The AI Shopper in 2026: Similarweb and Statista on How People Buy

New Similarweb and Statista data shows AI now appears in 11.4% of US online shopping journeys, and that shoppers use it alongside search, not in place of it. In this joint webinar on the AI shopper in 2026, Daniel, a principal insight analyst at Similarweb, presents clickstream data on how AI tools fit into the path to purchase. Kasia, a senior data journalist at Statista, follows with survey data from the US, UK and Germany on who uses AI to shop, how far they trust it, and why almost nobody wants an AI agent to buy for them yet.

TLDW

AI has more than doubled its share of US shopping journeys in two years, but 89% of those journeys also include search, and AI mostly sits in the middle, at the research and comparison stage. Shoppers who use both AI and search visit about twice as many sites and convert at 23%, against 12.5% for AI alone. AI sends under 1% of ecommerce visits as a last click, which Similarweb says hides its real influence, because an AI recommendation changes which brand a shopper visits over the following week. Statista’s surveys show that 30 to 40% of consumers trust AI to recommend or shortlist products, while only about a fifth would let it make the purchase.

Thoughts

The headline finding is that journeys using both AI and search convert best, and it is the weakest claim in the webinar. Daniel reads the extra site visits as validation: people check an AI answer with a search, feel sure, and buy. An audience member asks the obvious question in the Q&A, which is whether those shoppers were simply more likely to buy anyway. He admits the data is correlational. His defense is that a shopper who had already decided would take fewer steps, not more. I do not think that holds. Someone buying a laptop or a holiday visits 15 sites because the purchase is large and they are committed to making it. Someone idly asking a chatbot about running shoes visits seven and leaves. The type of purchase could explain both the longer journey and the higher conversion without the tools doing anything for each other.

The measurement argument is stronger, though it comes from a company that sells measurement. AI is present in 11.4% of journeys that include a transactional site and delivers under 1% of visits to those sites. The gap between those two numbers is the real story of the session. Most analytics tools credit the last click, so a channel that shapes the shortlist and then hands the shopper back to Google gets no credit. That is true, and it was already true of social ads and review sites. The part I would treat with care is the advice to watch growth instead of share. Growth of more than 200% a year is easy from a base below 1%, and it says little about where the number settles.

Kasia’s most useful observation is about what kind of objection people have to AI agents. The concerns she lists are payment security, how the data is used, and the agent choosing the wrong thing. None of those is a wish to do the shopping yourself. People are not attached to the weekly grocery order. They are worried about what happens when software does it. That is a problem that product design can solve, and her suggestions in the Q&A are sensible: a spending limit, a verified label for agents, payment details held by one agent only. It also suggests the low trust number for agentic commerce can move quickly once the first few safe experiences exist, in the way people stopped worrying about typing a card number into a website.

The demographic finding worth acting on is parents. Adults in households with children are more than twice as likely to use AI often when shopping, 24% against 11%. Age gets most of the attention in this kind of research, and the age gap is real, but time pressure may be the better predictor. The category data points the same way. People are most open to agents for travel, electronics and clothing, which are purchases with a lot of comparison work. The KFC example from China, where an agent orders a meal for two under a budget for pickup in 30 minutes, is a small purchase with a clear limit and little to get wrong. That is probably how agent buying starts in the West too.

Two things from the final minutes deserve more weight than they got. First, the seven day data shows visits, not sales. Daniel says so when asked, and his answer that visibility is “the gate that everything has to pass through” is fair, but a brand reading the Sephora number should know it measures attention. Second, Kasia mentions that some consumers suspect brands pay to be recommended. Today about 40% of UK shoppers trust AI product recommendations. That trust depends on the belief that the answer is not an advert. As AI products add paid placement, this is the number I would expect to fall. It also matters for publishers. The first journey Daniel draws includes a “top 10 best gifts” guide as a stop on the way to a purchase. The second puts AI in that middle position. He does not say what happened to the guide.

Key Takeaways

AI appears in 11.4% of US shopping journeys that include a transactional site, up from 4.5% about two years ago.
89% of those AI assisted journeys also include search. Only 11% use AI alone.
AI is a middle step 76% of the time, the first step 23% of the time and the last step 18% of the time. Search is spread evenly across the whole journey.
People who use both tools are nearly twice as likely to name AI as the most useful for research and comparison. The gap narrows when they are judging value.
AI only journeys visit about 7 sites, search only about 10, and journeys with both about 15.
Conversion rates are 12.5% for AI only, 17.3% for search only and 23% for both together. Similarweb says this is a correlation.
When an AI tool recommends Sephora, 7.9% of those users visit Sephora within seven days and 3.3% visit Ulta.
AI referrals are under 1% of visits to ecommerce sites and are growing more than 200% a year.
Statista forecasts global consumer ecommerce revenue above $4.9 trillion by 2030, more than 27% above 2026.
27% of US consumers use AI often or very often when shopping, against 23% in the UK and 19% in Germany. 44% of Germans never do.
34% of US Gen Z use AI often when shopping. For baby boomers the figure is 6%.
Adults living with children use AI often for shopping at more than twice the rate of those without, 24% against 11%.
Trust sits at 30 to 40% for recommending and shortlisting products, and at about a fifth for letting AI decide the purchase.
US consumers are most open to AI agents buying clothing and electronics. In the UK and Germany it is travel and electronics, and about 40% of UK Gen Z would let an agent book travel.

