A 2026 survey of 4,000 shoppers across nine European countries suggests that AI is becoming a meaningful product-discovery and comparison channel. It also reveals a tension worth naming for clients: consumers are open to useful AI assistance but remain dissatisfied with many retailer-deployed tools. The survey does not measure actual revenue effects or change over time. Its strongest value is as a snapshot of reported behavior and preferences. It does not prove that particular AI features drive or destroy sales, and this article pairs it with behavioral and experimental research that gets closer to those questions.
AI Shopping in Europe 2026: What 4,000 Shoppers Say They Do
The survey data comes from "The Agent Era," published by the World Retail Congress with Vercel as commissioning partner. It covered 4,000 shoppers in May 2026: 1,350 in the UK, 1,000 in Germany, 825 across Southern Europe (Spain, France and Italy pooled), and 825 across the Nordics (Sweden, Norway, Finland and Denmark pooled). Twenty questions. One snapshot, no earlier wave.
Be straight with clients about what this is. Vercel sells storefront infrastructure, so the editorial framing leans toward the sponsor's interest. The regional results are useful as an exploratory snapshot, but the report does not disclose enough methodological detail (sampling method, demographics, weighting, response rate or significance testing) to assess representativeness or sampling error fully. It covers nine countries, with the UK making up 33.8% of the sample, so a pooled "European" figure cannot automatically stand for European consumers. All findings below are self-reported, and stated intentions routinely overstate observed behavior.
With those limits named, the snapshot still carries a structural signal worth building strategy around: respondents now select AI search as a shopping starting point more often than any single social platform.
| Survey result | UK | Southern Europe | Germany | Nordics |
|---|---|---|---|---|
| Start shopping journeys on AI search | 20% | 27% | 24% | 19% |
| Use ChatGPT when shopping for new products | 33% | 49% | 38% | 40% |
| Use no AI search when shopping | 50% | 32% | 40% | 44% |
| Have purchased on an AI recommendation (AI users) | 55% | 58% | 55% | 51% |
| Would spend less or refuse with AI-only service | 57% | 52% | 54% | 61% |
| Prefer physical stores for high-value purchases | 65% | 79% | 61% | 76% |
| Likely to choose a retailer for better AI tools | 43% | 59% | 40% | 39% |
AI Search Now Rivals Individual Social Platforms in Shopping Discovery
In every region surveyed, AI search (ChatGPT, Gemini, Claude) was named as a journey starting point by 19–27% of respondents, more than named Facebook, TikTok or Instagram individually.
The regional splits: 27% in Southern Europe, 24% in Germany, 20% in the UK, 19% in the Nordics. Compare that to Facebook (14–23%), TikTok (13–18%), and Instagram (14–16%). Two qualifications keep this honest. First, "AI search" pools three platforms while social media is split per platform, so the study does not show AI exceeding social media collectively. Second, some gaps are within a point (20% versus Facebook's 19% in the UK) and the report publishes no significance tests, so treat narrow leads as parity, roughly.
Even so, the company is remarkable. ChatGPT launched in November 2022; TikTok arrived in 2016, Instagram in 2010, Facebook in 2004. A channel under four years old now sits level with or ahead of each social platform as a stated shopping entry point. Retailer websites still dominate as a starting point (76–80%), followed by Google (65–68%).
One more correction to the common framing: AI answers are no longer entirely unbuyable. OpenAI announced advertising in ChatGPT in May 2026, including CPC buying and product-feed ads. Organic recommendations remain distinct from paid placements, and visibility in them depends on machine-readable product data and citable content, alongside platform integrations, retrieval systems and model behavior. That mix is a different discipline from paid social and from classic SEO ranking, and no single lever guarantees presence.
ChatGPT Leads, and a Third to a Half of Shoppers Use No AI Search at All
ChatGPT is the clear leader everywhere: 49% of Southern European respondents report using it when shopping for new products, 40% in the Nordics, 38% in Germany, 33% in the UK. Gemini sits second at 22–37%. Claude, Bing AI and Perplexity each stay below 6%.
The "none of the above" figure runs from 32% in Southern Europe up to 50% in the UK. Population data says the adoption ceiling is far away and uneven: Eurostat reports that 32.7% of EU residents aged 16–74 used generative AI tools in 2025, ranging from 65% among 16–24-year-olds to 14% among those 55–74. Retail AI adoption will differ as sharply by demographic group, which matters for any client whose customer base skews older.
What Shoppers Use AI Search For: Price Comparison Dominates
Among respondents who use AI for shopping, price comparison is the top use case in every region, cited by 51–55%. AI is functioning as a comparison engine.
