- ChatGPT generated 92.4% of all trackable referral sessions from standalone AI assistants in a study covering 6.77 million visits across 166 websites.
- ChatGPT sent 28.8% of its referrals to internal search pages, including 34.6% for SaaS sites, creating avoidable friction for visitors who may already be close to a decision.
ChatGPT sends 92.4% of all trackable referral traffic from standalone AI assistants.
And now we know the #1 way to increase your ChatGPT traffic regardless of business niche.
A new study analyzed 6.77 million AI-driven visits and found one platform is currently responsible for… https://t.co/niKRjfMch4 pic.twitter.com/nDaaR8oTmK
— Alex Groberman (@alexgroberman) July 12, 2026
Among trackable referrals from standalone AI assistants, ChatGPT is not simply the market leader. It accounts for almost the entire measured category in Previsible’s dataset.
For its latest State of AI Discovery Report, Previsible analyzed 6.77 million referral sessions across 166 Google Analytics properties between November 2024 and May 2026.
Monthly sessions from standalone AI assistants increased from 65,249 to 644,478 over that period, representing 9.9x growth. ChatGPT accounted for 92.4% of all trackable sessions in the dataset.
The study only covers referrals from standalone platforms such as ChatGPT, Claude, Gemini, Perplexity and Copilot. It excludes AI discovery taking place directly inside Google Search through AI Overviews and AI Mode, which cannot be isolated in referral data in the same way.
ChatGPT Is Pulling Away From Every Other AI Platform
ChatGPT referral sessions increased from 47,606 in November 2024 to 610,910 in May 2026. That is 12.8x growth in 19 months and a level of referral volume no competing standalone assistant came close to matching.
The more interesting movement happened below it.
Gemini grew from 5,598 to 18,119 monthly sessions and finished the period as the second-largest source. Claude increased from just 133 sessions to 8,528 and overtook Perplexity in March 2026.
Perplexity moved in the opposite direction. After reaching 17,507 monthly sessions in March 2025, it fell to 6,788 by May 2026. Copilot dropped from a peak of 8,651 sessions in August 2025 to only 339.
This does not mean marketers should ignore every platform except ChatGPT. Claude may matter disproportionately for companies targeting developers, technical buyers and professional services, while Gemini’s integration across Google’s products makes its wider influence difficult to measure through referral sessions alone.
But for businesses deciding where to focus limited AI search resources today, the traffic data leaves little room for debate. ChatGPT comes first.
The Traffic Is Growing, but It Is Also Volatile
Previsible’s data also shows how quickly AI referral traffic can change.
Total monthly sessions fell by 50% in November 2025, driven largely by ChatGPT referrals dropping from 448,412 in October to 213,345 one month later. Traffic recovered to 442,609 in December.
Previsible said the drop was likely connected to a model or product change, although it could not confirm a single cause without more information from OpenAI.
That distinction matters. AI referral traffic is controlled by platforms whose recommendation systems, browsing capabilities and linking behavior can change without the visibility marketers are accustomed to receiving from traditional search engines.
A brand can therefore improve its AI visibility and still experience a sudden traffic decline caused by a product decision outside its control. Measuring several platforms and avoiding dependence on one landing-page type will become increasingly important as this channel grows.
ChatGPT Is Sending High-Intent Visitors to Internal Search Pages
The study’s most useful finding is not ChatGPT’s market share. It is where the traffic lands.
ChatGPT directed 28.8% of its referrals to internal site search pages. Across SaaS websites, the figure reached 34.6%.
Previsible interprets this pattern as domain-level confidence combined with page-level uncertainty: ChatGPT identifies a relevant website but does not always select the most appropriate individual URL. It sends the visitor to the site’s search function and leaves the final navigation step to them.
That turns an often-neglected utility page into an acquisition landing page.
Many internal search experiences were never designed for that role. They return loosely related results, hide the most commercially relevant pages or offer no clear next step. Some are barely usable on mobile.
For a visitor arriving from an AI recommendation, that is a particularly expensive failure.
The user has already explained what they need, reviewed an AI-generated comparison and chosen to visit the recommended website. Sending that person into a poor internal search experience adds friction at the point where the site should be helping them make a decision.
AI Visitors Can Be Worth More Than Their Traffic Share Suggests
The value of these sessions becomes clearer when referral volume is compared with conversion data.
