AI referrals at about 1% of traffic: read the benchmark's dates first
Conductor's 2026-titled benchmark reports 2025 data. How to put an older industry benchmark beside your own 2026 analytics without misleading a board.
The short answer
Check three things before an AI referral benchmark goes into a board pack: when the data was collected, whose sites it covers, and how AI referrals were counted. A report titled 2026 can describe last year, and a vendor's enterprise customers are not your market. Label the period on the slide and lead with your own trend.
Key takeaways
- A benchmark's publication date says little about its data: read the methodology for the collection period before quoting it.
- SE Ranking measured AI search at 0.32% of visits across 101,574 websites in January to April 2026.
- Two benchmarks can differ by a factor of three because they sample different sites, so the sample matters as much as the date.
- Referrer-based counts miss visits from apps and copied links, so any AI referral share is a floor rather than a total.
- Put your own monthly AI referral trend first on the slide and use an outside benchmark as a labelled, dated range.
In this article
Why the dates come first
An AI referral benchmark is only useful if you know when its data was collected, because the channel moves quickly. SE Ranking's study of 101,574 websites found AI search sent 0.32% of visits in January to April 2026, which it put at about 16 times the 2024 level.1 In a channel growing that fast, a figure that is a year old describes a different market.
Conductor published its 2026 AEO/GEO Benchmarks Report in November 2025 and last updated the page on 6 July 2026. The title says 2026. The methodology says the data is from 2025, in two separate windows. That is normal for benchmark reports and not a criticism of Conductor, which is open about its periods. It is a problem only when the headline figure travels without them.
Our view: the period belongs in the same sentence as the figure, every time. If a slide cannot carry it, the figure should not be on the slide.
What the Conductor report measured
The report holds two datasets with different periods, and they should not be mixed. The AI Overviews part covers 21.9 million US Google searches from 15 September to 12 October 2025, and found AI Overviews on 25.11% of them.3 Prevalence varied widely by industry: 48.75% of healthcare searches against 4.48% in real estate.3
The referral part is older and narrower: it covers a different window, earlier in 2025, and a smaller set of sites. Conductor is a vendor that sells AEO and SEO software, so that traffic sample is its own customers' sites. Both points matter more than the headline, and the next section gives the figure with them attached.
The headline figure, with its period
With its period attached, the headline reads as follows. The referral data runs from May to September 2025 and comes from US traffic on 1,215 enterprise customer domains on Conductor's platform, across ten industries.4 Across Conductor's sample, AI referrals were 1.08% of website traffic in May to September 2025, with information technology highest at 2.80%.4 ChatGPT sent 87.4% of those AI referrals.4 By the date of this note the referral data is between nine and fourteen months old.
Now set it beside the newer SE Ranking figure. SE Ranking found ChatGPT sent 74.78% of AI referrals in January to April 2026, with Gemini second.2 The older benchmark shows a higher AI share of traffic and a higher ChatGPT share. Time alone does not explain that, since the channel grew between the two periods.
| Conductor | SE Ranking | |
|---|---|---|
| Published | 12 Nov 2025 (updated 6 Jul 2026) | 18 Jun 2026 |
| Referral data period | May–Sep 2025 | Jan–Apr 2026 |
| Sites | 1,215 enterprise customer domains | 101,574 websites |
| AI share of visits | 1.08% 4 | 0.32% 1 |
| ChatGPT share of AI referrals | 87.4% 4 | 74.78% 2 |
Sources: Conductor 4; SE Ranking 1 2. Both vendors sell search software.
Why do two AI referral benchmarks disagree?
The two figures differ by a factor of about three because they sample different sites. Large enterprise domains in information technology and consumer goods attract more AI referrals than a broad web sample full of small sites. Neither figure is wrong. Each describes its own sample, and yours may look like neither.
The counting method adds a second gap. Both studies classify a visit as AI-referred from its referrer. Visits from mobile apps that strip the referrer, and from links a reader copies out of an answer, land in direct traffic instead. So every AI referral share is a floor. Our board-metrics guide covers which AI search numbers survive that problem.
Three data windows behind two reports quoted in 2026
Conductor's referral data, enterprise customer domains
Conductor's AI Overviews data, US searches
Conductor publishes its 2026 report
SE Ranking's referral data, broad website sample
SE Ranking publishes; Conductor updates its page
How to show a dated benchmark to a board
Show your own data first and the benchmark second, as a labelled range. A board compares the number on the slide with the number it remembers, so the slide has to carry the context.
- Lead with your trend. Twelve months of your own AI referral sessions and their share of total, from your analytics, with the counting rule in a footnote.
- Quote the benchmark with its period and sample. "1.08% of visits, May–Sep 2025, Conductor's enterprise customer domains" rather than "about 1%".4
- Show two benchmarks, not one. A range from two samples is more honest than a single point.
- Add what referrals miss. Mentions in AI answers and branded search lift do not appear as referrals. Our pipeline guide sets out how to connect them.
Our view: refresh the benchmark line each quarter and drop any figure whose data is more than a year old. A dated benchmark in a recurring report quietly becomes a target, and nobody remembers where it came from.
Sources
- SE Ranking, AI traffic study: 101,574 websites, 250 countries, Jan–Apr 2026 (Jun 2026)
- SE Ranking, AI traffic study: share of AI referrals by engine, Jan–Apr 2026 (Jun 2026)
- Conductor, 2026 AEO/GEO benchmarks: 21.9M US searches, 15 Sep–12 Oct 2025 (Nov 2025; page updated Jul 2026)
- Conductor, 2026 AEO/GEO benchmarks: AI referral traffic, 1,215 US enterprise customer domains, May–Sep 2025 (Nov 2025; page updated Jul 2026)
