Marie Haynes published a GA4 analysis on October 21 that puts ChatGPT at 0.62% of Google organic sessions, or 0.73% with all other LLMs included, across 12 months of GA4 data from 100 websites. On the 55 sites tracking conversions, she puts ChatGPT’s AI traffic conversion rate at 13.8% against 3.02% for Google organic, while Google organic drove 27x more conversions than all LLMs combined. Her conclusion, in her words: “chasing ChatGPT citations and LLM mentions is optimizing for the wrong target.”
How much traffic does ChatGPT send compared with Google organic?
In Haynes’ panel, ChatGPT sessions equal 0.62% of Google organic sessions, and all LLMs together equal 0.73%. That is a share of one channel, not of a site’s total traffic. A second GA4 benchmark, from The SEO Works, finds AI assistants sending 2.6% of what organic search sent across 95 sites, a different sample and definition.
In her charts, the “145.8M sessions” in the header matches the sum of 144,788,763 Google organic sessions and 1,062,833 LLM sessions. Direct, paid, social and email are not in that total (our arithmetic). The “964 out of 1,000” graphic in the conversion chart is a per-mille dot illustration, not a count of conversions. She also reports that 45 of the 100 sites she sampled were not tracking any key events.
Where the two GA4 datasets split, and where they land close
| Item | Haynes (Oct 21) | The SEO Works (updated Oct 14) |
|---|---|---|
| Sample | 100 GA4 properties, 12 months (dates not stated) | 95 sites, Sep 2025 to Aug 2026 |
| Denominator | Google organic sessions | Organic search sessions |
| AI share of organic | ChatGPT 0.62%; all LLMs 0.73% | All AI assistants 2.6% |
| Conversion-rate multiple | About 4.6x (ChatGPT, 13.8% vs 3.02%); about 4.3x for all LLMs (her chart) | 1.18x pooled (7.18% vs 6.08%), 1.21x median |
| AI conversions per 100 from organic | 3.68 (112,640 vs 3,061,316 key events) | 3.2 (5,001 vs 155,605) |
| How AI sessions were identified | Not stated | Session source regex, published |
By our arithmetic, the traffic share differs by about 3.5x (0.73% vs 2.6%) and the conversion-rate multiple by about 3.6x (all LLMs: 4.3x vs 1.18x). The two gaps roughly offset, and the published AI conversions per 100 from organic sit at 3.68 and 3.2. Neither author makes the offset point; it is our reading of their figures. In each study the conversion figures come from a smaller pool than the traffic share: 55 of Haynes’ 100 sites, 63 of SEO Works’ 95.
Why the conversion multiple moves
SEO Works’ own table shows how much the answer depends on the comparison channel: 0.67x against paid search, 1.18x against organic search, 1.37x against total sessions and 2.48x against direct. It states that “Anyone quoting a conversion multiple for AI traffic without stating the comparison channel has published a number that cannot be checked or reproduced.” Haynes compares against Google organic, which is a legitimate choice, but a multiple only means something next to that choice.
Identification is the second variable. SEO Works publishes the match it used: Session source matches partial regex bard|chatgpt|openai|claude|copilot|gemini|perplexity|grok. It says it matched referring domains because GA4’s own AI Assistant category is too recent to cover a 12-month window; our earlier piece covers GA4’s AI Assistant channel in acquisition reports. Haynes’ post does not say how a ChatGPT session was identified.
What neither study can see
GA4’s help page on (direct) / (none) traffic says: “When links to your site lack UTM parameters or your site isn’t integrated with marketing and advertising platforms, traffic source information is lost.” A session count by AI source can only include visits that arrive carrying a source. See Google’s explanation. SEO Works adds that Google AI Overviews clicks “arrive as ordinary organic sessions and cannot be separated”, that in-app browsers and stripped referrers would not be tracked by its regex, and calls its measured AI contribution a floor. It also notes that GA4 “counts key events, not people.” Haynes’ post does not address referrer loss or how key events are counted, so we cannot say how those affect her numbers.
The trend lines measure different things, too. Haynes charts absolute monthly ChatGPT sessions across her 100 properties, from a peak of 105,371 in October 2025 to 71,018 in September 2026, which she describes as growth ending in August 2025 and also as a plateau. SEO Works tracks all-AI sessions as a share of organic: 2.29% from September 2025 to April 2026, and 3.04% from May 2026. They cannot be laid on top of each other. We looked at a similar mismatch before in why two AI studies don’t agree.
A check you can run on your own property
Using SEO Works’ published regex, in GA4 Explore:
- Pick one date range and use it for both segments.
- Build an AI segment with Session source matching partial regex
bard|chatgpt|openai|claude|copilot|gemini|perplexity|grok. - Build a second segment for the channel you will divide by (SEO Works used organic search; Haynes used Google organic).
- Compare sessions and key events for both, and write down the comparison channel next to every multiple you report.
Treat the AI figure as a floor, treat key events as events and not people, and expect AI Overviews clicks to sit inside organic.