Discover in AI, Buy on Your Site: Why AI Checkout Stalled and What to Do

Shoppers use AI to find and compare products, but many still finish the purchase on the store’s own website, so the practical goal is to be recommended and then convert the visit on your site. In late 2025 the promise was that people would ask an assistant for a product and buy it without leaving the chat. A year later, the picture is more nuanced.

For stores, that changes what to optimize. This guide explains what happened, why in-chat checkout struggled, and how to set up your catalog and measurement for the channel as it actually behaves. The foundation is accurate data, which our Google Merchant Center agency work is built around.

What happened to in-chat checkout?

It launched with large ambitions and then grew more slowly than announced. The events below are the ones reported publicly.

• OpenAI launched Instant Checkout in ChatGPT in September 2025. By early 2026, reports said far fewer merchants were live than announced, and OpenAI was reported to be moving the feature into apps.
• Walmart was reported in March 2026 to see in-chat checkout convert at about a third of the rate of sending shoppers to its own site, while ChatGPT brought a higher share of new customers.
• Google, Microsoft and Shopify kept building: Google’s Universal Commerce Protocol and Microsoft’s Copilot Checkout launched in January 2026.

Treat these as reported signals rather than settled facts, and expect the picture to keep moving.

Why did in-chat checkout struggle?

Three problems stand out, and only one of them is about technology.

1. Data quality. Agents working with scraped or outdated product data showed wrong stock, delivery dates and shipping costs.
2. Transactions are hard. Payments, taxes, returns and customer accounts are complex to move into a chat.
3. Shopper habits. A survey by Semrush in December 2025 found that half of respondents had bought after researching with AI, but far fewer had bought directly inside an AI tool.

The first problem is the one a store can control. It is also the reason accurate, structured data is the best investment whichever way checkout evolves.

What does it mean for your store?

AI assistants are, for now, mostly a discovery channel. The goal is to be recommended, then to convert the visit on your site. That means:

• Feed the assistants accurate data. Structured, real-time catalogs beat scraped pages. See the product data AI shopping agents read.
• Make the landing experience match. The price, stock and delivery the assistant promised must be what the shopper sees.
• Measure AI as discovery. Track visits and new customers from AI assistants, not only purchases completed in chat.
• Keep the customer relationship. Checkout and customer data stay with you.

Where does the journey break between the assistant and your checkout?

A shopper who arrives from an assistant has already formed expectations. Each gap between what was promised and what the page shows costs you the sale. Use this table to find yours.

How do you set up measurement for AI as a discovery channel?

Start with what your analytics can already tell you, and be honest about what it cannot. Follow these steps.

1. Open your analytics and list traffic sources that come from AI assistants. Group them in a single report.
2. Compare conversion rate and average order value of that group against organic search.
3. Check what share of those visitors are new customers versus returning ones.
4. Review the landing pages those visitors reach. Note any that are not product pages.
5. Test five products yourself: ask an assistant for them and see what price, stock and delivery it shows against your site.
6. Record the gaps and fix the ones that appear most often.

Be careful with attribution. Some assistants pass no referrer information, so part of this traffic will look like direct visits. Treat your numbers as a floor, not a total.

How does this play out in a hypothetical store?

This is an illustration, not a client.

Imagine a home goods store that sees a modest but growing group of visitors arriving from AI assistants. Reviewing their landing pages, the owner finds that many land on a category page rather than the product they asked about, and the delivery estimate in the feed is two days shorter than checkout shows.

The fix has nothing to do with in-chat checkout. It is linking each feed item to the right product and variant page, and correcting the shipping settings. That is the kind of unglamorous work that decides whether discovery turns into orders.

What mistakes do stores make when planning for AI discovery?

Most mistakes come from treating AI like a new storefront instead of a new source of visitors. Here are the ones to watch for, with the symptom that gives each away.

• Chasing a checkout integration before fixing the feed. Symptom: product issues remain open in Google Merchant Center while the team debates protocols.
• Sending AI visitors to generic pages. Symptom: AI-sourced sessions show high bounce rates on category or home pages.
• Judging AI by last-click sales only. Symptom: AI looks irrelevant in reports because the shopper researched there and bought later from a different source.
• Letting promotions drift out of sync. Symptom: a sale price is live on the site but the feed still shows the old price for hours.
• Ignoring first-time buyer friction. Symptom: new-customer sessions reach the cart but abandon at account creation.

None of these needs new technology. Each needs someone to own the data and review it on a schedule, which is why we recommend naming one owner for the catalog before anything else.

What should you ask your team or agency before starting?

• Who owns the product catalog, and where is the single source of truth?
• How often does the feed refresh, and how do we know when it fails?
• Can we separate AI assistant traffic in our analytics today?
• Does each feed item link to the exact product and variant page?
• Which of the above would change if checkout moved into assistants?

When should you not worry about this?

Not every store needs to act now.

• If assistants send you almost no traffic and your catalog is small, basic Google Merchant Center hygiene is enough for now.
• If your products are bought mostly through repeat purchase or direct relationships, discovery in AI is a minor channel for you.
• If your data is unreliable, fix that before you invest in any checkout integration.

Should you ignore in-chat checkout?

No. Protocols are still developing, and some will stick. Read Google UCP vs OpenAI ACP to understand where they are going, and ChatGPT vs Gemini vs Copilot for how each assistant works today.

For the basics, start with agentic commerce explained, and use the readiness checklist to see what to prepare. To compare the platforms, see Google vs Microsoft Merchant Center vs ChatGPT Shopping.

How does Avafa work on this?

Avafa is a consultancy based in Spain that works remotely. We prepare the catalog data that assistants rely on, make feed and page data agree, and build one master feed for each channel. We do not promise placement in any assistant’s answers, since that is the platforms’ decision.

The service sits under our Google Merchant Center agency offer, with ChatGPT Shopping catalog preparation for stores that want to be ready there.

FAQs

Is AI shopping overhyped?

In-chat checkout grew slower than announced, but AI research before buying is already common. The opportunity is real; it just looks different.

What should I measure?

Traffic and new customers from AI assistants, and the conversion of those visits on your site.

Should I wait for in-chat checkout to mature?

No. The data work you do for discovery is the same work any checkout protocol will need, so it is not wasted either way.

Why do assistants sometimes show the wrong price or stock?

Usually because they used scraped or outdated data. Keeping your feed current and your page markup consistent reduces the chance.

Can I tell which visits come from AI assistants?

Partly. Some assistants identify themselves as a referrer and others do not, so expect an undercount.

Does this change how I run my checkout?

Mostly it means keeping it simple for first-time buyers, since AI brings new customers who do not know your brand yet.