Skip to main content
The Direct traffic block in AI Demand compares GA4 direct sessions with a baseline for pages DevTune can measure. It highlights changes that line up with AI answer exposure without claiming that AI caused or owns those sessions.

​
Open the Direct traffic view

  1. Open a project
  2. Select AI Demand in the project sidebar
  3. Open Overview
  4. Scroll to Direct traffic
The block requires connected GA4 data. Its comparison also needs enough Search Console and AI-answer coverage to evaluate the same page and time window.

​
Read the chart

The chart keeps the verifiable GA4 total separate from the pages used for the comparison:
  • GA4 full direct total bars show all direct sessions that GA4 reports for the project.
  • Measured-page actual shows the daily equivalent of direct sessions for pages with complete comparison data.
  • Measured-page baseline shows the same measured pages against their own recent baseline.
  • A shaded AI-correlated lift span marks a window where all three correlation checks passed.
The lines may sit below the bars because the full GA4 total can include streams, hosts, or landing paths that are outside the measured-page population. Compare the two lines with each other. Do not treat the gap between a line and the bars as lift. The population note below the chart tells you how many pages contributed to the lines on the latest date. A smaller population means the comparison covers less of the full GA4 total.

​
How AI-correlated lift is evaluated

DevTune evaluates trailing seven-day windows. A window is marked AI-correlated lift only when all three legs hold:
  1. Direct traffic rose materially. The measured pages are at least 10 sessions above baseline and at least 125% of baseline. DevTune looks at the four preceding seven-day windows and takes the median of those with enough historical coverage. Two, three, or all four can contribute; uncovered windows are left out. If fewer than two are available, the current window reads as Unmeasured instead of receiving a lift classification.
  2. AI answer exposure held. The page had at least one citation or at least five answer fetches, and the qualifying signal did not fall by more than 25% from the prior window.
  3. Another measured channel did not explain the rise. Search Console clicks, paid sessions, and non-AI referral sessions are checked against the direct-session increase. A channel rules out the AI-correlated label when it grew by at least 10% and by enough sessions to match at least 25% of the direct lift.
Citations take precedence over answer fetches when both are fully measured. Training crawls and visible AI referrals do not qualify the answer-exposure leg.
AI-correlated means the signals moved together under these checks. It does not attribute direct sessions to AI, identify an AI platform, or prove causation.

​
Inspect a page-level result

Open Audit → Pages, then select a page to open its inspector. The Direct traffic section shows one of five states:
  • AI-correlated lift — all three legs passed, with the measured values shown.
  • Unexplained lift — direct traffic rose materially, but AI activity did not qualify.
  • Other-channel lift — another measured channel grew enough to match the increase.
  • Baseline — no material lift was detected.
  • Unmeasured — GA4, Search Console, or AI-evidence coverage was incomplete.
An AI-correlated traffic lift item can also appear under Needs attention after the result persists across two non-overlapping seven-day windows in one uninterrupted episode.

​
Why a highlighted span can disappear

The baseline adapts. When a higher traffic level persists, it becomes part of the recent baseline and the AI-correlated lift highlight can fade. The label detects the onset of a change; it is not a permanent attribution label.

​
Next steps