About 70.6% of AI referrals arrive with the referrer stripped and land in Direct. Dark-AI sessions convert near 10.21% against a 2.46% genuine-direct baseline, so an inflated Direct conversion rate is the tell. This is a model, not a measurement. Label it as one in any deck.
The decomposition appears here.
What the Dark AI estimator does
Roughly 70.6% of generative AI referral sessions arrive with the HTTP referrer header stripped, so GA4 files them under Direct. Those sessions convert at around 10.21% against a 2.46% genuine-direct baseline, which means a Direct bucket with an unusually high conversion rate is hiding your highest-intent AI traffic. This tool solves the mix: given your Direct sessions and conversions, it backs out how many are likely dark AI and what revenue that represents, then cross-checks the count against the 70.6% referrer-stripped share.
How to use it
- Pull monthly Direct sessions and the conversions attributed to Direct from GA4.
- Enter your average order value or annual contract value.
- Read the implied dark-AI session count, the hidden conversions, and the monthly revenue they represent.
- Treat the output as a directional model, and confirm it by tagging placed URLs and cohorting Direct by landing page.
How does it separate dark AI from genuine direct?
It solves one equation: your total Direct conversions equal dark-AI sessions at a 10.21% rate plus genuine-direct sessions at a 2.46% rate. Given the totals, only one split of the bucket fits, which yields the implied dark-AI session count. It then divides by 0.706 to estimate total AI sessions, since only about 29% arrive with the referrer intact.
Where do the 10.21% and 2.46% rates come from?
Published 2026 measurement: Dark AI sessions convert at roughly 10.21%, a 4.1x premium over the 2.46% baseline for genuine direct traffic. Your own rates will differ, so treat the output as a direction. If you can measure your visible AI-referral conversion rate, substitute it for a tighter estimate.
Why is this a model and not a measurement?
Because the referrer is genuinely gone, so you cannot observe these sessions directly. The estimate assumes your conversion rates match the published benchmarks and that the Direct bucket is only dark AI plus genuine direct. Label it as an estimate in any deck, and reduce the guesswork by tagging every URL you place and cohorting Direct traffic by landing page, since dark AI lands on deep informational pages, not the homepage.
Does my data leave the browser?
No. The decomposition runs entirely in your browser. Nothing you enter is uploaded, stored or sent to any server.