Top AI Referral Sources Are Fragmenting: ChatGPT Leads at 722 Retailers, Down From 844, While Claude Jumps From 1 to 15

AI referral traffic is still small in volume, but stands apart in quality and growth

Ecommerce Trends: Which AI sources are sending referral traffic to online retailers?

AI referral traffic keeps growing for top online retailers; rankings reveal which LLMs are sending the most visitors to those ecommerce sites.

www.digitalcommerce360.com

Start with the top of the funnel. According to Adobe Analytics data for July 2026, AI referral traffic to U.S. retail sites rose 62% year over year, and 1,219% compared with October 2024, when generative AI tools began to spread. Quality diverges as well: AI-referred visitors generate 53% more revenue per visit and convert at a rate 60% higher than other traffic, a lead that has now held for 11 straight months. We covered the same metrics as of May in this article.

Even so, as Digital Commerce 360 notes, AI referrals have not overtaken traditional search engines in total volume. A narrow channel that keeps widening is where things stand in the summer of 2026. In the same July release, Adobe reported that the average machine readability of U.S. retail homepages was only 61%, so the gap on the receiving end remains wide.

All of that is the familiar "AI traffic is growing" story. What is new in this Digital Commerce 360 piece is a quarterly, retailer-by-retailer comparison of which AI platform that traffic is coming from.

How the top AI referral source shifted from Q1 to Q2

The data comes from the AI Commerce Rankings that Digital Commerce 360 and ReFiBuy launched in July 2026. Using the Top 1000 Database, which covers North America's largest online retailers by annual web sales, the rankings score each retailer's readiness for AI shopping every quarter. We explained how to read the first edition in an earlier article.

The Q2 edition added a new cut: for each retailer, which engine is the largest source of its AI referral traffic. Scot Wingo, founder and CEO of ReFiBuy, described it to Digital Commerce 360 this way.

"

One of the things that moved, for example, is in Q1, 844 retailers had ChatGPT as No. 1. That decreased to 722 in Q2.

Source: Scot Wingo (Founder and CEO, ReFiBuy)

The platforms behind it filled the gap. According to Wingo, retailers with Gemini as their top source doubled from 16 to 32, and Wingo said Perplexity picked up 12 more retailers. ReFiBuy's published Q2 findings put Perplexity at 8 to 21 and Claude at 1 to 15.

Of the four engines, ChatGPT was the only one to lose top-source positions in Q2. But 722 is still more than 70% of the 1,000 retailers. Leadership has begun to spread, yet ChatGPT remains by far the largest single supplier of AI referrals.

How to read the metric: a count of No. 1 positions, not share

This is the part of the story that deserves the most care. These figures are not traffic share. They are the number of retailers for which each engine is the largest AI referral source. ReFiBuy's own chart notes say exactly that: counts of retailers, not traffic shares.

By construction, a drop in ChatGPT's top-source count does not mean fewer visits came from ChatGPT. If a retailer's ChatGPT referrals stay flat while Gemini referrals surge past them, the count moves by one. The accurate reading of "122 fewer retailers with ChatGPT as No. 1" is therefore "at 122 retailers, another engine overtook ChatGPT."

Third-party data supports that reading. Similarweb's AI search statistics show ChatGPT's share of generative AI website visits falling from about 76% in June 2025 to about 53% in May 2026, while its absolute visit count stayed roughly flat. Over the same period Gemini rose from under 9% to 27 to 28%, and Claude from about 2% to about 9%.

Similarweb's figures measure visits to each assistant's own site, not referrals sent to retailers. Still, both datasets agree on the shape of the story: ChatGPT did not shrink, the others grew.

Three caveats apply. First, the comparison covers a single quarter, and ReFiBuy states that its methodology is refined with each edition, so it is too early to call this a trend. Second, Gemini, Perplexity and Claude combined lead at only 68 retailers, under a tenth of the field. Third, ReFiBuy coined the term "Agentic Commerce Optimization" and sells it as a service, giving it an incentive to stress the need for multi-engine work. The data has value on its own, but the interpretation should be read with that in mind.

What is behind Claude's jump from 1 to 15

Small in absolute terms, Claude's move is the largest of the four engines by growth rate. Wingo called the jump from one retailer to 15 "a really big move," and said that if it continued for eight quarters, Claude would be in the hunt alongside Gemini and ChatGPT.

