What AI does when shoppers in six countries ask it to buy, from 360 questions on four AI surfaces
Opttab Research. Data collected 9 November 2026.
Not one of 1,440 AI answers bought anything. On 9 November 2026, Opttab Research asked four AI surfaces to buy, compare and buy, or choose a store for shoppers. The surfaces were ChatGPT, Gemini, Google AI Mode and Bing AI Mode. We ran the test in six countries: the United States, Canada, the United Kingdom, Germany, France and Türkiye.
Each market got the same 60 questions in its local language. Sessions were logged out and geo-targeted to each country. That gave 360 searches and 1,440 answers, and all 1,440 returned text. We read and coded every one.
This study follows two earlier Opttab Research studies. In our Black Friday 2026 study (collected 31 October 2026), 64.4% of “buy it for me” answers said AI couldn't complete the purchase. In our Cyber Monday 2026 study (collected 4 November 2026), that share was 82.1%. This time we look only at the buying moment.
The design differs. Prompt sets, surfaces and dates changed. This study adds Bing AI Mode and captures ChatGPT product cards. It leaves out Google AI Overviews and Copilot. Treat every comparison with the earlier studies as directional, not a measured change.
The short version: no AI assistant we tested bought anything. 0 of 1,440 answers claim a purchase. Only 2.9% offer an in-chat cart or checkout. Instead, 69.6% name or link a store in text and send the shopper there. ChatGPT and Google AI Mode show product cards, while Gemini and Bing AI Mode almost never do. The first store named is most often a multi-brand retailer, ahead of marketplaces. In Canada, that first store is Canadian in only 27.1% of answers. For a logged-out shopper, agentic commerce today is a referral business.
Key findings
No purchases, little checkout. 0 of 1,440 answers say the assistant completed a purchase. Only 41 (2.9%, cluster 95% CI 1.2–4.5%) offer an in-chat cart or checkout. 69.6% name or link a store and stop there.
AI says it can't buy. 71.7% of 240 “buy it for me” answers say the assistant can't complete the purchase. That falls between our Black Friday (64.4%) and Cyber Monday (82.1%) results.
Surfaces shop differently. ChatGPT shows product cards in 44.7% of answers. Google AI Mode does so in 27.5%. Gemini shows shoppable cards in 0.0% and Bing AI Mode in 0.3%.
Retailers get the hand-off. A multi-brand retailer is the first store named in 37.8% of answers. A marketplace goes first in 23.0%. The brand's own store goes first in 9.4%.
Canada leaks abroad. Only 27.1% of answers to Canadian shoppers name a Canadian store first. In the US, 72.9% name a home-country store first. The UK is next lowest at 38.8%.
The hand-off lacks facts. Only 23.8% of answers give a delivery or returns detail. 54.4% state no concrete price.
What it means: the store AI names, and the facts it can repeat about that store, now decide where an AI-assisted purchase happens. Brands should work on that hand-off now instead of waiting for in-chat checkout. To see which store AI names on your own buying prompts, track them in Opttab's AI visibility tracking.
The numbers at a glance
Number | What it measures |
|---|---|
0 of 1,440 | answers that say the assistant completed a purchase |
2.9% | of answers offer an in-chat cart, checkout or buy button (cluster 95% CI 1.2–4.5%) |
69.6% | of answers name or link a store in text, without product cards |
71.7% | of 240 “buy it for me” answers say the assistant can't buy |
44.7% | of ChatGPT answers show product cards (161 of 360, counted exactly) |
37.8% | of answers name a multi-brand retailer as the first place to buy |
27.1% | of answers to Canadian shoppers name a Canadian store first (n=240) |
23.8% | of answers give a delivery date, cut-off, delivery time or returns window |
54.4% | of answers state no concrete price |
Finding 1: No AI assistant bought anything, and few offered checkout
We asked AI to buy, compare and buy, or choose a store 1,440 times. Not one answer said it had placed an order (0 of 1,440). Only 41 answers offered an in-chat cart, checkout or buy button. That is 2.9% (cluster 95% CI 1.2–4.5%).
The usual reply is a referral. 69.6% of answers name or link a store in text. 13.1% show shoppable cards without checkout. 14.5% give no buying path at all.
No surface crosses 10%. ChatGPT offers checkout most often, in 8.3% of its 360 answers.
