Across 100 websites, Marie Haynes found ChatGPT delivered 0.62% as many sessions as Google organic search, in a study published on 21 October 2026. Two more studies that week, from The SEO Works and Kevin Indig, pointed the same way, so marketing leaders should keep AI search budgets modest and stop judging them on referral clicks.
Key takeaways
- Marie Haynes counted ChatGPT sessions at 0.62% of Google organic volume across 100 websites and over 145 million GA4 sessions, rising to 0.73% with every other AI assistant added.
- The SEO Works counted 72,095 AI assistant sessions against 2,825,594 organic sessions across 95 sites between September 2025 and August 2026, so organic sent 39 times more traffic.
- AI visitors converted better than organic visitors in both studies, at 13.8% against 3.02% in Haynes’s data and 7.18% against 6.08% in The SEO Works’ data, but at 0.67 times the rate of paid search.
- Kevin Indig found on 19 October 2026 that direct sessions with identifiable AI traces made up 0.003% of direct traffic across 12 sites.
- The SEO Works left Google AI Overviews out of its count, because GA4 records their clicks as ordinary organic sessions.
Three studies, one direction
Search Engine Roundtable grouped the three together on 22 October. Each used a different sample and asked a slightly different question, which makes the agreement more useful than any one of them.
| Study | Published | Sample | AI traffic found | Conversions |
|---|---|---|---|---|
| Marie Haynes, GEO is a distraction | 21 October 2026 | 100 sites, 12 months of GA4, over 145 million sessions | ChatGPT at 0.62% of Google organic sessions | ChatGPT 13.8% against Google organic 3.02%, with Google organic driving 27 times more conversions than all AI assistants combined |
| The SEO Works, AI referral traffic benchmark | Data to August 2026 | 95 GA4 properties, 63 in the conversion comparison | AI assistants at 2.6% of organic sessions | AI 7.18% against organic 6.08%, or 3.2 AI conversions for every 100 from organic |
| Kevin Indig, Growth Memo | 19 October 2026 | 12 sites, mostly B2B, August against September 2026 | AI traces in 0.003% of direct sessions | Not measured |
Haynes also reports that ChatGPT’s referral growth stopped in August 2025, around the GPT-5 release. The SEO Works found a gentler rise, with AI referrals climbing from 2.29% to 3.04% of organic sessions from May 2026. Both put AI assistants below about 3% of organic search sessions.
Do AI visitors convert better?
Against organic search they mostly do, and the gap shrinks as the samples get larger. Haynes’s ChatGPT visitors converted at more than four times the Google organic rate, but only 55 of her 100 sites tracked conversions at all. The SEO Works pooled 63 sites and found a 1.18 times advantage over organic, with AI converting worse than organic on 23 of them. AI traffic converted at 0.67 times the rate of paid search and at 2.48 times the rate of direct traffic.
The SEO Works also split its results by business type, putting B2C sites at 1.51 times the organic rate and B2B product sites at 0.62 times. Its conversions are GA4 key events, which the study treats as upper-bound counts, so a form view or a download can sit beside a sale.
A higher conversion rate on 2.6% of the traffic still adds up to a small number of customers, so for most brands the conversion story isn’t enough to justify a large AI search budget on its own.
Is GEO a distraction?
Haynes argues it is, for most sites, and recommends that teams stop prioritising ChatGPT citations and work on Google Search and Google’s own AI features instead. Her data supports the first half. The second half needs a caveat that The SEO Works spells out, because clicks from Google AI Overviews arrive in GA4 as organic sessions. When AI Overviews appear on most of a brand’s searches, as DemandSphere’s September data suggested, some of what these studies call Google traffic already comes through an AI answer.
The studies also measure what happens after a click. An assistant that names your brand in an answer can shape a shortlist without sending anyone to your site, which is the argument we made in the citation moved off the page. Indig went looking for that effect in direct and branded traffic and couldn’t find it reliably. Across his filtered January to September sample, total direct sessions rose 140.9% while engaged direct sessions fell 1.0% in monthly counts, and one site logged 409,000 direct sessions of which 198 were engaged. Teams that read rising direct traffic as an AI dividend are often counting bots.
Manual Focus sells AI visibility measurement through Light Meter, so weigh what follows with that in mind. Our view at Manual Focus is that these studies settle the referral question, since AI assistants are a small traffic source for most brands, and leave the influence question open, so AI search work should be sized to a modest share of the SEO budget and judged on how often buyers see your brand in AI answers and on what new customers say, with referral sessions as the third signal. Indig recommends a similar monthly report, with commercial results such as qualified leads, purchases and revenue as his third signal.
