ChatGPT still sends the overwhelming majority of identifiable AI referral traffic in a new analytics panel spanning 5,777 GA4 properties, but the distribution is materially less concentrated than another major dataset published this week. Trakkr’s September 22 snapshot puts ChatGPT at 86.9% of identified AI referral pageviews over the trailing 30 days, while Claude has climbed to 8.9% after growing 177.4% from the preceding period.
The figures come from Trakkr’s AI Search Traffic Index, a changing panel of connected Google Analytics 4 properties. Across that panel, the overall AI referral index increased 46.1% over the latest 30-day period. Gemini accounted for 2.9% of identified AI referrals, while Perplexity and Microsoft Copilot each represented 0.5%. You.com followed at 0.3%, with the remaining tracked sources rounding to near zero in the published source mix.
The result does not mean ChatGPT has 86.9% of AI usage, AI citations or global AI search. Trakkr explicitly defines the metric as the share of identifiable AI referral pageviews detected through GA4 session-source data. It is a measurement of observable traffic leaving AI products and arriving at websites in the panel.
Claude now represents almost one identifiable AI referral in eleven
The most important movement is not ChatGPT’s continued lead but Claude’s acceleration. Claude’s 8.9% share is up 4.2 percentage points from the prior period, while its measured referral traffic increased 177.4%. In practical terms, roughly one in every eleven identifiable AI referral pageviews in Trakkr’s current 30-day source mix came from Claude.
ChatGPT still dwarfs every competitor individually. Its 86.9% share is nearly ten times Claude’s and roughly 30 times Gemini’s. ChatGPT referral volume also grew 40.6% over the period, even as its share declined by 3.4 percentage points because other sources—most notably Claude—expanded faster.
That distinction matters. A declining share does not necessarily mean declining traffic. ChatGPT is sending more measured referral pageviews in Trakkr’s panel while accounting for a smaller percentage of a rapidly expanding category.
The entire identifiable AI referral category grew 46.1%
Trakkr reports that its normalized AI referral index rose 46.1% over the latest 30-day period compared with the preceding 30 days. The index tracks aggregated screen pageviews from sessions attributed to supported AI sources across qualifying GA4 properties.
Growth is uneven across industries. Technology and Finance recorded the largest 30-day increase in the current dashboard at 106.4%, followed by Entertainment at 48.5%, Education and Public at 39.7%, Commerce at 34.4%, Other at 33.3% and Lifestyle and Health at 14.5%. The number of contributing properties also varies sharply by category, with Commerce representing 2,686 properties in the current sector table while Entertainment contains 170.
Those differences are a reminder that a global panel average is not a substitute for first-party analytics. A publisher, SaaS company or retailer can have a radically different AI source mix depending on audience, content type and which assistants are most likely to cite or link to its pages.
Why Trakkr says 86.9% while BrightEdge says 95.1%
The new figure arrives immediately after BrightEdge reported a much higher concentration. NetContentSEO covered that dataset in our analysis of ChatGPT’s 95.1% share of AI referrals in BrightEdge’s measurements. At first glance, 86.9% and 95.1% appear to disagree substantially.
They should not be treated as competing estimates of one universal market share. Both are proprietary measurement environments with different site populations, collection systems, source classifications and time windows. Trakkr explicitly warns that its connected-property panel changes over time and is not representative of the entire web. BrightEdge’s dataset likewise reflects the sites and measurements available to its platform rather than a census of all internet traffic.
The Claude difference demonstrates why methodology matters. In Trakkr’s current panel, Claude accounts for 8.9% of identifiable AI referrals by itself. That is enough to push ChatGPT’s share below 90%, even though ChatGPT remains overwhelmingly dominant. A panel containing a different mix of industries or sites can produce a much smaller Claude share and therefore a much larger ChatGPT share.
The useful conclusion is not that one provider has found the “correct” number. It is that multiple large datasets currently agree on the direction—ChatGPT dominates measurable AI referrals—while disagreeing materially on the degree of concentration.
Trakkr is measuring pageviews with visible source attribution
The denominator is especially important. Trakkr detects supported AI sources from GA4 sessionSource values and aggregates screen pageviews on those attributed sessions. Source mix is each platform’s percentage of identified AI referral pageviews over the trailing 30 days.
That means the index cannot see every visit influenced by an AI assistant. Trakkr explicitly excludes traffic whose referrer is unavailable from the identifiable referral measure. A user can research a product in an AI application, later navigate directly to the brand, or arrive through another channel without preserving an AI source signal. That influence will not necessarily be credited to the assistant.
This blind spot is not unique to Trakkr. NetContentSEO previously examined how native AI applications can strip referral information and disappear from conventional referral measurements. Referral analytics is therefore best understood as observable outbound traffic, not total AI-assisted discovery.
Google AI Overviews and AI Mode are largely outside this source mix
Another important limitation concerns Google. Trakkr’s methodology notes that the index is based on identifiable AI referral sources. Visits generated through Google AI Overviews or AI Mode can arrive in analytics as ordinary Google organic traffic rather than a separate Gemini or AI-search source.