Surface | Offers checkout | Shoppable cards | Points to a store | No buying help |
|---|---|---|---|---|
ChatGPT | 8.3% | 24.2% | 56.4% | 11.1% |
Google AI Mode | 1.1% | 27.8% | 60.8% | 10.3% |
Bing AI Mode | 1.1% | 0.3% | 89.7% | 8.9% |
Gemini | 0.8% | 0.0% | 71.4% | 27.8% |
Asking directly barely helps. Some prompts ask outright whether AI can check out for the shopper. On those, 14.2% of 240 answers offer checkout. 48.3% give no buying path at all.
North America and the UK see a little more checkout. Offers reach 4.6% in Canada, 4.2% in the US and 3.8% in the UK. They drop to 2.9% in Türkiye, 1.2% in Germany and 0.4% in France (n=240 each, indicative).
Why it matters: Treat the hand-off as the agentic moment. For a logged-out shopper, the store page AI points to has to close the sale on its own. Opttab's Agentic Commerce Readiness Audit scores that path from 0 to 100 and gives you a roadmap to fix it.
Finding 2: AI tells shoppers plainly that it can't buy for them
Across all answers, 41.5% say the assistant can't complete the purchase, place the order or pay (597 of 1,440, cluster 95% CI 33.1–50.3%). On direct “buy it for me” prompts, the share climbs to 71.7% of 240 answers. On compare-and-buy prompts it is 55.0%.
That 71.7% sits between our two earlier studies. Our Black Friday 2026 study found 64.4% of 348 answers (31 October 2026). Our Cyber Monday 2026 study found 82.1% of 162 (4 November 2026). Prompt sets and surfaces differ, so read the comparison as directional.
Store-choice prompts never trigger a refusal (0.0%). There, the shopper only asked where to buy.
Surface | Says it can't buy | Says buying in chat needs something extra |
|---|---|---|
Gemini | 52.2% | 8.1% |
ChatGPT | 46.7% | 13.1% |
Bing AI Mode | 39.4% | 7.5% |
Google AI Mode | 27.5% | 9.7% |
Requirements surface when shoppers ask. Only 9.6% of all answers say buying in chat needs something extra, such as signing in, a connected account or a supported merchant. On checkout-capability questions, that reaches 37.5%.
Germany declines most. 49.6% of German answers say the assistant can't buy. In the US, 35.8% do (n=240 each, indicative).
Why it matters: AI is telling shoppers to buy elsewhere. The store named in that same answer receives them. Today, being named matters more than being bought through. Use Opttab's AI visibility tracking to see which stores AI names on your “buy it for me” prompts, market by market.
Finding 3: ChatGPT and Google AI Mode show products, while Gemini and Bing AI Mode name stores
We counted product cards exactly from the cards we collected. ChatGPT shows them in 44.7% of answers (161 of 360, 1.37 cards per answer). Google AI Mode shows them in 27.5% (99 of 360, 0.96 cards per answer). Across both surfaces, 36.1% of 720 answers show cards.
The coded buying ladder confirms the gap on the other two surfaces. Gemini shows shoppable cards in 0.0% of answers and Bing AI Mode in 0.3%.
Bing AI Mode works as a text referrer. 89.7% of its answers name or link a store without cards. Gemini gives no buying path in 27.8% of answers, the most of any surface.
Country | Answers with product cards | Cards per answer |
|---|---|---|
United States | 51.7% | 1.68 |
United Kingdom | 50.8% | 1.77 |
Canada | 48.3% | 1.58 |
Germany | 24.2% | 0.63 |
France | 24.2% | 0.8 |
Türkiye | 17.5% | 0.52 |
Cards cluster in English-language markets. On the two card surfaces, cards appear in about half of US, UK and Canadian answers. Elsewhere they appear in a quarter or fewer (n=120 per country, indicative).
Big retailers fill the cards. On Google AI Mode, Walmart appears most often (21). Walmart.ca (20), Amazon CA (18), Amazon.co.uk (18) and Best Buy (13) follow.
Why it matters: Cards draw on price, stock and a merchant link AI can read. Keep that product data complete wherever cards appear. Opttab's Feed Optimizer checks catalog quality for ACP and UCP. On Gemini and Bing AI Mode, your store has to be named in the answer text itself.