What to change in your reporting
- Work out your own ratio. In GA4, match session source against the pattern The SEO Works used,
bard|chatgpt|openai|claude|copilot|gemini|perplexity|grok, and compare those sessions with organic search over 12 months. If you land near 2% to 3%, plan your budget around that figure. - Clean direct traffic before anyone reads it. Split direct sessions by engagement, landing page and browser, as Indig did, and look for spikes of unengaged visits before crediting AI with any growth.
- Ask new customers where they heard of you. Add AI assistants as an option in the lead form or post-purchase survey. Indig cites Ramp, whose self-reported answers attributed 30% to 40% of its direct traffic to answer engines, a figure no analytics tool would have shown. Our attribution tear-down playbook covers how to weigh self-reported answers against platform data.
- Track exposure on a fixed prompt set. Pick the questions your buyers ask, run them on the same engines each month and record whether your brand appears. Repeat each question several times, because single answers vary from run to run.
- Fund Google’s AI surfaces first. Google’s AI Overviews and AI Mode sit inside the channel that still drives most of your traffic, and our post on Google’s new AI Mode links covers how brand sites earn those clicks.
What the studies can’t tell you
Haynes calls hers “my small dataset” and doesn’t say how the 100 sites were chosen beyond random sampling. The SEO Works gives customer types but not sectors or countries, and doesn’t break results down by assistant. Indig’s sample skews B2B and US, and his visibility tests rest on 17 and 35 usable weekly observations, which he notes limits their precision. None of the three measures AI Overviews or AI Mode clicks separately, because GA4 doesn’t show them. The trend is also unsettled, with Haynes seeing ChatGPT referrals flat since August 2025 and The SEO Works seeing a rise since May 2026. Retail could look different in the holiday season, where Adobe forecast AI referrals to US retail sites up 130% year on year over November and December.
Further reading
- The citation moved off the page, on why AI influence often leaves no click to count
- Manual Focus Light Meter, on measuring brand visibility across AI engines with repeated runs
- AI Overviews hit 82% of tracked brand searches, on Google’s AI answers sitting inside organic traffic
- Attribution tear-down, for testing channel claims against your own evidence
- What Adobe’s AI shopping forecast means for Black Friday, for the retail case where AI referrals are growing fastest
If you want help setting an AI search budget your finance team will accept, our services page explains how we work.
Frequently asked questions
How much traffic does ChatGPT send compared with Google?
Very little, in the studies published in October 2026. Marie Haynes found ChatGPT sent 0.62% as many sessions as Google organic search across 100 websites and over 145 million GA4 sessions. The SEO Works found AI assistants together sent 2.6% of organic search traffic across 95 websites between September 2025 and August 2026, so organic search sent 39 times more.
Does traffic from AI assistants convert better than Google?
Against organic search, it did in two October 2026 studies. Marie Haynes measured ChatGPT visitors converting at 13.8% against 3.02% for Google organic. The SEO Works measured AI referrals at 7.18% against 6.08% for organic, but at 0.67 times the rate of paid search, and AI converted worse than organic on 23 of the 63 sites it compared. Volumes stayed small either way.
Is GEO worth investing in for marketing teams?
For most brands it deserves a modest budget, sized to what AI assistants return today. Three studies from October 2026 found they send a small share of Google's traffic and conversions. The Manual Focus view is to keep work on how AI answers describe your brand, and to judge it on how often buyers see you in those answers and on what new customers report.
Does AI visibility increase direct or branded search traffic?
Not measurably yet, according to Kevin Indig. In a Growth Memo study published on 19 October 2026 he ran 632 tests across four brands and could not establish a reliable link between AI visibility and later branded search or direct traffic. Across 12 sites, direct sessions with identifiable AI traces made up 0.003% of direct traffic, and much of the recent growth in direct sessions came from bots.