The driver Wingo pointed to is product cards, the shoppable recommendation format Claude now shows. ReFiBuy's monitoring agents first reported seeing Claude product cards in mid-June 2026. That leaves only about two weeks of Q2 in which the cards were live, so most of the Q2 gain happened on the pre-product-card version of Claude. If the cards have a real effect, it will show up from Q3 onward, and the next edition will test it.

Anthropic's approach to commerce has differed from the checkout integrations OpenAI and Google pursued in 2025. As Digital Commerce 360 laid out, the company ran an internal agent-to-agent trading experiment called Project Deal, and in late August 2026 announced Claudeforce with Salesforce.

Then on September 2, the day before this article, Anthropic released commerce agent blueprints for Claude: a shopper agent that handles catalog search through cart building inside a retailer's own site or app, and a merchant agent that answers questions about sales and inventory. Accenture, Mastercard and Visa are named as partners, and Shopify is building a reference implementation. One distinction matters here. These agents are embedded in the retailer's own properties, which is a different mechanism from Claude's app sending visitors to retail sites. The referral growth and the blueprint launch are separate moves by the same company and should be read as such.

Reading Gemini's doubling and Perplexity's rise

Gemini's doubled top-source count tracks closely with its user growth. Similarweb estimates Gemini's monthly visits rose 450% from May 2025 to May 2026. Google has added buy buttons to search results and AI Mode and is widening the direct path to retailer catalogs through the Universal Commerce Protocol (UCP).

ReFiBuy's Q2 data includes a notable UCP figure. Only 225 of the Top 1000 had a detectable UCP endpoint, and those 225 had a median AI1000 rank of 304 versus 550 for the rest. As ReFiBuy itself notes, this is association, not attribution: it is not possible to tell whether UCP support lifted rankings or whether companies with the capacity to adopt it were already ahead.

Perplexity's growth from 8 to 21 is harder to explain. In June, ReFiBuy reported that Perplexity had withdrawn its buy capability and removed product cards. That its top-source count still grew 2.6x suggests Perplexity is used for research rather than purchasing, and that research is showing up as visits to retail sites. It is a reminder that having a checkout feature and sending traffic are not the same thing.

What this means for ecommerce teams: from single-engine to multi-engine

The four signals in the AI Commerce Rankings include "diversity of AI sources" alongside AI traffic volume. The fact that diversity is in the index at all reflects a view that optimizing for one engine is a risk. This quarter's spread in top-source counts adds a quarter of evidence to that view.

The same Q2 edition saw 972 of 1,000 retailers change rank, with a median move of 35 positions, and the index average fell from 42.0 to 39.7. ReFiBuy frames that as a bar that rises as traffic spreads across more engines and selling surfaces. AI shopping reshuffles far faster than traditional SEO ever did.

Measurement comes first in practice. If AI-referred visits are lumped together as "AI," a shift like this one cannot be confirmed in a retailer's own data. Referrals need to be split by ChatGPT, Gemini, Claude, Perplexity and Copilot, and the top source reviewed quarterly. Similarweb's analysis also shows that since May 2026, more than 60% of ChatGPT referrals land on homepages rather than product pages, which creates its own attribution blind spot.

Product data comes next. In August, Wingo said AI shopping moves on a different clock from traditional ecommerce and that retailers cannot treat the work as a one-time project. ReFiBuy's findings page adds that a catalog tuned for one engine now has to perform across a growing number of AI surfaces. Before engine-specific tactics, the shared foundation delivers more reliably across engines: price, availability and specifications present in the HTML before JavaScript runs, and structured data complete at the product level. Support for connection standards such as UCP and ACP is a step that follows that foundation.

Conclusion

What this data shows is not ChatGPT's retreat but the beginning of the end of an era in which AI referrals flow through a single door. By top-source count, ChatGPT still holds more than 70% of retailers, and Gemini, Perplexity and Claude together account for just 68.

Even so, the amount of movement in one quarter makes the next edition worth watching. Q3 will be the first full quarter with Claude's product cards live, Shopify merchants may adopt Anthropic's blueprints, and UCP support will spread ahead of the holidays. The question for ecommerce teams has shifted from "which AI do we bet on" to "are we readable no matter which AI shows up."