If you’ve been hedging your AI search strategy across ChatGPT, Perplexity, Gemini, and Copilot, new data suggests it’s time to stop hedging.
A fresh AI traffic study from Previsible, tracking 6.77 million LLM-driven sessions across 166 websites over 19 months, shows the AI referral landscape has consolidated hard around a single platform. ChatGPT now accounts for 92.4% of all trackable LLM referral traffic, up from roughly 84% just six months earlier. Everyone else is fighting over single-digit scraps.
For growth-stage founders trying to figure out where to put limited content resources, this data point matters more than most of the “AI is changing search” think pieces out there. Here’s the breakdown, and what it should change about your roadmap.
Growth Didn’t Plateau, It Paused
Total monthly LLM-referred sessions across the tracked sites grew nearly 10x over the study period, climbing from 65,000 in November 2024 to over 644,000 by May 2026.
That growth wasn’t a straight line, though. Sessions nearly halved in a single month last November, almost entirely because ChatGPT referrals dropped sharply, likely tied to a model update that temporarily favored other sources over site content. Traffic recovered within a month and kept climbing from there.
The takeaway for anyone building a content strategy around AI visibility: a single vendor’s product decision can cut your AI-referred traffic in half overnight, with no warning and no appeal process. If ChatGPT is your primary AI channel (and for most sites, it now is), build for volatility rather than a smooth upward line.
Where the Other Platforms Landed
The real story isn’t at the top. It’s in who’s moving underneath ChatGPT.
- Claude grew roughly 64x over the study window and overtook Perplexity in referral volume in March 2026, a lead it has held since. Growth was flat through most of 2025 before accelerating sharply, which lines up with the platform’s push into agentic and enterprise tools. For sites targeting technical buyers or professional-services audiences, this is a signal worth acting on early rather than after the trend is obvious.
- Gemini has grown steadily without much volatility, though its deep integration into Google Workspace and Android likely means its true discovery footprint is undercounted by referral data alone.
- Perplexity peaked in early 2025 and has since fallen by more than half.
- Copilot dropped even further, down roughly 96% from its 2025 peak.
Neither of the last two looks like a traffic-acquisition bet anymore. Both appear to be optimizing for keeping users inside their own product experiences rather than sending them elsewhere.
One caveat worth flagging: this data covers standalone LLM platforms only. AI-driven discovery inside Google’s own results, including AI Overviews and AI Mode, almost certainly generates more AI-influenced traffic than every standalone chatbot combined. It’s just measured differently, so it isn’t in these numbers.
The Real Finding: AI Sends Users to Your Search Bar, Not Your Content
The most useful part of this data isn’t the market-share breakdown. It’s where AI platforms are actually sending visitors once they arrive.
Roughly a quarter of all AI-referred traffic across industries lands on internal site search pages, not on the specific piece of content that answered the user’s question. ChatGPT alone sends close to 29% of its referred traffic straight to internal search.
The interpretation: the model trusts your domain enough to send someone there, but isn’t confident enough to pick the exact right page, so it hands the decision back to the visitor via your search box. That pattern held steady across industries and the full 19-month window, which suggests it’s a structural feature of how these models retrieve information, not a temporary glitch.
Practically, that means your internal site search is no longer just a navigation feature. It’s become an acquisition surface in its own right, often for your highest-intent visitors. If it’s clunky, poorly indexed, or an afterthought in your site build, that’s now a conversion leak.
The landing pattern also varies meaningfully by industry:
- SaaS traffic disproportionately lands on search results pages
- Ecommerce traffic tends to land on product pages, with purchase intent already formed
- Publisher traffic mostly lands on news pages, though publishers capture only a sliver of the organic traffic volume relative to how often their content gets cited, meaning they do the work and see little of the payoff
- Education traffic lands directly on course pages, skipping marketing content entirely
- Health-related traffic lands heavily on About pages, suggesting users are evaluating source credibility before engaging with the content itself
There’s also a split in how the platforms behave. ChatGPT and Gemini act like search engines: they trust a domain but hedge on the exact page. Perplexity and Claude behave more like content curators, picking specific pages and leaning toward long-form material. If editorial content is core to your strategy, Perplexity and Claude carry more weight than their traffic share alone suggests.
What This Means for Your Roadmap
A few practical shifts worth making based on this data:
- Prioritize ChatGPT optimization first. It’s not one channel among several anymore. It’s the dominant one, by a wide margin. Expand to other platforms once ChatGPT-driven volume justifies the extra effort.
- Keep an eye on Claude. Its overtaking of Perplexity in March 2026 is recent enough that early positioning still has real value.
- Treat your internal search like an acquisition channel, not a back-office feature: audit it, index it properly, and make sure it surfaces the right content fast.
- Structure product and pricing data so it’s machine-readable. Vague pricing language gives AI systems nothing to summarize or compare, which likely costs visibility in comparison-style queries.
- Measure AI traffic by page type, not just site-wide. A site-wide average can hide the fact that one page, often pricing or a comparison page, is getting several times the AI traffic of the rest of the site combined.
The open question the data doesn’t yet answer is which platforms send visitors who actually convert versus visitors who bounce. That’s the next frontier for anyone serious about AI-driven acquisition. Traffic volume alone was never the whole picture, and it’s becoming less so by the month.
Data referenced in this piece comes from Previsible’s third AI Traffic Study (6.77 million sessions, 166 GA4 properties, November 2024–May 2026), as reported by Search Engine Land.