Sources
- GEO is a distraction, Marie Haynes
- How much traffic do AI assistants actually send, and does it convert better?, The SEO Works
- Does AI drive more direct traffic?, Growth Memo
- 3 New Studies Suggest AI Visibility and Search Not Worth Investing In, Search Engine Roundtable
From the podcast
James and Hannah talk through three studies covered together on 22 October 2026, from Marie Haynes, The SEO Works and Kevin Indig, which all find that AI assistants send a small share of the traffic and conversions Google organic search does. They cover whether AI visitors really convert better, why Google’s AI Overviews blur the line between Google and AI traffic, why a spike in direct traffic is often bots, and how to size and measure AI search work. Manual Focus sells AI visibility measurement through Light Meter, and James says so on air. Then three jobs to hand to AI.
Get AI on this
Paste these into whichever AI assistant you use. Swap the bracketed parts for your own details. All three need file uploads, and the third works best set up to run monthly in a tool that can run on a schedule.
1. Work out your AI traffic ratio
I run marketing for [brand] at [website], selling mainly in [market].
Studies published in October 2026 found AI assistants sent between about 0.7%
and 2.6% as much traffic as organic search. I want to know where we sit.
Attached is a Google Analytics export of sessions and key events by session
source / medium, by month, for the last 12 months.
1. Treat as AI assistants any source matching this pattern, ignoring case:
bard|chatgpt|openai|claude|copilot|gemini|perplexity|grok
2. Treat as organic search any row with medium "organic".
3. Give me a table with the columns: Month | Organic sessions | AI assistant
sessions | AI as % of organic | Organic key events | AI key events |
Organic conversion rate | AI conversion rate.
4. Add a totals row for the 12 months, and a second table splitting AI
sessions by assistant.
List any sources you weren't sure how to classify, and don't estimate numbers
that aren't in the file. Note that clicks from Google AI Overviews appear as
organic sessions and can't be separated in this data. If you couldn't check
something, say so.
2. Clean up direct traffic
I need to know how much of our direct traffic is real before anyone reports
it as an effect of AI search.
Attached is a Google Analytics export of direct / (none) sessions for the last
6 months, by day, with landing page, browser, country, sessions and engaged
sessions.
1. Flag days where direct sessions jumped but engaged sessions didn't.
2. Flag landing pages, browsers or countries with a high share of unengaged
direct sessions.
3. Leave out home, pricing, login and demo pages, then give me a monthly
table with the columns: Month | Direct sessions | Engaged direct sessions |
Engaged share | Change in engaged direct from previous month.
4. Write three sentences I can put in a report on whether engaged direct
traffic is growing.
Say which patterns you think are bots and why, and mark anything you can't
tell from the data rather than guessing. If you couldn't check something,
say so.
3. Read what new customers say
Set this up as a routine that runs on the first working day of each month.
Our lead form and post-purchase survey ask "How did you hear about us?" and
include "An AI assistant (such as ChatGPT or Gemini)" as an option, with a
free-text box.
Attached are last month's survey export and last month's AI assistant
sessions from analytics.
1. Count responses by answer, and give AI assistants as a share of all
responses.
2. From the free text, list every assistant people name, with a count.
3. Give me a table with the columns: Month | Responses | Said AI assistant |
Share | Assistants named | AI referral sessions in analytics.
4. Compare with previous months if I attach them, and say in two sentences
whether self-reported AI discovery is rising, falling or flat.
Quote any free-text answers that mention a specific question someone asked
an AI. Don't infer an AI source where the respondent didn't give one, and
if you couldn't check something, say so. Read the episode transcript
James: If someone in finance has asked you what AI search is worth to the business, three studies published this week give you an awkward answer. AI assistants send a sliver of the traffic Google does. I'm James Vickers, this is Manual Focus, and Hannah's here to help me work out what that means for the budget.
Hannah: Hi James. Awkward rather than fatal, because all three measure clicks, and clicks aren't the whole story.
James: Give me the clearest one.
Hannah: Marie Haynes published a piece on the twenty-first of October called "GEO is a distraction". GEO means generative engine optimisation, getting your brand into AI answers. She looked at a year of Google Analytics data across a hundred websites. ChatGPT sent nought point six two per cent as many sessions as Google organic search. Add every other AI assistant and it only rises to nought point seven three.
James: So well under one per cent.
Hannah: Well under. The second study is from The SEO Works, an agency in Sheffield. Ninety-five sites, September last year to August this year. About seventy-two thousand sessions from AI assistants, against about two point eight million from organic search. So organic sent thirty-nine times more traffic.
James: And the third?
