AI referrals to online stores rose more than 200% in a year yet drive little click-through, while a chatbot recommendation can double a brand's visits over rivals. B2C ecommerce is forecast to top 4.9 trillion dollars by 2030.

In Short

Similarweb studied how people shop online and found that when an AI like ChatGPT recommends a brand, shoppers are much more likely to visit that brand than a competitor, sometimes twice as likely. But those AI chats don't send many direct clicks to stores compared with regular search, and most people who use AI to shop still use search too. What changes is that being the brand an AI names now matters a lot, even though almost no click shows up in a company's traffic reports to prove it.

The finding

Similarweb published its State of Ecommerce 2026 report today, September 10, 2026, describing how conversational AI tools are reshaping the path shoppers take toward a purchase without generating the traffic that once made that path measurable. The report, issued from Tel Aviv over Business Wire at 8:00 a.m. Eastern Daylight Time, frames its central claim in a single line: consumer AI conversations augment search rather than replacing it, but AI chat recommendations hand brands a material advantage.

The measured effect is uneven across two dimensions. According to Similarweb, direct referrals from dedicated AI platforms increased more than 200% over the past year yet still do not produce a large volume of traffic to ecommerce marketplaces and online stores. Set against search, AI conversations do not result in as much immediate click-through traffic. The influence arrives instead at the recommendation itself. AI buying recommendations, the company found, exert an outsize pull on purchases, in some cases giving the recommended brand a two-to-one advantage over its competitors.

That advantage is documented brand by brand in a chart the report builds from United States desktop data covering July through December 2025. When AI recommends American Express, 7.2% of users visit American Express within seven days against 3.1% who visit Capital One. Reverse the recommendation and the gap widens: with Capital One recommended, 14.2% of users visit Capital One within the week while 3.8% visit American Express. The pattern holds across categories. A Skyscanner recommendation draws 9.5% of users to Skyscanner and 7.6% to Kayak; a Kayak recommendation draws 12.0% to Kayak and 3.4% to Skyscanner. In beauty, a Sephora recommendation sends 7.9% to Sephora and 3.3% to Ulta, while an Ulta recommendation sends 7.6% to Ulta and 4.6% to Sephora.

Stacking, not switching

The report's argument rests on a behavioral observation rather than a displacement thesis. Consumers are folding ChatGPT, Gemini and other conversational tools into a routine part of the buying process, according to Similarweb, but they are not abandoning search. The company puts a figure on the overlap: 89% of the time, consumers who use AI in their shopping research also use search.

Daniel Reid, Principal Insight Analyst for Consumer Goods and Retail at Similarweb and the lead author, described the pattern in terms of accumulation. "Consumers are not switching tools - they are stacking them," Reid said. "People are using AI to explore and narrow options while still turning to Search to move toward a decision. The most complex journeys, the ones that use both, convert the best. Retailers and brands who figure out how to master these new paths to product discovery stand to benefit."

The claim that AI influence persists past the moment of the chat, without a corresponding click, is one Similarweb has advanced before. Its panel-based work on tracing AI recommendations to real traffic set out the methodological basis for asserting that a recommendation shapes a later visit the recommending platform never gets credit for. Independent research points the same direction on quality while disputing the magnitude. A study of 973 ecommerce sites with 20 billion dollars in combined revenue found ChatGPT referrals underperforming traditional channels on both conversion rate and revenue per session, while Criteo reported in May 2026 that conversion from ChatGPT-referred traffic approached twice that of traditional search in consumer electronics, lifestyle and home categories. The spread across the published literature runs from three times to twenty-three times depending on method and site category.

The measurement gap underneath the number

The 200% referral increase carries a caveat that the same report supplies: the volume remains small in absolute terms. That combination, rapid growth on a low base with disproportionate downstream influence, is precisely the condition that resists conventional attribution. A shopper can receive a recommendation inside a chat window, then open a browser or a retailer app to complete the purchase, and the referral that shaped the decision generates no click, no session and no line in an analytics report.

Ahrefs research published on February 4, 2026 found ChatGPT sends roughly 190 times less traffic to websites than Google despite handling about 12% of its query volume, and Google Analytics only added a dedicated AI assistant channel grouping in May 2026. Reduced organic click-through compounds the problem from the search side. Ahrefs found Google's AI Overviews correlated with a 58% reduction in click-through for top-ranked pages, and SISTRIX data put the cost to the German market at 265 million lost organic clicks per month, with position-one click-through falling from 27% to 11% when a summary appears.

