If you’re trying to figure out where to put your AI search optimization effort in 2026, there’s a new dataset worth paying attention to. Previsible, an SEO analytics firm, just published its third AI traffic study, and it’s the largest one yet: 6.77 million LLM-driven sessions pulled from 166 GA4 properties across ten industries, tracked from November 2024 through May 2026. Written by David Bell, the report is titled the 2026 State of AI Discovery Report, and it paints a fairly clear picture: this market may not be as fragmented as a lot of us assumed.
Here’s what stood out to me.
ChatGPT isn’t just winning, it’s pulling further ahead
The main number is worth sitting with. ChatGPT now accounts for 92.4% of all trackable LLM referral traffic, and that share is still growing. Sessions went from about 47,600 in November 2024 to over 610,000 by May 2026, a 12.8x increase. For context, the report’s own prior estimate put ChatGPT at around 84% share back in December 2025, so this is a genuine acceleration, not a plateau.
Worth noting: this study only measures referral traffic from standalone LLM platforms. It deliberately excludes AI Overviews inside Google Search, which the report suggests likely represents even more AI-driven traffic on its own. So “ChatGPT dominance” here is specific to the standalone chat-app frame, not the whole AI search picture.
Claude is the one to watch, even at a small base
Claude’s numbers are still tiny in absolute terms (8,528 sessions in May 2026 versus ChatGPT’s 610,000+), but the growth curve is the interesting part. Claude grew 64x over the study period and passed Perplexity in monthly sessions for the first time in March 2026, then stayed ahead. Much of that came from a sharp acceleration in early 2026: sessions roughly doubled month over month from January through March.
The report ties this to Claude’s expansion into coding and professional workflows rather than general search habits. If your audience skews toward developers or technical buyers, this is probably the platform shift worth tracking closely, even though it’s still a small slice of overall traffic.
Gemini is the quiet, steady one
Gemini grew 3.2x over the same period, from 5,598 to 18,119 monthly sessions, without much volatility along the way. The report notes that Gemini’s integration into Google Workspace and Android likely means its real discovery influence goes beyond what referral tracking alone can measure.
Perplexity and Copilot are both fading
Both platforms peaked and then declined significantly. Perplexity peaked in March 2025 and has since dropped 61%. Copilot’s decline is steeper: an 8,651-session peak in August 2025 fell to just 339 sessions by May 2026, a 96% drop. The report frames Copilot’s earlier growth as more of a short-lived spike than a durable trend, and points to Microsoft’s shifting pricing model as one possible factor.
Where the traffic actually lands matters more than the platform
One of the more useful findings has less to do with market share and more to do with behavior. The report splits platforms into two personality types:
- Search-pattern models (ChatGPT, Gemini) trust a domain but often can’t pinpoint the exact page, so they route users to internal search results or topical hub pages.
- Content-selection models (Perplexity, Claude) pick specific pages, with a notable lean toward long-form and educational content.
What that seems to mean is your internal site search experience matters about as much as your content itself, since a meaningful share of AI-referred visitors are landing there first rather than on the page you’d want them to see.
Industry differences are wide
AI traffic penetration isn’t even across sectors. E-commerce grew 37x off a near-zero base, and product pages now capture 43% of e-commerce LLM traffic, arriving with purchase intent already formed. Insurance saw 18.9x growth to the highest penetration rate of any vertical (1.51%). Health was the only sector where AI traffic penetration actually declined, and About pages served as the top landing spot there, suggesting users are checking source credibility before trusting a health answer.
A few notes from the report’s recommendations
A handful of these are worth sitting with, paraphrased here:
- Prioritize ChatGPT visibility first, given its overwhelming share of standalone LLM traffic.
- Keep an eye on Claude even while its absolute numbers stay small, especially for technical or professional audiences.
- Treat product pages as a genuine AI discoverability requirement, not just a conversion optimization detail.
- Make pricing information machine-readable rather than gated behind a “contact us” form.
- Invest in your internal search experience, since AI-referred users land there roughly a quarter of the time.
- Measure AI traffic by page type instead of relying on a single site-wide average, since it can hide where the real concentration is happening.
Worth reading in full
This summary only scratches the surface. The full report includes a detailed breakdown by industry and page type, plus methodology notes on how sessions and penetration rates were calculated. If any of this is relevant to your traffic strategy, it’s worth reading directly: previsible.com/seo-strategy/ai-traffic-report-july-2026.
All data and findings referenced above come from Previsible’s 2026 State of AI Discovery Report by David Bell.