How AI referral traffic behaves differently from traditional search., as served to AI agents
# What AI agents get at /research/ai-referral-traffic-vs-search
--- title: How AI referral traffic behaves differently from traditional search. description: A study of what happens after someone arrives from ChatGPT, Perplexity or another AI assistant instead of a traditional search engine. url: https://tanujsharma.xyz/research/ai-referral-traffic-vs-search author: Tanuj Sharma status: Ongoing dataset: External benchmarks + first-party measurement in progress confidence: Directional published: 2026-10-01 modified: 2026-10-15 --- # How AI referral traffic behaves differently from traditional search. A study of what happens after someone arrives from ChatGPT, Perplexity or another AI assistant instead of a traditional search engine. ## 01 Research question Does someone arriving from ChatGPT behave differently from someone arriving from Google? AI assistants increasingly sit earlier in the decision process. A user can spend several minutes asking ChatGPT to understand a problem, compare approaches, shortlist products, eliminate alternatives and clarify technical questions before clicking a source. That means the click reaching a website may represent a different level of intent from a conventional search click. The question is: are AI referrals simply another source of traffic, or are they a smaller but more qualified class of visitor? ## 02 Working finding The evidence so far points to a nuanced answer: AI referral traffic is still relatively small, but it is growing quickly and frequently shows unusually strong engagement or conversion behaviour. Semrush analysed 17 months of clickstream data and reported that outbound referral traffic from ChatGPT grew 206% in 2025, comparing January 2025 with January 2026. [1] Ahrefs’ tracker chatgpt-vs-google.com, which covers 74,752 websites, found that all AI chatbots combined sent about 0.28% of total web traffic in March 2026, while Google sent 28.12%. [3] What is more interesting is what happens after the click. Ahrefs reported that on its own website, AI search visitors made up just 0.5% of visitors but drove 12.1% of sign-ups. [3] This is one company’s first-party result and must not be treated as a universal benchmark. Adobe has also reported stronger engagement from AI-referred retail visitors. By February 2025, visits to US retail sites from AI referrals had a 23% lower bounce rate, 12% more page views and lasted 41% longer than other traffic, though they were still 9% less likely to convert (down from 43% less likely in July 2024). [4] By October 2025 that had flipped: Adobe reported that visitors arriving at US retail sites from generative AI sources were 16% more likely to convert than visitors from non-AI sources such as paid search, email, organic search and social media. [5] > **INTERPRETATION:** AI traffic appears to be sparse but often highly considered. The chatbot may have already performed part of the discovery and comparison process before the user reaches the website. This does not mean every site should expect AI visitors to convert better — industry, query intent, product complexity and the point in the buying journey all matter. ## 03 Methodology AI referral analysis needs more than a standard GA4 channel report. Create a dedicated AI referral group covering sources such as: [2] - chatgpt.com - perplexity.ai - claude.ai - gemini.google.com - copilot.microsoft.com Then compare AI traffic against organic search across four groups. ### Traffic - Sessions - Users - New users - Share of total traffic ### Engagement - Engagement rate - Bounce rate - Average engagement time - Pages per session ### Commercial behaviour - Sign-up rate - Conversion rate - Revenue per session - Lead rate - Assisted conversions ### Landing behaviour - Top AI landing pages - Top Google landing pages - Research content vs product pages - Branded vs category pages Also segment by device (desktop / mobile) and return behaviour (first visit, return visit, subsequent direct visit, subsequent branded Google search). There is an attribution problem here. Some AI-assisted journeys will not retain a measurable AI referrer. A user may discover a brand inside ChatGPT and later type the URL directly, Google the brand, or return from another device. Some AI influence may therefore appear as direct or another channel rather than a clean AI referral. [2] A useful attribution question is: “How did you first hear about us?” - ChatGPT - Claude - Gemini - Perplexity - Google - Other Referral attribution tells us who clicked. Self-attribution helps reveal who was influenced. ## 04 Results These are externally observed patterns, not my own first-party results yet. Across the external datasets reviewed, three patterns appear repeatedly. ### 1. AI referral volume remains small relative to Google Even with fast growth, AI referrals are not currently replacing traditional search traffic on most websites. [3] ### 2. The channel is growing quickly Semrush’s clickstream analysis reported ChatGPT outbound referrals growing 206% year over year, from January 2025 to January 2026. [1] ### 3. Visitors often appear more considered once they arrive Adobe has repeatedly measured stronger engagement among AI-referred retail users. Ahrefs has also reported unusually strong conversion performance from AI referrals on its own site. [3][4][5] Those results vary significantly by category and should not become a universal “AI traffic converts better” rule. **AI referrals may represent a different stage of the user journey, and measuring them as just another referral source misses the interesting part.** ## 05 Interpretation Traditional search often begins the evaluation process. AI assistants can participate in the evaluation process before the click happens. That creates a plausible reason why some AI-referred users appear unusually engaged. They may arrive after asking follow-up questions, comparing alternatives, understanding pricing, eliminating unsuitable options and clarifying terminology. This could make the referral smaller in volume but stronger in intent. That remains an interpretation until the effect is reproduced consistently using first-party data. ## 06 Limitations AI referral measurement is currently imperfect. Not every AI-assisted journey passes a detectable referrer. Different AI products handle outbound links differently. Some users discover a brand inside an assistant and later return through Google or direct traffic. External studies also cover different industries, time periods and customer journeys. The Ahrefs result is specific to Ahrefs. The Adobe findings come primarily from retail behaviour. Neither should automatically be generalised to every SaaS, ecommerce or content website. This study therefore treats external datasets as context rather than proof of my own site’s behaviour. ## 07 What I am testing next The next step is to compare AI and organic search visitors using first-party analytics over a sufficiently long period. I want to measure: - Session quality - Landing-page type - Conversion rate - Revenue per session - Return behaviour - Branded-search behaviour - Assisted conversions The most important question is not “how much traffic does ChatGPT send?” It is: what kind of visitor does ChatGPT send? ## Sources 1. Semrush: ChatGPT traffic analysis: Insights from 17 months of clickstream data. https://www.semrush.com/blog/chatgpt-search-insights/ 2. Semrush: How to track, measure, and boost AI referral traffic. https://www.semrush.com/blog/ai-referral-traffic/ 3. Ahrefs: AI chatbot traffic: what it is, and how to get more. https://ahrefs.com/blog/ai-chatbot-traffic/ 4. Adobe: The explosive rise of generative AI referral traffic. https://business.adobe.com/uk/blog/the-explosive-rise-of-generative-ai-referral-traffic 5. Adobe: U.S. online spending hits $88.7B in Oct. 2025, up 8.2% YoY. https://business.adobe.com/uk/blog/adi-october-2025-holiday-shopping-actuals ## Changelog - 15 Oct 2026: Checked every cited figure against its source. All figures held up. Added exact dates and comparison groups, moved Adobe’s October 2025 conversion figure to its correct source [5], and moved the Semrush tracking guide [2] to the measurement section it supports. - 01 Oct 2026: External evidence review and first-party measurement framework published. ## Related - Note: [Count what ChatGPT visitors do, not just how many arrive](https://tanujsharma.xyz/notes/measure-what-chatgpt-visitors-do) - Essay: [SEO Is Not Dead. Search Has Become Bigger Than Google.](https://tanujsharma.xyz/writing/seo-is-not-dead) ## Research into practice I publish what I learn here. At MarketingPal, I use what holds up to run search and marketing for lean teams. https://marketingpal.co