Skip to main content
DevTune checks your AI Traffic every day. When an AI platform fetches your pages for answers, or crawls your site, much more or much less than usual, DevTune records a detected change. Each change states what moved, on which platform and over which dates. It never states a cause. For example: OpenAI fetched your pages less often. 120 answer fetches from Oct 16 – 22, down from a weekly average of 410 from Sep 18 – Oct 15.

​
What is detected

DevTune compares each AI platform separately, such as OpenAI, Anthropic or Perplexity, and two kinds of traffic:
  • Answer fetches: an AI system retrieving a page to build an answer.
  • Crawls: training crawlers and search-index crawlers reading your site.
Each day, the last 7 days are compared with the weekly average of the 4 weeks before them. Referral visits are not part of detection.

​
Thresholds

A change is recorded only when all of these hold:
  • The last 7 days differ from the weekly average by at least 50%.
  • The difference is at least 50 events. Small sites and quiet platforms do not produce changes from a handful of requests.
  • At least 3 of the 7 days were 50% or more above or below the daily average, in the same direction, so a single unusual day is not enough.
Changes appear on your charts only when DevTune is confident in them: at least 5 of the 7 days moved the same way and the difference is at least 100 events. Rises and drops use the same rules.

​
Sensor outages are excluded

Only edge middleware and the Cloudflare log pull see answer fetches and server-side crawls, so detection needs at least one of them. A project with only the tracking snippet is not compared. Detection counts only traffic on hostnames, and for Cloudflare the paths, that edge middleware or the Cloudflare log pull covers. Traffic the tracking snippet records on any other hostname is left out. www. and the bare hostname count as the same site. DevTune compares a project only when its sensors reported throughout both periods. It does not compare a project when:
  • A sensor started, stopped or was replaced during the 5 weeks being compared. Rotating the edge middleware’s ingest key counts as a replacement.
  • Edge middleware has not reported any event since the last day being compared.
  • A Cloudflare integration is paused, has a sync error, or has not synced past the last day being compared.
DevTune also compares each platform and kind of traffic only when, on every hostname where it appeared during the 5 weeks, it was recorded on every day. A day with none could be a sensor that stopped and recovered. Other platforms and other kinds of traffic don’t count toward this, so a busy crawler can’t hide a gap in answer fetches. Because of this:
  • A complete stop isn’t reported, because it can’t be told apart from a sensor outage.
  • Traffic that doesn’t arrive every day isn’t compared.
  • A sensor that misses only some requests can’t be detected, so a partial outage can look like a real change.
If traffic for the 5 weeks is still being processed, DevTune checks the project again later instead of comparing partial counts. While a project, or one platform’s traffic, cannot be compared, its open changes are hidden. If the sensors recover and the change still holds, it reappears. After 3 daily checks that cannot compare, the change closes and is removed.

​
Where changes appear

Detected AI Traffic changes appear as markers on the Visibility, Audit and Results charts, marked Detected by DevTune. Use the Events filter on any of these charts to show or hide AI Traffic changes. They are hidden until you turn them on, and your choice applies to every chart in that browser. When a change ends, its marker stays as history with a Resolved label. If the same movement returns within a week, the same marker is updated rather than a new one added. Detected changes are available on every plan that includes AI Traffic. They are not sent by email or Slack.

​
API and MCP

The same changes are returned by GET /api/v2/projects/{projectId}/timeline/events and the devtune_get_timeline_events MCP tool, with kind ai_traffic_change and a trafficChange evidence object. See the API reference.