ChatGPT sends 0.62 sessions per 100 Google organic sessions. Three October 2026 studies disagree on what that means.
ChatGPT sent 0.62 sessions for every 100 sessions Google organic sent, according to Marie Haynes, who published the number on October 21, 2026. Her sample covered 100 websites and more than 145 million sessions of GA4 data over 12 months. She titled the post "GEO is a distraction." The SEO world has argued about it all week.
The AI referral traffic vs organic gap is real. The conclusion she draws from it is where people split. This post lays out what three October studies found, where they disagree, and what you should measure in your own analytics before you move budget in either direction.
Track this metric for your brand → nobori.ai
Haynes found that ChatGPT sessions equaled 0.62% of Google organic sessions across 100 sites. Add every other LLM referral and the figure rises to 0.73%. She also reports that ChatGPT referral growth stopped being exponential around August 2025, near the GPT-5 release, and that ChatGPT has sent fewer visitors since.
The conversion numbers cut the other way. On the 55 sites that tracked conversions, ChatGPT visitors converted at 13.8%. Google organic visitors converted at 3.02%. That makes a ChatGPT visit worth about 4.6 times a Google visit.
Volume still won. Google organic produced 96.5% of the conversions in her dataset, which is 27 times what all LLMs produced combined. Her conclusion: "The AI search that is actually sending people to businesses today isn't ChatGPT. It's Google."
One caveat sits inside her own sample. 45 of the 100 sites tracked no key events at all, so the conversion comparison rests on 55 properties.
Yes on volume, with a twist on conversion. The SEO Works published an AI referral benchmark on September 7 and updated it on October 14. It covers 95 UK websites from September 2025 to August 2026. Organic search sent 2,825,594 sessions. AI referrals sent 72,095. Organic sent 39 times more traffic.
Conversion was closer than Haynes found. AI referral traffic converted at 7.18% and organic at 6.08%, a ratio of 1.18x. Against paid search the ratio was 0.67x, and against direct it was 2.48x. On 23 of the 63 sites with conversion data, AI traffic converted worse than organic. Five sites recorded no AI conversions.
B2B is the weak spot. For B2B products, AI referrals equaled 3.40% of organic sessions, and AI traffic converted at 0.62 times the organic rate. If you sell to businesses, the "AI visitors convert 4x better" line you have seen in vendor decks does not hold in this sample.
The same benchmark found AI and organic referral growth move together, with a +0.557 correlation across 61 websites. Sites that grow one tend to grow the other.
Run the math on 10,000 monthly organic sessions. At 0.62%, ChatGPT sends about 62 sessions. At Haynes's 13.8% rate that yields roughly 8.6 conversions. The same 10,000 organic sessions at 3.02% yield about 302 conversions. This is a Nobori calculation from her published rates, and your numbers will differ.
The SEO Works put the same trade-off in per-site terms. Doubling AI traffic adds 6.6 conversions per website per month. Growing organic by 5% adds 10.3. Doubling organic adds 205.8.
Read those numbers as a budget signal. A 5% organic gain beats a doubling of AI referral traffic. Nobody serious disputes that. The dispute is about whether referral sessions are the right yardstick for AI visibility at all.
Rand Fishkin of SparkToro said ChatGPT's 0.62% is less traffic than DuckDuckGo sends, and called AI "a digital billboard if you will." Mike King of iPullRank answered Haynes with "The numbers are right. The category is wrong."
King's argument rests on what happens before a click. He cites Semrush figures showing about 34% of sessions end without a click when no AI Overview appears, about 43% with one, and about 93% in AI Mode. He also points to 1.2 billion weekly ChatGPT users. People read the answer and leave. A referral report never records that visit.
Google's own data shows the surface is growing. A Semrush study reported on October 21 tracked 400 commercial keywords on US desktop. AI Overviews appeared on 88% of them, up from 45.75% in September 2025. Weighted by search volume, the share was 97.08%.
Similarweb noted that Google revamped AI Overviews and AI Mode on October 15, moving source cards higher in the answer. Google is also piloting payments to about 100 publishers whose content grounds its AI answers, with payouts visible in an experimental Search Console panel even when users never click.
