AI traffic benchmark: our data and the published range
We have no measurement of our own to publish yet. This page says so, shows what we will count and how, and puts the published numbers next to each other with the reasons they differ.
Last checked · How we check
Our own data
Sessions measured
n = 0
No sessions have been counted for this benchmark yet, so there is no percentage to show. A percentage from so few sessions would say nothing about the web, and we will not publish one. When counting starts, the first figure appears here with its n, its period and its method.
The published range
| Source | AI share of traffic | Sample | Status |
|---|---|---|---|
| SE Ranking | 0.32 % | 101,574 websites, Jan 2025 to Apr 2026 | Original read |
| Clickport | 0.7 % of sessions | 2,619 detected AI sessions, 30 days | Original read, self-selected sample |
| Conductor | 1.08 % | 13,770 enterprise domains | Secondary only |
Full table with periods, definitions and links: published AI traffic numbers. Do not average these rows: they count different things.
How we will measure
The plan is a first-party count on the sites the operator runs, taken from server logs without cookies: visits aggregated per day, bots excluded by user agent, and the share with an AI referrer or tag. Every figure will be published with its absolute number of sessions and the period.
- Below 1,000 sessions in a month we show raw counts only and no percentages.
- We name the number of sites as "websites of the operator" and do not name them.
- Numbers reported by others are shown separately as "reported, unchecked" and are never averaged into ours.
Why the numbers differ
Published figures for the AI share of traffic range from 0.32 % to 1.08 %, and for ChatGPT's share of AI referrals from 74.78 % to 92.4 %. They are not contradicting each other as much as counting different things.
- The denominator. All websites, enterprise domains or the customers of one analytics vendor. Large and small sites behave differently.
- The definition. Only visits with a matching referrer, visits with a tag too, or visits to the AI platforms themselves (Similarweb counts those, which is not traffic to your site).
- Direct. In Clickport's sample 35.7 % of detected AI sessions had a tag and no referrer. A study that counts referrers only misses them. Another vendor, Loamly, reports 70.6 % of AI visits without a referrer; its sample size is stated inconsistently (20,428 or 446,405 visits depending on the source), so we only mention it here and do not rely on it.
- Apps. Mobile and desktop apps often send no referrer, so they are invisible to referrer-based counts.
- The period. Behaviour changed during 2025 and 2026, and again in October 2026.
That is why this site does not publish a single "AI share of the web". It shows each number with its sample, period and source.
Sources
Read on . Links lead to the original pages.