NP Digital analyzed data from 412 companies and found that AI platforms generated 6.2% of total traffic but accounted for 11.4% of sales. The figures suggest that AI referrals contribute a substantially larger share of revenue than their traffic volume alone would imply.
Semrush reached a similar conclusion, estimating that the average visitor from a non-Google AI platform was worth 4.4x as much as the average traditional organic visitor based on conversion rate.
The exact multiplier will vary by industry, website and conversion model. But both datasets point in the same direction: visitors referred by AI systems tend to arrive further into the decision process.
That makes the 28.8% of ChatGPT traffic landing on internal search pages more than a technical curiosity. It is a conversion problem.
AI Referral Intent Looks Different in Every Industry
The Previsible study also found that AI assistants do not send every industry to the same type of page.
For e-commerce sites, product pages were the primary landing destination. These visitors often arrive after asking an AI system to compare products, features or prices, making clear and structured product information critical.
Education traffic concentrated on course pages, which received 52% of referrals. Users frequently bypassed general marketing content and moved directly to the program or course matching their query.
Health visitors behaved differently. About pages received 42.1% of AI referrals, suggesting that people were evaluating the credibility and identity of the organization after receiving health-related information.
Legal traffic was more evenly distributed across blog, About, contact and location pages. That reflects a longer evaluation process involving expertise, service availability and geographic relevance.
Publishers received 54% of their AI referrals on news pages, but referral penetration remained extremely low compared with their overall organic traffic. Publishers may supply much of the source material used in AI answers while receiving only a small fraction of the resulting visits.
The practical lesson is that there is no universal AI landing-page template. The right entry point depends on what users need to verify or do next within a particular industry.
Measuring AI Visibility Is Finally Becoming Easier
Until recently, much of this traffic was mixed into Referral, Direct or other channel groupings, making it difficult to isolate.
In May 2026, Google Analytics added an AI Assistant channel to its default reporting. Recognized referrals receive the medium value ai-assistant and appear separately from standard referral traffic.
GA4 shows what happens after someone clicks, including landing pages, engagement, conversions and revenue. It does not show how often a brand was mentioned inside an AI response without receiving a visit.
Google also introduced dedicated generative AI performance reports in Search Console in June 2026. These provide separate visibility data for appearances within Google’s generative Search and Discover features, including AI Overviews and AI Mode.
Microsoft moved earlier. Its AI Performance report in Bing Webmaster Tools shows citations, cited URLs, visibility trends and samples of the grounding queries used to retrieve a site’s content across Copilot, Bing and supported partner experiences.
Third-party platforms can add broader monitoring across multiple assistants, but these native reports provide a much stronger starting point than marketers had a few months ago.
The reporting landscape is still fragmented, as we covered in our analysis of Google’s increasingly complex approach to AI-readiness signals. Referral traffic, citations and appearances inside Google AI features remain separate layers that need to be measured together.
What Marketers Should Do Now
Start with the pages already receiving AI traffic. Segment GA4’s AI Assistant channel by landing page, page type, conversions and revenue. A site-wide AI traffic percentage can hide major differences between product, pricing, course and internal search pages.
Test the internal search experience as if it were a campaign landing page. Check the queries visitors are likely to use, the relevance of the results, mobile usability and whether high-value pages are easy to reach.
Strengthen navigation and internal linking. Clear category structures, contextual links and descriptive page titles help users reach the right content and may also reduce the likelihood that AI systems fall back to a generic search page.
Make decision-stage information easy to extract. Product specifications, pricing, eligibility requirements, locations, credentials and comparisons should be available in clear text rather than hidden inside images, PDFs or vague sales copy.
Measure visibility and business results separately. Citations and mentions show whether a brand is entering AI-generated answers. Referral sessions and conversions show whether that visibility produces commercial value. Neither metric tells the complete story by itself.
The Traffic Has Arrived Before Most Sites Were Ready for It
The headline figure is ChatGPT’s 92.4% share of trackable standalone AI referrals. The more consequential number may be the 28.8% of its traffic being sent to internal search pages.
AI assistants are already choosing brands, recommending websites and delivering visitors who appear to be closer to a decision than the average organic search user.
The problem is no longer limited to whether a brand can appear in an AI response. It is whether the website is ready for the handoff when it does.