- Compare prices: 51–55% across regions
- Advice on what to buy and product features: 38–53%
- Discover new brands or products: 31–42%
- Find products from brands they already shop with: 22–30%
- Buy inside the AI platform without redirect: 18–24% (24% in the UK)
Reported action follows. Among AI users, 58% in Southern Europe, 55% in the UK and Germany, and 51% in the Nordics say they have purchased a product following an AI recommendation. That is self-reported action, and it says nothing about incrementality or verified purchases, but as a statement of willingness it is hard to dismiss.
Does AI Shopping Actually Grow? What Behavioral Data Adds
The Vercel survey cannot establish that AI shopping is growing, because it contains no earlier wave. Behavioral data from the US nevertheless supports the direction of travel. Adobe Analytics, drawing on a dataset it says covers more than one trillion visits to US retail sites, recorded 138% year-on-year growth in AI-referred retail traffic in May 2026, and growth of 1,324% since October 2024. Those visitors converted 54% more often and generated 53% more revenue per visit than non-AI referrals, with 53% longer visits and 23% more pages viewed. The figures cover US sites and identifiable AI click-throughs only, so treat them as directional evidence, and never as a European benchmark.
The strategic read for clients holds even with the caveats: retailers with incomplete or inconsistent product data are ceding comparison visibility to competitors with clean feeds, in a channel whose measurable traffic is growing fast where it can be measured at all.
AI Discovery Often Ends in a Physical Store, and Attribution Is the Hard Part
For high-value purchases, stores still win in the survey. Between 61% and 79% of respondents across the four regions prefer to buy expensive items in a physical shop: 79% in Southern Europe, 76% in the Nordics, 65% in the UK, 61% in Germany. AI-assisted discovery feeds stores as well as websites: 3–5% of respondents say they discover via AI search and then buy in a physical store, and 10–13% discover via AI and buy on the retailer's website. Whether those shares are growing is something this snapshot cannot say.
The attribution problem may be larger than visible AI-referral traffic suggests. A 2026 observational study linked consenting users' assistant conversations (ChatGPT, Claude, Gemini) with their subsequent browsing. When an assistant recommended a previously unengaged brand, brand-name searches rose by 4.3 percentage points, visits to the brand's own site by 2.4 points, and visits to brand-specific retailer pages by 1.0 point. Most of that influence was invisible to ordinary referrer and last-click attribution. The study observed no transactions and no offline visits.
Closing the AI-to-Store Attribution Gap Takes More Than Analytics Alone
Be precise with clients about what each measurement layer can and cannot do. Web analytics sees the AI-referred session. Store analytics (footfall, zone dwell, queue behavior) measures the physical part of the journey. Neither, on its own, can tell you that a given store visit originated in ChatGPT. Connecting an AI exposure to a store outcome causally requires a linking mechanism or an experiment: tagged offers, loyalty identity, location-based consent, matched cohorts, geo-tests or incrementality testing.
There is a consulting opportunity sitting in exactly that gap. Retailers pouring budget into AI discovery without a measurement design are flying blind on ROI. A phased roadmap (instrument the store, choose a linking mechanism, then test) is the kind of engagement a well-positioned advisor can own, and the store-side instrumentation is its foundation.
How Consumers Rate Retailer AI Tools
Only two on-site AI capabilities earn net-positive experience ratings across all four European markets: visual search and personalized product recommendations. Virtual try-on and chatbots rate net-negative in most markets, and virtual try-on is the weakest-rated tool in the whole study. Net-positive here means positive ratings exceed negative ones; it is a satisfaction measure, and it says nothing about ROI or incremental revenue.
Usage is high, which makes the satisfaction gap worse. Personalized recommendations are used at least occasionally by 67% of Southern European respondents, 59% in the UK, 58% in the Nordics and 57% in Germany. Consumers meet these tools regularly and often find them inadequate: repetitive recommendations (26–33%), incorrect chatbot answers and irrelevant suggestions are each reported by roughly a quarter to a third depending on market.
Self-reported spending intent points the same way. More than half say they would spend less or refuse to spend if served solely by an AI assistant: 61% in the Nordics, 57% in the UK, 54% in Germany, 52% in Southern Europe. Only about one in ten say they would spend more. These are reported willingness figures, and they measure sentiment about a hypothetical, so do not read them as measured revenue loss.
Where do the failures come from? Some originate in incomplete or inconsistent product data; others come from retrieval, model behavior, interface design or a missing human escalation path. An audit of AI shopping agents found recommendations concentrating on a few "modal" products, shifting substantially after model updates, and showing strong position biases, with simple seller-side description changes able to move the agent's choices. Retailers need to test the complete system, since better data alone will only fix part of the problem.
