AI referrals are 1.08% of visits, and the referral log is the wrong instrument

Conductor's 2026 AEO / GEO Benchmarks Report puts AI referral traffic at 1.08% of total website visits across 13,770 domains in 10 industries between May and September 2025, growing at roughly 1% month over month during the study window ^[1][2]. That is a number to log, not a number to re-plan around, and the more expensive error is treating referral data as a proxy for AI-driven discovery in the first place ^[10][15].

Start with the size. One in roughly a hundred visits ^[3] is inside the noise band of most analytics stacks. Read the 1% monthly figure as compounding relative growth, and the channel needed about 70 months, near six years, to double from where it sits ^[13]. Back out four monthly intervals from May to September and the share entered the study period at about 1.04% and left at 1.08% ^[16]. The entire measured growth across the window is roughly four hundredths of a percentage point. Nothing in that justifies a reorganised content roadmap.

Then the concentration. ChatGPT produced about 87.4% of AI referrals in the analysis ^[4], which works out to about 0.94% of all visits from one platform and about 0.14% from everything else combined, or roughly one visit in 735 ^[11][12]. A benchmark that is seven-eighths one vendor is a report on that vendor's product decisions, not on a channel. Industry spread is wider than the trend: IT came in at 2.8% and Consumer Staples around 1.9%, while Communications Services and Utilities were lower ^[5][6]. IT's share is about 2.6 times the cross-industry average ^[14], which means the average is a directional reference and not a target for anyone.

The substantive point is the instrument. A referrer records a visit that arrived from a traceable source, so it describes traffic that reached the site and nothing else ^[7]. A user can see a brand cited in an AI answer, run a branded search later, type the URL directly, or take the answer and never click at all ^[7]. Every one of those paths is invisible in referral reporting, and the last one produces no visit to attribute. Conductor's own framing is that AI answers and AI Overviews form an additional discovery layer that shapes consideration before any measurable session exists ^[15]. So the 1.08% is a floor on influence and a ceiling on what the log can see, and the gap between the two is unmeasured rather than zero.

Publishers should also stop borrowing the headline number. Conductor studied brand domains across industries; publisher-specific reporting cited in the same research put AI sources at under 1% of publishers' pageviews ^[8]. Different populations, different metrics, both true.

What to watch: whether the 1% monthly rate holds or bends once assistant products change their linking behaviour, since an 87.4% concentration means one product decision moves the whole benchmark ^[4]. Watch your own branded search and direct trends alongside AI-referred sessions as context rather than causation, and track citations and brand inclusion in AI answers as a separate exposure metric ^[9]. Watch, too, for internal ownership of indexing, content access, and link-sharing policy, because those rules determine whether you appear in the layer you cannot measure ^[9]. Baselines by industry and segment are worth more here than any single cross-industry percentage ^[9].