Finding 4: Multi-brand retailers get the first hand-off more often than marketplaces
We expected marketplaces to lead. They don't. A multi-brand retailer is the first store named in 37.8% of answers (cluster 95% CI 31.7–44.2%). A marketplace goes first in 23.0% (18.6–27.0%). The brand's own store goes first in 9.4% (6.0–13.1%). 19.5% of answers name no store at all.
Retailers lead in all six countries. The margin is narrow in Türkiye (31.7% to 28.3%). Price comparison sites go first in 13.3% of Turkish answers and 10.8% of German ones. In US, Canadian and UK answers, they go first in under 2%.
Country | Multi-brand retailer | Marketplace | Brand's own store | Price comparison | None named |
|---|---|---|---|---|---|
United States | 42.9% | 22.9% | 10.4% | 1.7% | 18.8% |
Canada | 39.2% | 25.0% | 10.8% | 1.2% | 20.4% |
United Kingdom | 40.0% | 22.5% | 10.8% | 1.7% | 20.8% |
Germany | 35.0% | 20.0% | 9.2% | 10.8% | 20.4% |
France | 37.9% | 19.2% | 10.4% | 5.4% | 20.8% |
Türkiye | 31.7% | 28.3% | 5.0% | 13.3% | 15.8% |
Brands do better on gifts. On gift-with-deadline prompts, 19.2% of answers name the brand's own store first. Marketplaces (31.7%) edge retailers (30.0%) on those prompts. On store-choice prompts, retailers go first in 66.7% of answers.
Surfaces lean differently. ChatGPT names a retailer first in 46.4% of answers. On Bing AI Mode, retailers (33.6%) and marketplaces (32.8%) run close to even.
Citations follow the same pattern. Retailers and marketplaces carry 38.9% of 6,707 answer-domain citation pairs. Review and comparison sites carry 24.1%.
Why it matters: The first store named is where the shopper goes. Check that the retailers carrying your products have accurate listings. Then give your own store a reason to be named first, such as gifting and delivery promises. Opttab's GEO/AEO actions map each buying prompt to your pages and list the content and product fixes against the most-cited pages.
Finding 5: Canadian and UK shoppers are often sent to foreign stores
Overall, 52.8% of answers name a store headquartered in the shopper's country first (cluster 95% CI 46.2–59.2%). The US (72.9%) and Türkiye (71.7%) are the most local. Canada is the least local. 27.1% of Canadian answers name a Canadian store first. 48.8% name a foreign store with a Canadian storefront.
We expected Canada to stand alone. The UK sits closer to Canada than to the US, at 38.8%. 5.0% of UK answers name a store with no UK storefront. 16.7% quote a price in a foreign currency.
Our Black Friday and Cyber Monday 2026 studies found 37.5% and 38.3% Canadian home-grown retailer mentions. Those studies counted every retailer named, not only the first. They measure something different.
Country | Home-country store first | Foreign store, local storefront | Foreign store, no local storefront | Local sources cited | Price in foreign currency |
|---|---|---|---|---|---|
United States | 72.9% | 2.1% | 1.7% | 81.3% | 0.8% |
Canada | 27.1% | 48.8% | 2.1% | 48.0% | 7.5% |
United Kingdom | 38.8% | 31.2% | 5.0% | 58.6% | 16.7% |
Germany | 53.8% | 21.7% | 0.0% | 87.5% | 0.4% |
France | 52.9% | 20.8% | 0.4% | 84.5% | 0.0% |
Türkiye | 71.7% | 9.6% | 1.7% | 85.8% | 0.8% |
Sources tell the same story. Only 48.0% of Canadian citations come from Canadian sites (n=1,002). In the UK, 58.6% come from UK sites (n=1,162). In Germany, 87.5% are local (n=1,171).
Why it matters: In Canada and the UK, domestic stores compete with foreign storefronts AI already treats as local. State country, currency and shipping in text so AI can tell you apart. Opttab's AXP bot-ready pages serve AI bots a clean, structured version of each page and keep it in sync when the page changes.
Finding 6: AI hands over the store without the delivery date or the price
Only 23.8% of answers give a concrete delivery date, cut-off, delivery time or returns window (343 of 1,440, cluster 95% CI 16.2–32.9%). Gift-with-deadline prompts reach 55.0%. Store-choice prompts reach 45.4%. Reorder prompts reach only 3.3%.