Chapters

03:32 How the Purchase Journey Changed

Daniel compares two journeys. The traditional one runs from an idea to a search, then to a brand site or a publisher’s buying guide, and ends at a store. The newer one has the same start and end, with an AI tool in the middle where the shopper refines the choice and gets brands recommended, before going back to search to check. His point is that more of the influence now happens between the first and last step.

05:20 AI Is in 11.4% of Shopping Journeys

Using clickstream data, Similarweb finds AI in 11.4% of US journeys that reach a transactional site. Nearly nine in ten of those also use search. Shoppers rate AI higher for research and comparison, and the two tools are closer when it comes to judging value. AI is concentrated in the research and consideration stages and is rare at discovery and at the final conversion.

07:33 Longer Journeys, Higher Conversion

Journeys that use both AI and search visit about 15 sites, roughly double the AI only figure, and convert at 23%. Daniel argues the extra visits are shoppers confirming what one tool told them with the other. He describes the longer journey as a sign that the shopper intends to buy.

09:06 Seven Day Influence and the Attribution Gap

Similarweb tracked what users did in the week after seeing an AI recommendation in finance, travel and beauty. They were consistently more likely to visit the recommended brand than a competitor. Last click attribution cannot see this, which is why AI shows up as less than 1% of ecommerce visits. Daniel says the growth rate is the number to watch.

11:41 Who Uses AI to Shop

Kasia presents Statista’s survey data. Shopping research is the third most common use of AI this year, after online research and education. Around 21% of people in the US and Germany and 26% in the UK have used AI tools for it. Frequent use is highest in the US and lowest in Germany, much higher among Gen Z and millennials, and notably higher in households with children.

14:54 Attitudes and Trust

Younger consumers are more positive about AI, and UK millennials are the most excited and the keenest to try new features early. Just under a third of people in all three countries are concerned about how fast AI is developing. The most common uses in shopping are finding new products, cutting research time and finding better value. Trust is fairly even across those functions and drops sharply for agentic commerce.

17:51 Agentic Commerce and the KFC Example

Handing the whole purchase to an agent is far from how people shop or use AI now. The specific worries are payment security, data use and wrong choices. Kasia sees openings where an agent has a clear advantage, such as a strict budget or a product that will sell out quickly. She calls shopping agents an unrealized opportunity in the West and points to China, where KFC began taking agent orders through an AI app in June.

23:10 Three Takeaways for Brands

Daniel closes with three points. Shoppers stack tools, so a brand needs to appear consistently in search, AI and social within the same session. Measurement has to move past the last click. And trust in AI has a ceiling at the purchase decision that varies by generation and category, so brands should invest where their own audience already is.

25:50 Q&A: Trust, Causation and Personalization

Kasia suggests building trust by having tools explain why a product was recommended, and by giving agents verification labels and spending limits. Daniel answers the causation question and says the seven day data did not track whether visits became sales. Asked which tools will be most useful next year, Kasia names price comparison, recommendations and personalized offers, with shopping agents last. She warns that an ad that follows someone across channels feels like surveillance.

Notable Quotes

“AI is kind of joining this journey, not owning the journey.”
Daniel, Similarweb, on why AI has not replaced search

“The complexity here is that signal of intent, not a signal of indecision.”
Daniel, Similarweb, on why longer journeys convert better

“It’s not that people are saying, ‘Oh, no, I want to go to the supermarket and do my weekly grocery shop myself.’ It’s more that, ‘Oh, what will happen if the AI is doing it for me?’”
Kasia, Statista, on what consumers object to about AI agents

“Trust in AI does appear to have a ceiling.”
Daniel, Similarweb, in his closing takeaways

“I want to be careful to not overclaim the causation, but I think it does point towards them working better together.”
Daniel, Similarweb, answering whether AI and search cause the higher conversion

“It doesn’t guarantee conversion, but it’s the gate that everything has to pass through.”
Daniel, Similarweb, on being recommended by AI tools

Watch the full webinar, The AI Shopper in 2026: Insights from Similarweb and Statista, on Similarweb’s YouTube channel.