TL;DR: After ChatGPT or another assistant recommends your product, the customer who clicks through judges your own page in seconds: who you are, what delivery costs, how returns work, whether paying feels safe. US survey data from PYMNTS shows that a majority of recent AI-assisted purchases still end on Amazon. The recommendation gets you the visit. Your page has to win the sale.
What happens after an AI recommends my product?
Most of the advice out there stops at the recommendation. Get into the feed, fix your schema, allow the crawlers, done. Many of those guides come from software vendors, so their story ends where the product ends. Fair enough.
Mine starts one click later.
A customer asks an assistant for a gift, a pan, a pair of running socks. The assistant names you. They tap the link and arrive on a page they have never seen, from a company they have never heard of, run by someone they have never met. That someone is you. And the assistant that just vouched for you is no longer in the room.
What they do next is old fashioned. They look for reasons to trust you, or reasons to leave. AI changed how they found you. It did not change what makes a stranger hand over a card number.
Do customers actually buy where the AI sends them?
Often, no. They buy where they already trust the checkout.
PYMNTS published a report on 30 October 2026 that puts some numbers on this. These are US consumer survey figures, and the Amazon numbers come from separate distributions, not one tracked journey. Still, the shape is hard to ignore.
What PYMNTS measured (US adults)
- Have bought something with AI assistance: about 132 million
- Begin online retail research with an AI tool: 58.4 million
- Still buy on Amazon even when an AI tool found the product: 54.5 million
- Amazon's share of online retail research starts: 29%
- Amazon's share of main purchase completions: 49%
- Amazon's share of most recent AI-assisted purchases: 59%
- Would let an agent both shop and buy, Sept 2025: 32%
- Would let an agent both shop and buy, Aug 2026: 24%
Some blunt arithmetic. Amazon starts 29% of the research and finishes 49% of the buying. That 20 point gap is people discovering things somewhere else and then walking over to the till they know. Put 54.5 million next to 132 million and you get roughly 41%, with the caveat that those two figures may not come from the same question. And willingness to let an agent shop and buy dropped 8 points in under a year. That's a quarter of the people who said yes, gone.
PYMNTS gives the reasons Amazon keeps the checkout: stored payment details, a known identity, familiar delivery expectations, loyalty benefits, an established returns process. None of those is a better product. All of them are trust.
The platforms seem to have noticed. TechCrunch reported on 5 October 2026 that TikTok launched an AI shopping assistant that answers questions on shipping, sizing and availability, with in app checkout so people can "buy directly from a brand from their For You feed." Meanwhile VTEX notes that OpenAI pulled Instant Checkout from ChatGPT in March 2026, and Forrester found in assistant purchasing was the least used feature. So the assistant finds the product. The buying still happens wherever the customer feels safe.
What do they check on my page first?
I don't have eye tracking data for this, so here is my reasoning, not a study. If those five things are why people finish on Amazon, they are also the five gaps a stranger will look for on your page. In roughly this order:
- Is this a real business? A name, a face, a place. "Family workshop in X since Y" beats a stock photo of a laptop.
- What will it cost to get here, and when? Delivery price and timing near the price, not three clicks away in a footer.
- What if it's wrong? Returns, in one plain sentence. People who don't know you assume the worst policy until you show them otherwise.
- Is paying safe? Payment methods they recognise, shown before the last step.
- Is this the thing the AI described? VTEX reports that assistants sometimes offered sold out items or products never meant for that channel, because they pulled stale data. If the assistant promised a colour you stopped making, the customer arrives already let down.
That last one is quietly brutal. You can lose the sale before the page even loads, because the assistant oversold you.
What can I fix this week?
Nothing here needs a developer or a budget. It needs a slow afternoon and a little honesty.
- Ask the assistants about yourself. Type your product category and town into ChatGPT and one other assistant. Read what they say about you. VTEX, citing Publicis Sapient, says few retailers check this regularly. Be one of the few.
- Fix the facts at the source. If the description is wrong, the problem is usually your own product data: vague attributes, old stock, missing sizes. VTEX recommends clean structured attributes and a robots.txt that lets AI assistants reach the catalogue.
- Move delivery and returns up. One line under the price. Cost, timing, how to send it back.
- Put a human on the page. Your name and a photo of you or the workshop. The assistant can't vouch for you once the customer has clicked through. You have to do that part yourself.
- Check the page on a phone, from a cold start. Not logged in, not on your wifi, not knowing the brand. Count how many seconds pass before you know what it costs to arrive.
Where this doesn't work
Some honest limits, because the data is thinner than the headlines.
The numbers come from one report on US consumers. Spain is not the US, and a survey is not a log of what people actually did. The Amazon measures come from separate distributions, not a funnel anyone tracked from first question to final payment, so the 29% and 49% are two snapshots, not one journey.
The PYMNTS article does not look at independent brands at all. The link I draw to small, founder led shops is my reasoning, not a measured result. It could be that people trust small brands more when an assistant names them. I don't know, and neither does the report.
And you probably don't know yet what share of your own visitors arrive from an AI assistant. Most analytics setups lump them in with direct or referral traffic. If it's 1%, this is a nice to have. If it's growing, it's urgent. Find that out before you rebuild anything.
FAQ
Can I pay to get recommended by ChatGPT? Not in the way you can buy an ad, as far as these sources show. The practical levers VTEX describes are clean product data, structured attributes and letting assistants access your catalogue.
Should I sell on Amazon instead? Maybe, for some products. But the data only says people trust a familiar checkout. It doesn't say they can't learn to trust yours. Fixing your own page is cheaper than handing over a margin.
Will AI agents soon buy everything for my customers? The PYMNTS numbers point the other way for now: willingness to let an agent shop and buy fell from 32% to 24%, and only 36.8 million US adults would let the AI hold their card.
How do I know if AI traffic matters for me? Start by asking a handful of recent customers how they found you, and look at your referral sources. You need your own number before you trust anyone else's.
Sources
- A: PYMNTS, "Amazon Wins 59% of AI-Assisted Purchases as Checkout Becomes the Prize", 30 October 2026. https://pymnts.com/commerce/ecommerce/2026/amazon-wins-59percent-ai-assisted-purchases-as-checkout-becomes-the-prize
- B: TechCrunch, "TikTok rolls out an AI shopping assistant and one-click checkout", 5 October 2026. https://techcrunch.com/2026/10/05/tiktok-rolls-out-an-ai-shopping-assistant-and-one-click-checkout/
- D: VTEX News, "Where agentic commerce pays off today", 1 November 2026. https://vtex.com/en-us/news/where-agentic-commerce-pays-off-today
Want a stranger's eyes on your page?
I run care_less from Santander, a one team studio for founder led businesses. The free 5 second test is simple: send me your site and I'll tell you what a stranger sees before they decide to stay or leave. That's the exact moment an AI referral lands. Book it here: https://zcal.co/i/O0g63S4u