Prices go missing too. 54.4% of answers state no concrete price (cluster 95% CI 45.6–63.1%). 4.4% quote a price in a foreign currency.
Surface | Local-currency price | Foreign-currency price | No price | Delivery or returns detail |
|---|---|---|---|---|
ChatGPT | 59.7% | 0.8% | 39.4% | 25.0% |
Google AI Mode | 45.0% | 5.0% | 50.0% | 23.1% |
Gemini | 33.9% | 9.7% | 56.4% | 20.6% |
Bing AI Mode | 26.1% | 1.9% | 71.9% | 26.7% |
Warnings arrive with store choice. 19.3% of answers tell shoppers to verify the seller, price, stock or authenticity. When shoppers ask which store to use, that rises to 49.6%. Coder agreement on this field was moderate (kappa 0.479), so read it as indicative.
Why it matters: Shoppers will ask whether an order arrives in time. AI can only repeat what it can read. Publish Christmas cut-offs, delivery times, returns windows and prices in plain text. Opttab's MCP server also lets AI agents search your products, check stock and apply your policies.
Also worth knowing
Answers name many stores. 37.7% of answers name four or more stores. Only 10.3% name exactly one. ChatGPT names a single store in 21.9% of its answers, the most of any surface.
Follow-up actions are rare. 10.6% of answers offer something short of checkout, such as price tracking, alerts or a reorder schedule. That rises to 31.7% on checkout-capability prompts and 24.2% on reorder prompts. Coder agreement here was low (kappa 0.172), so treat it as indicative.
Payment and AI companies get cited. chatgpt.com is the most-cited domain, in 4.6% of answers. openai.com (2.8%) and stripe.com (2.0%) make the top 25. Software vendors carry 6.4% of all citations, and 10.2% on ChatGPT.
Google AI Mode cites stores most. Retailers and marketplaces carry 53.2% of Google AI Mode citations. On ChatGPT they carry 34.0%. Gemini leans hardest on review and comparison sites, at 32.2%.
Local champions top the citations. The most-cited domains include trendyol.com (4.2% of answers), walmart.com (4.0%), hepsiburada.com (3.7%), bestbuy.ca (3.4%) and mediamarkt.de (3.2%).
The playbook
Opttab helps 400+ brands get found and recommended by AI answer engines. Each play below comes from a finding above and names the Opttab feature that does the work. To see where your brand stands today, start with a free AI Visibility Report.
1. Test the hand-off path Evidence: 0 of 1,440 answers complete a purchase, and 69.6% name or link a store in text. Do this: Ask each assistant to buy your top products in every market. Follow the store it names to the product page and cart. Fix anything on that path that breaks. In Opttab: Prompt Generator writes the buying prompts your shoppers ask, for every intent. Agent and website analytics shows which AI bots reach your product pages and what they read.
2. Get into the product cards Evidence: ChatGPT shows product cards in 44.7% of answers and Google AI Mode in 27.5%. Cards appear in about half of US, UK and Canadian answers. Do this: Keep merchant feeds complete and current. Every key product needs a price, stock, currency, shipping and a working product link. In Opttab: Connect your Shopify store through Opttab's integrations. Then use the Feed Optimizer in AI Commerce to check catalog quality for ACP and UCP.
3. Equip the retailers AI names first Evidence: Multi-brand retailers are named first in 37.8% of answers and marketplaces in 23.0%. Price comparison sites go first in 13.3% of Turkish and 10.8% of German answers. Do this: Check your listings at the retailers, marketplaces and comparison sites AI names in each market. Each should be accurate, in stock and priced as on your own site. In Opttab: Answer Engine Analytics shows which pages AI cites for each prompt, by country and language, so you can see the stores behind each answer.
4. Give AI a reason to name your own store Evidence: The brand's own store is named first in 9.4% of answers overall and in 19.2% on gift-with-deadline prompts. Do this: Publish what only your store offers in plain text. Cover gifting options, delivery promises, exclusives and warranty terms. In Opttab: Content Studio writes briefs and articles from the visibility gaps where a retailer is named instead of you. GEO/AEO actions list the content and product fixes.
5. Publish delivery, returns and prices in plain text Evidence: Only 23.8% of answers give a delivery or returns detail, and 54.4% state no concrete price. Do this: Put Christmas delivery cut-offs, delivery times and returns windows on one crawlable page per market. State prices in text on product pages. In Opttab: The Agentic Commerce Readiness Audit in AI Commerce gives you a 0–100 score and a roadmap. GEO/AEO actions list the content and on-site technical fixes and publish changes to your tools.