That makes the 2.9% Gemini figure easy to misread. It should not be interpreted as the total share of web traffic influenced by Google’s generative search experiences. It represents identifiable referral pageviews attributed to the Gemini source within Trakkr’s classification.
This distinction becomes more important as Google embeds AI deeper inside Search. A publisher may lose or gain clicks because of AI Overviews or AI Mode while those visits remain classified inside the familiar google / organic channel. AI influence can therefore move without the AI-referral dashboard capturing the movement directly.
Claude’s growth may matter more than its absolute share
An 8.9% share is still small beside ChatGPT, but the 177.4% growth rate makes Claude strategically harder to ignore. If that acceleration persists—which the current data cannot guarantee—the AI referral market could become less concentrated even while ChatGPT continues growing in absolute terms.
The qualification is crucial because 30-day growth rates can be volatile. A smaller platform can produce a large percentage increase from a lower baseline, and Trakkr’s panel composition itself changes over time. The 177.4% figure is a measured movement inside the current panel, not a forecast that Claude will continue expanding at that rate.
Still, the data provides a useful counterweight to dashboards that make AI referral traffic look effectively synonymous with ChatGPT. In Trakkr’s current environment, Claude is already large enough to represent a meaningful secondary acquisition channel.
Perplexity’s visibility does not translate into a large referral share here
Perplexity accounts for only 0.5% of Trakkr’s current identified AI referral pageviews and declined 3.2% over the measured 30-day comparison. Microsoft Copilot also holds 0.5%, although its measured traffic grew 6.7%.
Those numbers reinforce a point NetContentSEO explored in our analysis of platform-weighted GEO metrics: engines can differ dramatically depending on whether the KPI is citations, audience exposure, usage or outbound traffic. A platform that is highly visible in AI-search discussions does not automatically send a proportionate number of measurable website visits.
For marketers, that makes equal weighting increasingly difficult to justify when the business objective is traffic. A dashboard that assigns ChatGPT, Claude, Gemini and Perplexity 25% each may be mathematically tidy while bearing little resemblance to the acquisition mix visible in the company’s analytics.
First-party GA4 should outrank any industry benchmark
Trakkr’s panel is valuable because it gives marketers a broad directional reference across thousands of properties, but the company’s methodology is unusually explicit about its limits. The panel contains only connected properties that produced qualifying AI referral data, changes over time, publishes no property identities or absolute public pageview totals, and does not claim representativeness for the whole web.
That means the most actionable benchmark remains a company’s own analytics. If Claude already represents 20% of a site’s AI referrals, a global 8.9% figure should not cause the team to de-prioritize it. If Perplexity produces qualified leads despite tiny overall referral volume, its business value may exceed its share of pageviews.
The external benchmark is most useful for detecting whether a first-party pattern is unusual and for watching directional movement. It can tell teams that Claude referral growth is not necessarily isolated to their own site, while still leaving their own conversion and revenue data as the final decision layer.
AI referral measurement is becoming a panel-comparison problem
The emergence of competing referral datasets is a sign of a maturing category. BrightEdge, Trakkr, StatCounter, Cloudflare and other measurement providers observe different slices of AI-to-web behavior. Their numbers will not converge automatically because they are not observing identical populations or using identical attribution rules.
This is familiar territory for web analytics. Search market share, browser share and advertising benchmarks have always varied by measurement methodology. AI referrals add another layer of difficulty because native applications, stripped referrers, generated answers and Google’s integration of AI into ordinary search can hide parts of the journey from conventional attribution.
The correct response is not to choose whichever percentage produces the strongest headline. It is to preserve the denominator. Trakkr’s 86.9% means 86.9% of identified AI referral pageviews in its current qualifying GA4 panel over the latest 30-day window. BrightEdge’s 95.1% belongs to BrightEdge’s own observed environment. Both can be informative without either becoming a universal market-share statistic.
ChatGPT dominates, but the second tier is becoming measurable
The most important update from Trakkr is therefore not that ChatGPT has suddenly become less important. Its referral volume grew 40.6%, and it still accounts for nearly seven out of every eight identifiable AI referral pageviews in the panel. It remains the dominant source by an enormous margin.
The change is that the rest of the market is becoming easier to see. Claude’s 8.9% share and 177.4% growth create a meaningful second source, while Gemini remains a smaller but visible contributor. At the same time, total measured AI referral activity across the panel is rising quickly.
For GEO teams, that argues for a measurement strategy with two ideas held at once: ChatGPT deserves disproportionate attention when the KPI is identifiable referral traffic, but the distribution is not fixed and no single vendor’s panel should be mistaken for the internet. The better benchmark is a range of external sensors checked against first-party analytics.
ChatGPT still dominates the traffic that AI assistants visibly send back to websites. Trakkr’s latest data simply shows that dominance is not absolute—and Claude is now large enough, and growing fast enough, to make the space behind ChatGPT strategically relevant.