Experimental evidence supports that implementation-first read. A large set of randomized field experiments at a cross-border retailer tested generative AI across seven workflows, from a pre-sale chatbot to product descriptions. Sales effects ranged from no detectable effect to a 16.3% increase, with four of the applications producing positive sales effects. The useful conclusion for a client roadmap: AI tools are neither inherently revenue-positive nor inherently destructive. Their effect depends on the job, the baseline system, implementation quality and customer segment. Treat each deployment as a hypothesis to be tested against conversion, margin, returns and customer-service outcomes.
The Human-AI Balance Consumers Say They Want
Appetite for AI-only service is a small minority, worth stating explicitly when a client proposes replacing their support team with a chatbot. Across regions, 42–49% of respondents prefer a mix of human and AI assistance online, 42–52% prefer humans only, and just 6–11% want AI alone.
The top stated concern is loss of human connection: 47% in the UK, 46% in the Nordics, 42% in Southern Europe, 40% in Germany. Privacy ranks second at 33–36%. Position AI as extending human service capacity. A chatbot deployment with no human escalation path is the highest-risk configuration in any market.
AI Quality as a Stated Retailer-Selection Factor
In every surveyed market, a large share of respondents say they would choose one retailer over another because it makes better use of AI tools on its website: 59% in Southern Europe, 43% in the UK, 40% in Germany, 39% in the Nordics. Note the wording: choosing between retailers in a hypothetical, which is weaker evidence than switching away from a retailer they already use. And since the Nordics sit below 40% and the report supplies no weighted Europe-wide figure, avoid repeating the headline claim that "over 40% of Europeans" behave this way.
On top of that, 26–37% say they would spend more with a retailer offering the latest AI guidance. Stated intent, yes, but directionally useful for prioritization. What respondents want improved is telling: more accurate personalized recommendations (41–51%) and smarter chatbots with human-like responses (33–43%). Emerging features rank lower: AR shopping experiences at 13–22% and AI styling assistants at 17–20% depending on market.
That points to an investment hierarchy for client roadmaps:
- Fix recommendation engines and product data first.
- Deploy or improve visual search second.
- Test emerging features like virtual try-on and AR as hypotheses, after foundational AI quality is demonstrably positive.
Agentic Commerce Infrastructure Is Arriving, With Caveats
The transactional infrastructure is being built now: Perplexity launched US shopping capabilities in November 2024, OpenAI added shopping to ChatGPT Search in April 2025 and US purchases from Etsy sellers in September 2025, Google unveiled AI Mode shopping with its Shopping Graph in May 2025, and Shopify announced Agentic Storefronts in March 2026. Availability differs by country, merchant and platform, and several of these channels remain US-focused or early access, so check the map before promising a client any of them in Europe.
Retailer sentiment is running ahead of the data. In an unpublished poll World Retail Congress ran among senior retailers from the US, Europe and Asia ahead of its 2026 Berlin event, 98% named fear of their brands failing to be selected by agentic search engines as their number one issue. No sample size or question wording was disclosed, so read it as the mood of one conference audience.
Treat the popular "brands, not aggregators, are winning the AI traffic" thesis with the same care. It comes from Vercel's CTO Malte Ubl as an observation from Vercel's own traffic, and it is a plausible one. But the 2026 conversation-matching study found assistant recommendations lifting visits to brand sites and retailer pages alike, with no general shift in the split between them. Corporate anecdotes point the same ambiguous way: JD Group's CEO told the congress that JD Sports receives ChatGPT orders in the US with two-thirds coming from new customers (no denominator or period disclosed), and Zalando's co-founder said 90% of its product-marketing content is now AI-generated. Suggestive, and worth watching; proof of a leverage reversal it is not yet.
For clients with strong direct-to-consumer infrastructure and clean product data, agentic commerce is an argument for investing in owned channels, and for measuring what those channels deliver both online and in the store where high-value purchases still close. If you are advising on AI discovery without a plan to measure the store visit at the end of it, that is the first gap worth raising.
Sources
- World Retail Congress / Vercel: "The Agent Era" (2026), survey of 4,000 European shoppers, May 2026; primary source for all survey statistics
- Adobe Digital Insights (June 2026), behavioral data on AI-referred traffic to US retail sites
- From Prompt to Purchase (2026 preprint), observational study matching AI assistant conversations with subsequent web activity
- Generative AI and Sales Productivity (preprint), randomized field experiments across seven retail AI workflows
- What Is Your AI Agent Buying? (2025 preprint), audit of AI shopping-agent recommendation behavior
- Eurostat (December 2025), generative AI usage in the EU population by age group
- OpenAI (May 2026), advertising in ChatGPT announcement
- Shopify, Agentic Storefronts announcement, March 2026