AI referral traffic reports disagree. Measure your own.
Two 2026 studies put ChatGPT's share of AI referral traffic 30 points apart. Here is why they disagree and how to measure the split for your own site.
Two well-run studies measured AI referral traffic in 2026 and landed about 30 points apart. Goodie put ChatGPT at 63% of AI referrals. Previsible put it at 92%. Same metric, same year, opposite headline. The share you quote depends on whose panel you read, so the number that should drive your plan is your own.
What the two studies actually found
Goodie's report compared two waves of GA4 referrer data from an anonymized brand panel, triangulated against SimilarWeb and SensorTower. Between May to August 2025 and March to April 2026, ChatGPT's share of AI referrals fell from 89.1% to 62.6%. Claude climbed from 1.4% to 18.5%. Gemini went from 2.4% to 10.6%, Perplexity from 3.1% to 7.3%, Copilot from 3.2% to 4.0%.
Previsible's third AI Traffic Study, published July 6, read 166 GA4 properties across nine industries: 6.77 million assistant-driven sessions from November 2024 to May 2026. Its number for ChatGPT was 92.4% of AI referral traffic as of May. Claude overtook Perplexity in March 2026, and Copilot collapsed 96% from its 2025 peak.
Notice what agrees and what doesn't. Both studies see the same trend lines: ChatGPT still first, Claude rising fast, Copilot falling off. The direction rhymes. The level does not. One says ChatGPT owns roughly six in ten AI referrals, the other says more than nine in ten.
Why AI referral traffic numbers disagree
Most of the gap comes from one quiet choice: how you average.
Goodie weights each brand equally. A tiny site and a huge site count the same, so smaller tools that a few brands lean on show up larger in the blended share. Previsible weights by sessions. Big properties dominate the total, so whatever their audiences use sets the headline. Brand-average lifts the challengers. Session-weight concentrates on the leader. That single decision can move ChatGPT 30 points without anyone measuring anything wrong.
Two more factors widen it. Panel composition matters: a SaaS-heavy sample behaves differently from one packed with publishing or ecommerce, and each study drew a different mix. So does attribution. Many assistants send people through in-app browsers or redirects that strip the referrer, so a chunk of real AI traffic lands in GA4 as direct, and how each study recovered that traffic changes the count.
What a 30-point gap means for your plan
It means you can't set an AI strategy off a chart you didn't build. Trust the 92% figure and you pour everything into ChatGPT and treat Claude and Gemini as rounding error. Trust the 63% figure and you spread effort across four assistants. Those are different quarters of work, chosen by which blog post you happened to read.
The honest read is that AI referral traffic behaves like one channel for most sites, with ChatGPT far in front, while the challengers are real and growing fast enough that ignoring them carries risk. Where your own site sits on that spread is a question only your data answers.
Measure your own AI referral traffic in GA4
You can see your split in about ten minutes. Open Reports, then Acquisition, then Traffic acquisition, and add Session source / medium as the dimension. Then build one regex filter that groups the assistants:
chatgpt.com|chat.openai.com|openai.com|claude.ai|gemini.google.com|perplexity.ai|copilot.microsoft.com
That gives you a clean count of the sessions each assistant sent, and their shares for your site rather than a panel's. Two cautions. Traffic from mobile apps often arrives as direct, so your measured AI share is a floor, not a ceiling. A short "how did you find us" field on your contact form helps recover the rest. And read last-non-direct attribution with care, since a first ChatGPT visit can convert weeks later under a different source. The point is to measure the clicks yourself instead of borrowing someone else's average.
Do this once a month and you get something no study can hand you: the actual mix of assistants sending you revenue, tracked over time. When the next report claims a new headline number, you check it against your own split and move on.
If you want that AI referral view built into your GA4 and read every month, that's the kind of measurement work we do at vibhe.