Hannah: Kevin Indig, in his Growth Memo newsletter on the nineteenth. He asked whether AI visibility shows up later as direct traffic, or as people searching for your brand by name. Across twelve sites, direct sessions with any identifiable trace of AI were nought point nought nought three per cent of direct traffic. And across six hundred and thirty-two tests on four brands, he couldn't find a reliable link.
James: The argument for AI traffic has always been that it converts better. Does that hold?
Hannah: Against organic search, mostly yes. Haynes found ChatGPT visitors converting at thirteen point eight per cent, against about three per cent for Google organic. But only fifty-five of her sites tracked conversions. In The SEO Works' bigger sample, AI converted at about one point two times the organic rate. It was worse than organic on twenty-three of the sixty-three sites they compared, and at about two thirds the rate of paid search.
James: So a better rate on a tiny slice.
Hannah: Still a small number of customers. And it splits by business type. The SEO Works put consumer sites at one and a half times the organic rate, and business-to-business product sites at about six tenths.
James: When we covered Adobe's holiday forecast last month, a hundred and thirty per cent growth in AI referrals to US retailers, we said that could still be a small slice. Is this the slice?
Hannah: It's the first decent look at its size. Adobe gave growth rates, these give shares, and the shares are low single digits. The trends disagree, though. Haynes says ChatGPT referrals stopped growing around August last year, when GPT five came out. The SEO Works saw AI rise from about two point three to just over three per cent of organic since May.
James: So is Marie Haynes right? Is GEO a distraction?
Hannah: She says stop chasing ChatGPT citations and work on Google Search and Google's own AI features. Her data backs the first half. The second has a catch, which The SEO Works spells out. Clicks from Google's AI Overviews turn up in your analytics as ordinary organic sessions. So some of what these studies call Google traffic already came through an AI answer.
James: And at the start of the month we wrote about AI Overviews showing on eighty-two per cent of tracked brand searches.
Hannah: Right, so Google versus AI isn't a clean split. And none of the three can see an answer that names your brand and shapes a shortlist with no click at all. That's the citation moving off the page, which we've argued since June.
James: Let me be straight with listeners here. We sell AI visibility measurement, through Light Meter, so weigh this with that in mind. Our view is that these studies settle the referral question. For most brands AI assistants are a small traffic source.
Hannah: But not the influence question.
James: No. So size the work to a modest share of the SEO budget, and judge it on how often buyers see you in AI answers, what new customers say when you ask how they found you, and referral sessions third.
Hannah: Indig suggests something close, one monthly report, though his third signal is commercial results, qualified leads and revenue. He also warns about direct traffic. From January to September, total direct sessions in his sample rose about a hundred and forty per cent, while engaged direct sessions fell one per cent.
James: Bots.
Hannah: Mostly. One site logged four hundred and nine thousand direct sessions, and a hundred and ninety-eight were engaged. So check any direct traffic spike before calling it proof that AI is working.
James: So, Get AI on this. First, work out your own ratio. Hand AI twelve months of your session data by source.
Hannah: Give it a Google Analytics export of sessions and key events by source. It groups the AI assistants the way The SEO Works did, and hands back AI traffic as a share of organic, with conversion rates side by side.
James: Second, clean your direct traffic before anyone presents it.
Hannah: Give it direct sessions split by landing page, browser and engagement. It flags the spikes of unengaged visits and gives you an engaged-only line to report instead.
James: And third, read what new customers tell you.
Hannah: Once "an AI assistant" is an option in the how-did-you-hear question on your forms, give it the export each month. It tallies the answers and the assistants people name, beside your referral sessions. Set it as a monthly routine, though adding the question is a person's job.
James: The prompts are in the show notes.
Hannah: As for what we don't know. Haynes calls hers a small dataset. The SEO Works doesn't give sectors or countries and doesn't split results by assistant. Indig's sample leans towards business-to-business and American sites, and his visibility tests had only seventeen and thirty-five usable weeks of data. Nobody can yet separate clicks from AI Overviews or AI Mode in standard analytics.
James: So the takeaway. AI assistants are a small traffic source today, so budget for them like one, and measure whether buyers see you rather than counting the clicks they send. The full article, with all three studies linked, is on manual-focus dot co dot uk.
Hannah: Along with the prompts.
James: If this was useful, follow us in Apple Podcasts, Spotify or wherever you get your shows, and the next one will turn up by itself. And if you'd like help putting a number on AI search that finance will accept, book a free working session at manual-focus dot co dot uk. Thanks, Hannah.
Hannah: Thanks, James. Go and check those direct sessions.