Which study you read decides the answer: AI assistants send between 0.2% and 6.2% of website sessions. The three recent datasets disagree because they measured different sites in different periods.

Each figure describes a different type of site, and none of them describes yours.

What the three studies found

Conductor found that 1.08% of sessions came from AI referrals. Its 2026 benchmarks report covers May to September 2025 and draws on 1,215 enterprise customer domains. ChatGPT supplied 87.4% of those referrals. Information technology sites saw 2.8% and consumer staples sites 1.9%.

Contentsquare found 0.2% of visits. Its 2026 Digital Experience Benchmark covers 99 billion sessions across 6,500 sites. AI-referred traffic grew 632% year on year, and those visitors converted at 1.3% in 2025, below the 1.9% for email. Services sites had the highest share at 0.5%, with software at 0.3%.

First Page Sage found 6.2% in July 2026, up from 0.1% in January 2023. It analysed just over 3.4 billion sessions across 218 client websites. B2B SaaS sites reached 11.8% and professional services sites 9.4%. It counts only sessions that arrive from an AI assistant, and treats Google AI Overviews separately.

Why the numbers differ

Two things explain the spread: the period each study covers and the sites it measured.

Conductor's window closed in September 2025. First Page Sage's own quarterly series shows how fast the number moves: 1.29% in the fourth quarter of 2024, 5.13% in the first quarter of 2026 and 6.24% in the second.

First Page Sage's 218 sites are clients of an agency that sells SEO and AEO services, so they are not a random sample of the web. Conductor's sites are large enterprise domains. Contentsquare's 99 billion sessions are the largest sample of the three.

The 1.08% figure is a 2025 measurement of enterprise sites. It is not a current benchmark for a UK services firm.

You can check your own share directly. Open your analytics, filter referral sources for chatgpt.com and perplexity.ai, and divide by total sessions for the last 90 days.

Does a low referral number mean AI has little influence?

Not necessarily. But the evidence for more influence is one study, and it has limits.

Similarweb ran the only study I found that tests this directly, and Salience's write-up sets out its method. It tracked US desktop users from July to December 2025 in finance, travel and beauty. Each had received a brand recommendation from ChatGPT and had not visited that site in the previous four weeks. Within seven days, they visited the recommended brand's website at 2.5 times the rate of a competitor that was not recommended.

Over half of those visits did not arrive as AI referrals. 55.9% came through search, against around 40% for standard visits. The recommendation happened inside ChatGPT, and your analytics logged a search.

Rand Fishkin, who contributed to the study, names the open questions himself. Would these people have found the brands anyway? Does it hold for small brands? How does an AI recommendation compare with ranking first in Google?

The brands tested were large consumer names such as American Express and Sephora. Fishkin says he is not entirely sure whether the pattern applies to B2B.

What to measure instead

A Goodfirms survey of 100 SEO and marketing professionals found that only 11% monitor branded search volume or share of voice. That is the metric that would show someone searching for your name after seeing it in an AI answer. We cover the tools teams use for this in what marketing leaders use to monitor brand visibility in ChatGPT and Perplexity.

Branded search in Search Console is the first place to look. If searches for your name rise after your AI visibility improves, that matches Similarweb's pattern. It does not prove cause.

I also looked for independent evidence on how often buyers name an AI assistant when asked how they found a business. I found vendor blog posts quoting percentages and no study I could verify, so I have left those figures out. That makes your own enquiries the evidence.

Add a free-text field to your enquiry form: "How did you hear about us?" Free text lets people write "ChatGPT" themselves. A dropdown can only record the options you thought of. Read the answers each month.

A referral share and a recommendation share are different numbers. Only the second tells you whether AI is sending you buyers, and only your own buyers can supply it.

A few common follow-up questions

Is AI referral traffic only 1% of visits?

It depends on the study. Conductor measured 1.08% in 2025 data, Contentsquare 0.2%, and First Page Sage 6.2% in July 2026 on its own clients' sites. Your share is in your referral report.

Should I stop investing in Google?

No study here suggests it. Organic search was still 42.8% of sessions in First Page Sage's July 2026 panel, down from 51.3% in January 2023. AI platforms were 6.2%.

Why would AI influence show up as search traffic?

Someone sees your name in a ChatGPT answer and searches for it later. Similarweb found that 55.9% of AI-influenced visits arrived through search, and your analytics logs those visits as search.