The infrastructure question drew a formal response earlier this month. NIQ and Similarweb announced a joint product to measure how AI assistants influence purchases, scheduled for an initial release in the fourth quarter of 2026 across a limited set of categories and markets, covering ChatGPT, Gemini, Google AI Mode, Perplexity and Claude. The reasoning behind that build is the same absence the ecommerce report describes: agentic transactions carried through Google's Universal Commerce Protocol and OpenAI's Agentic Commerce Protocol can move a shopper from question to completed sale without the referral, click and session record that two decades of digital measurement assumed.

What the report measures beyond AI

Similarweb produced the report in collaboration with Statista, which contributed a "What Comes Next" section of predictions. Among them is a forecast that global business-to-consumer ecommerce revenue will exceed 4.9 trillion U.S. dollars by 2030, an increase of more than 27% from 2026. The company's own at-a-glance summary sets the 2030 figure at 4.9 trillion dollars, up 27% on 2026, and places worldwide ecommerce site traffic up 6.8% year over year against a 1.3% rise in the comparable prior period.

The report also documents a continuing shift of ecommerce from the web toward apps. According to Similarweb, ecommerce app sessions are growing at roughly 1.3 times the rate of web visits, and 86.5% of U.S. consumers say they primarily shop on a smartphone or tablet. That channel migration runs alongside a retention problem other researchers have flagged, with Adobe finding that 72% of shoppers delete retail apps after a single use, and app install economics tightening as United Kingdom and Ireland cost per install climbed to 3.85 dollars.

Within marketplaces such as amazon.com, the report identifies clothing, shoes and jewelry as the strongest category, with unit sales up 33.7% in the United States, while the electronics category barely moved at 1.5% growth. The document breaks its global and regional analysis into five market segments: Marketplaces, Beauty and Cosmetics, Consumer Electronics, Fashion and Apparel, and Home and Garden. On regional revenue, Similarweb's summary lists the U.S. at 1.5 trillion dollars, up 23.9%, the U.K. at 159 billion dollars, up 16.6%, and Germany at 133 billion dollars, up 21.2%.

The report separately reports Gen AI referrals to ecommerce websites up 203% year over year, against 1.2% growth in direct traffic, and puts the proportion of shopper journeys that utilize AI at 11.4% of sessions, with 23.0% of those AI-and-search sessions converting.

Why this matters for marketers

The report crystallizes a problem the marketing community has been circling for a year: the channel gaining the most influence over what people buy is the one that reports the least about itself. A media buyer can watch AI referral volume rise 200% and still be unable to attribute a single completed sale to the recommendation that produced it, because the recommendation and the purchase happen in different places with no shared identifier between them. The brand-versus-competitor gaps Similarweb published quantify what being the recommended name is now worth, and a fourteen-point spread in seven-day visitation is not a rounding error in a category where customer acquisition cost is the binding constraint.

The commercial stakes have already surfaced in adjacent measurement moves. Similarweb opened a dataset showing 26% of ChatGPT replies carrying sponsored ads, and the emergence of paid placement inside AI answers means the recommendation surface is becoming an auction, not only an organic output. For brands that master the new paths, as Reid framed it, the reward is disproportionate visitation. For those that do not, the risk is a competitor named in their place, in a channel where the loss shows up nowhere in the numbers a marketing team currently reviews. The gross merchandise value flowing through agentic surfaces is projected to grow faster than the tooling built to attribute it, and whether measurement catches up before budgets shift is the open question the fourth-quarter NIQ and Similarweb release is meant to answer.

Timeline

Summary

Who: Similarweb, a digital data and analytics company listed on the New York Stock Exchange under SMWB, with a forecasting section contributed by Statista and analysis authored by Principal Insight Analyst Daniel Reid.

What: The State of Ecommerce 2026 report, which finds that AI referrals to online stores rose more than 200% in a year while still driving limited click-through, that an AI recommendation can give a brand up to a two-to-one visitation advantage over a competitor, and that global B2C ecommerce revenue is forecast to exceed 4.9 trillion dollars by 2030.

When: Published September 10, 2026, drawing on United States desktop recommendation data from July through December 2025.

Where: Issued from Tel Aviv, with brand-level findings measured across United States desktop shopping and global and regional segments spanning marketplaces, beauty, electronics, fashion and home.

Why: The AI recommendation is becoming decisive to what people buy at the same time that it generates the least measurable trail, leaving brands able to win or lose shopper visits in a channel that reports almost nothing to the analytics teams accountable for acquisition cost.