, with At least N beside it when edge middleware or the Vercel log drain recorded visits.
A host running both sensors switches between them as these conditions change. Crawls and answer fetches from Cloudflare, the Vercel log drain or an older middleware version never make a referral count read 0.

When the project-wide count is measured

The project-wide referral count is measured on a day only when every required host is measured that day. With no required hosts, it reads Unmeasured. Required hosts are:
  • your primary domains that have no path, such as example.com but not example.com/docs;
  • every host you connect to Cloudflare, and every host the Vercel log drain or edge middleware has reported.
example.com and www.example.com count as one host. A host Cloudflare, the Vercel log drain or edge middleware has reported stays required until you exclude it, even after you remove that sensor. Only a primary domain that Cloudflare, the Vercel log drain or edge middleware hasn’t reported, or a Cloudflare hostname whose connection never synced, leaves the list when you remove it. A host only the snippet has reported is never required, and its referral visits still count. When you filter to one host, the count reads that host’s own days. So one required host with no referral sensor holds the project-wide count at Unmeasured on every day in the window; for a Cloudflare hostname, from the day you connected it. Install the snippet or edge middleware on it, or exclude it.

Exclude a host

In Settings › Integrations, AI referral hosts lists the required hosts and the hosts you excluded. Select Exclude on a host no referral sensor can run on, such as a retired docs host or a staging host a sensor saw once. Its referral visits still count, but it no longer holds the project-wide count at Unmeasured. If you exclude every required host, the count reads Unmeasured. Select Include to make it required again. Members who can manage the project’s sensors see Exclude and Include; other members see the list without them. See roles and permissions.

Event allowance

Every referral visit the snippet, edge middleware or the Vercel log drain records counts against your monthly AI Traffic events allowance. A visit is charged once for each sensor that records it, but counted once in your figures.

Keep each site on one crawler sensor

Pick one crawler sensor per site and stay with it. Nothing stops you from switching both on: where Cloudflare already covers a hostname, its data wins for bot traffic there, and the other sensor’s bot requests for that host are not counted or charged. In version 0.13.0 or later, edge middleware also records AI referral visits on that host, and those are charged. For bot traffic, edge middleware or the Vercel drain takes precedence over the tracking snippet on a host. So if you find both active on the same hostname, remove or pause the one you do not want. Edge middleware is switched on for a whole project rather than for a single hostname, so on a project running both, Cloudflare wins on every hostname it covers.

How current each sensor stays

A sensor reports as Active while data from it has arrived recently. How recently depends on the sensor: The windows differ because the sources deliver differently: crawler and snippet events reach DevTune as they happen, while GA4, PostHog and Search Console are collected on a schedule and fill in history afterwards. Do not read a quiet site as a broken one - a hostname that goes an hour without an AI crawler hit still reports as Active.

Next steps