Probably not with the tools you have today. Kevin Indig tested this in his October 19 Growth Memo using 12 GA4 properties. Direct traffic rose 140.9% from January to September, while engaged Direct sessions changed by -1.0%. One site logged 409,000 Direct sessions, and 198 were engaged.
Indig ran 632 tests across four brands and found no reliable link between AI visibility and later branded or Direct visits. AI traces made up 0.003% of Direct traffic. The theory that AI answers quietly feed your Direct channel has no support in his data.
That result matters for anyone who defends AI spend with "it shows up as brand search later." Indig's test did not find it. It does not prove the effect is zero, since 12 properties is a small sample, but it removes the easy excuse.
Read this together with our earlier look at AI Overviews on branded searches. If AI answers now sit on most brand-name results, the click you lose on your own name is a measurable cost, and it is a different cost from the missing AI referral.
SISTRIX released a citation worthiness study on October 22, written by Johannes Beus and Julia Weissbach. It covers 96,418 URLs from 17,380 domains in the DACH region for Q2 2026, including 48,527 cited pages, each paired with an uncited page from the same domain.
The multiples show how much more often cited pages carry a feature:
Pages with many of these features were cited 2.1 times as often, yet only 10% of cited pages had many of them. llms.txt moved the other way: 11.7% of cited URLs had one, against 14.0% of all domains. Our post on llms.txt for AEO covers why that file has not earned a place on your list.
SISTRIX also drew a line on method: "Don't trust a GEO study with fewer than 5,000 URLs." Apply that test to every number in this post. Haynes and the SEO Works measure sessions from 100 and 95 sites, and Indig works from 12.
No, and that is the core problem. Haynes, the SEO Works, and Indig all measure clicks that arrive on your site. King, Fishkin, and the Semrush data point at exposure that never produces a visit. SISTRIX measures which pages get cited. These are four questions with four different answers.
A citation is not a session. We covered a version of this gap in AI mentions vs citations, where fewer than half of the top 100 mentioned brands were also top 100 cited domains. A brand can win an answer and earn no click, or earn a click and miss the answer.
Referral traffic also depends on which engine sends it. Our look at AI chatbot usage vs referral share showed engines with large audiences and tiny click-out rates. A single "AI traffic" line in GA4 hides that spread.
Measure three layers and keep them separate. Each answers a different question, and the studies above only cover the first.
If layer one is flat and layer two is rising, you are winning exposure without clicks, and Haynes's math describes your situation. If layer two is flat, you have a visibility problem that no referral report will show.
Fix tracking before you cut anything. Haynes found 45 of 100 sites tracked no key events. You cannot judge any channel's value without conversion tracking, and that holds for organic as much as for AI.
Track this metric for your brand → nobori.ai
Three actions, each doable in a few days.
For a side-by-side of the tools that automate the second step, see our comparison of AI search analytics platforms for B2B growth teams.
Expect more datasets, and expect them to disagree. The Haynes and SEO Works samples run to roughly 100 sites each. Larger panels will land between "distraction" and "billboard." Google's publisher payment pilot and the October 15 layout change also suggest Google plans to keep more of the answer on its own page.
The safe read for a B2B team is narrow. Organic search still produces nearly all measurable conversions. AI referrals are small, growing in line with organic, and converting inconsistently. Exposure inside answers is large and unmeasured by sessions. Treat the first as your revenue base and the second as a separate line you track on its own terms.
Marie Haynes found ChatGPT sent 0.62 sessions per 100 Google organic sessions across 100 websites and 145 million sessions. The SEO Works found organic sent 39 times more sessions than AI referrals across 95 UK sites.
It depends on the sample. Haynes found 13.8% for ChatGPT against 3.02% for Google on 55 sites. The SEO Works found 7.18% against 6.08%, and 0.62 times the organic rate for B2B products.
Referral data alone cannot answer that. Referral sessions miss answers people read without clicking. Measure presence in answers and referral conversions separately before you cut or raise spend.
GA4 added an "AI Assistant" channel only in June 2026, and many AI answers produce no click. Build your own channel group from known AI referrer domains and read it next to citation tracking.
Get daily AI visibility alerts for your brand → nobori.ai
