AI referral traffic and conversion: what the 2026 data shows

In 2025 the standard objection to GEO was that AI assistants send almost no traffic. In 2026 the data providers answered with numbers, and the answer is more interesting than either side expected. AI referrals are still a small share of visits, but they have become the best-converting source most retailers have, and the visible referrals are only the part of the effect that carries a referrer header.

How much traffic AI actually sends

Similarweb's 2026 Generative AI Landscape measured referral visits, meaning clicks out of answers, from ChatGPT, Gemini, Claude, Perplexity, Grok and DeepSeek across desktop and mobile web worldwide. The average was 770.7 million visits a month between June 2025 and May 2026, up 117.4% on the prior twelve months, with marketplaces (46.8 million a month, up 237.3%), news and travel (44.5 million each, up 115.6%) and finance (19.1 million, up 236.5%) receiving the most (Similarweb, September 2026). Semrush's clickstream study, built on more than a billion rows of U.S. panel data from October 2024 to February 2026, found ChatGPT's referrals to the web grew 206% between January 2025 and January 2026, while total ChatGPT visits plateaued near a billion a month from November 2025 (Semrush, April 2026).

The source mix is lopsided. BrightEdge reported ChatGPT at 95.1% of the AI referral traffic it measures in August 2026, with Gemini at 2.4%, Perplexity at 1.1% and Claude at 0.8%; ChatGPT's referrals rose 101% between January and August 2026, including a 36% jump in August alone (BrightEdge, September 2026). Similarweb's mid-2025 analysis of the top 1,000 global domains likewise put ChatGPT above 80% of AI referrals (Similarweb, September 2026).

Against total site traffic it is still small. Bain's May 2026 analysis puts AI at up to a quarter of referral traffic for some retailers but under 1% of total traffic (Bain, "Agentic AI in Retail", May 2026). Adobe's cumulative figure gives the growth curve: AI referral traffic to U.S. retail sites in July 2026 was 1,219% above its October 2024 level, and 62% above July 2025, a slowdown from the 393% year-on-year growth of Q1 2026 and 138% in May (Adobe Analytics, as reported by Digital Commerce 360, June and August 2026).

The conversion flip

The most striking series of the year is Adobe's. Adobe Analytics covers online transactions across more than a trillion visits to U.S. retail sites, including more than 200 of the Top 2000 online retailers. In March 2025, visitors arriving from AI sources converted 38% worse than other visitors. By March 2026 they converted 42% better, by May 54% better, and by July 2026 60% better, the eleventh straight month in which AI traffic outperformed (Adobe Digital Insights, as reported by TechCrunch, April 2026, and Digital Commerce 360, June and August 2026). On Prime Day 2026 AI-referred shoppers converted 40% better than paid search, email and social, a reversal from 23% worse a year earlier (Adobe, July 2026).

Engagement moved with conversion. In March 2026 AI-referred visitors spent 48% longer on site and viewed 13% more pages; by July they spent 59% more time, bounced 33% less, added items to cart 28% more often and generated 53% more revenue per visit than other shoppers, up from 37% more in March (Adobe Analytics via TechCrunch and Digital Commerce 360, 2026). Similarweb's panel shows the same pattern for ChatGPT against Google in the United States: 15 minutes on site versus 8, 12 pages versus 9, and a 7% conversion rate on transactional sites versus 5% (Similarweb 2025 Generative AI Landscape, U.S. desktop, September 2025 data, as summarised June 2026).

One caution applies to every one of these comparisons. A shopper who clicks out of an AI answer has already compared options inside the assistant, so part of the conversion advantage is selection, not persuasion. The figures show that AI delivers pre-qualified visitors; they do not show that the same person would have converted less had they arrived from search. That distinction matters for how the numbers are used in a forecast.

Where AI referrals go

Not every category receives referrals at the same rate, because not every answer carries a citation. Similarweb's analysis of ChatGPT answers in the United States in May 2026 found that 22.6% of travel answers included a web citation, the highest of any industry measured, against 13.5% for retail and 6.8% across all conversations; in beauty, 54.7% of the sources cited were retail or ecommerce domains (Similarweb, September 2026). Semrush found that web search was enabled on only 34.5% of ChatGPT queries in February 2026, down from 46% in late 2024, which caps how many answers can send a visitor anywhere (Semrush, April 2026).