6. Localize Canada and the UK Evidence: Only 27.1% of Canadian and 38.8% of UK answers name a home-country store first, and 16.7% of UK answers quote a foreign currency. Do this: State storefront, local currency and shipping country on key pages and in feeds. Give local publishers the right facts. In Opttab: Opttab's AI visibility tracking runs the same prompts by country and language, so you can compare Canadian, UK and US answers. Prompt Volume shows which countries the buying questions come from.
7. Be ready for the agent that can buy Evidence: Only 2.9% of answers offer checkout, but ChatGPT reaches 8.3%. 37.5% of checkout-capability answers name a requirement such as a supported merchant. Do this: Connect feeds, stock and checkout so agent hand-offs land on a working path. Make your products reachable by the agents shoppers already use. In Opttab: Opttab's MCP server, part of AI Commerce, lets AI agents search your products, check stock and apply your policies. AXP bot-ready pages give agents a clean, structured version of every page.
8. Track which store goes first Evidence: Which store goes first varies by surface, from 30.0% retailer-first on Gemini to 46.4% on ChatGPT, and by market. Do this: Measure which store each answer names first, by surface and market, prompt by prompt. Watch which pages AI agents read. In Opttab: Opttab's AI visibility tracking follows your buying prompts across AI models and markets. Agent and website analytics shows which AI bots visit, what they read, and the traffic and conversions AI search sends you. Start a 7-day free trial or book a demo.
Methodology
Prompts. We wrote 60 shopper questions across six intents: delegated purchase (“buy it for me”), compare and buy, store choice, checkout capability, reorder and replenish, and gift with a deadline. Each intent yields 240 answers.
Markets and languages. We covered six markets: the United States, Canada, the United Kingdom, Germany, France and Türkiye. We wrote prompts in English for the US, Canada and the UK. An AI model (Claude) rewrote them in German, French and Turkish as a local user would type them.
Surfaces. Every question went to ChatGPT, Gemini, Google AI Mode and Bing AI Mode. Google AI Overviews and Copilot were not part of this design. The surfaces give context for the shopper's experience. This is not a ranking of models.
Collection. We collected data on 9 November 2026 through the Cloro collection API. Sessions were logged out and geo-targeted to each country. 360 searches on four surfaces gave 1,440 answer slots, and all 1,440 returned text.
Citations and cards. 81.5% of answers cite at least one source. We classified 6,707 answer-domain citation pairs covering 3,404 domains by type and market. We counted product cards exactly from the collected cards for ChatGPT and Google AI Mode (720 answers).
Coding. We coded each answer against a written ten-field codebook, with model assistance. The fields cover buying level, refusals, follow-up actions, purchase requirements, stores named, first store type and market, price currency, delivery detail and risk warnings.
Quality checks. An independent second coder re-coded 120 answers (1,200 judgements). Overall agreement was 91.2%, ranging from 85.8% to 98.3% by field. Kappa ran from 0.172 (follow-up actions) to 0.966 (refusals to buy).
Statistics. Headline figures use 95% Wilson intervals and a cluster bootstrap by prompt. Comparisons with our Black Friday 2026 (31 October) and Cyber Monday 2026 (4 November) studies are directional.
Limitations
One day, one run. Answers vary from run to run. Treat the figures as a snapshot, not a trend.
Logged-out only. Signed-in shoppers, connected accounts and merchant integrations may see more checkout options. This study cannot show them.
Observational. We describe what answers say, not what drives them. The plays are our recommendations based on these observations, not tested interventions.
Two weak fields. Follow-up actions (kappa 0.172) had low coder agreement. Risk warnings (kappa 0.479) had moderate agreement. Treat both figures as indicative.
Rules-based classification. Home-grown and storefront status followed written rules. We did not check them against a registry.
Cards on two surfaces. Product card counts cover ChatGPT and Google AI Mode only. The coded buying ladder describes Gemini and Bing AI Mode.
Small cells. Country cells hold 120 to 240 answers, and surface-by-country cells hold fewer. Read them as indicative.
Different designs. Prompt sets, surfaces and dates differ from our Black Friday and Cyber Monday 2026 studies. No comparison with them is a measured change.