Destinations are concentrated too. Google alone received 21.6% of ChatGPT's outbound clicks in February 2026 and the top ten domains just over 30%, while the number of distinct domains receiving any referral fell from a peak of about 260,000 in October 2025 to about 170,000 (Semrush, April 2026). For a mid-sized brand, the practical reading is that AI referral volume is won or lost on whether the engine cites you at all, not on how it ranks you.

The attribution gap

Three independent measurements say the referrer undercounts AI's effect. Similarweb notes that visits without referrer data are recorded as direct traffic, so its referral figures are a floor (Similarweb, September 2026). Its 2026 Downstream Impact study tracked user journeys from July to December 2025 in finance, travel and beauty and found that brands recommended in ChatGPT answers received more site visits within seven days than competitors that were not mentioned, with much of that traffic arriving through Google search and credited to it (Similarweb, June 2026). Pew's browsing data shows the same at the Google end: users clicked a link inside an AI summary in only 1% of visits, yet 58% of panellists encountered at least one summary in a month (Pew Research Center, July 2025).

Our own study of more than 2 million AI conversations matched to subsequent browsing found the same shape: brand-site visits rose 1.5 to 2.5 times in the seven days after an AI mention, but only about 2.5% of those visits carried a trackable AI referral parameter even after ChatGPT's May 2026 link change (flowgen, The AI mention effect, July 2026). The visible referral is the receipt, not the purchase.

The Korean platform mix

Korean teams face a different source list. Opensurvey's second-half 2026 report found 58.1% of search users had used ChatGPT and 52.2% Gemini in the past three months, with Gemini up 23.3 percentage points in six months; generative AI was the first screen for 36.5% of work and study searches and 27.2% of knowledge searches, while Naver still held 67.1% of daily-life and 60.5% of news searches (Opensurvey, July 2026). Naver's share of Korean search reached 81.34% on 24 May 2026, up from 59.26% a year earlier, with Google at 14.87% (Internet Trend via Newsway, May 2026), after AI Briefing reached 20% of integrated search queries with a stated target of about 40% (Naver via kt nasmedia, April 2026).

Two consequences follow. First, a Korean referrer list has to include chatgpt.com, gemini.google.com, perplexity.ai, copilot.microsoft.com and claude.ai, and keep Gemini separate rather than folded into Google. Second, Naver AI Briefing does not arrive as a separate referrer; a click from an AI Briefing answer looks like a Naver search click. For most Korean brands the largest AI-answer surface is therefore invisible in the source report, and the only way to see it is a change in Naver organic clicks and branded queries for the questions AI Briefing covers.

What Korean marketers should measure

What this means for Korean brands

• Analyze the questions, not the keywords. Referral volume follows citation, and citation follows the specific questions buyers ask. Prompt Demand and GEO Insight exist to find those questions and show who is cited for them today.
• Create pages that earn the citation. With only a third of ChatGPT queries searching the web and 6.8% of answers citing anything, the page has to be the kind an engine cites: specific, current and structured. Content Agent drafts it; Channel Agent places it where engines look.
• Measure the whole effect. Report AI referrals as a floor, conversion per source as the value, and branded search and direct lift as the size. Analytics joins the three so the discussion moves from "AI sends no traffic" to "AI sends the best traffic we have, and here is how much we are missing".

Sources

1. Similarweb, "AI Referral Traffic by Industry: What 770 Million Monthly Visits Reveal", 3 September 2026
2. Similarweb, "AI Search Trends: What Changed in the Past Year", 24 June 2026 (2025 Generative AI Landscape, 2026 AI Brand Visibility Index, 2026 Downstream Impact study)
3. Semrush, "ChatGPT traffic analysis: Insights from 17 months of clickstream data", 7 April 2026
4. BrightEdge, "ChatGPT Referral Traffic More Than Doubles in 2026, Reaching Record 95.1% Share of AI Traffic", 24 September 2026
5. Adobe Digital Insights Q1 2026 data, as reported by TechCrunch, 16 April 2026
6. Adobe Analytics May 2026 data, as reported by Digital Commerce 360, 17 June 2026; July 2026 data, 19 August 2026
7. Adobe, "AI shoppers convert 40% better as Prime Day hits $26.4B", July 2026
8. Bain & Company, "Agentic AI in Retail: How Autonomous Shopping Is Redefining the Customer Journey", 14 May 2026
9. Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results", 22 July 2025
10. Opensurvey, AI Search Trend Report 2026 H2, 20 July 2026
11. Internet Trend search share data, as reported by Newsway, 28 May 2026; kt nasmedia on Naver AI Briefing, 15 April 2026
12. flowgen, "The AI mention effect", 1 July 2026