ChatGPT shopping: how retailers can prepare product data for AI discovery?

OpenAI launched Instant Checkout in September 2025, letting people buy directly from Etsy sellers without leaving ChatGPT. Six months later, it pulled the feature back because merchants found the checkout step too rigid to work with. Since then it's been focused on something more basic, getting ChatGPT to understand what a retailer is actually selling well enough to answer a direct question about it.

People already ask ChatGPT for the best running shoe under £100, and what it recommends comes down to how much it can actually work out about each retailer's shoes, starting with something as basic as the price.

A UK retailer's page for a running shoe called the "Ridge Trail 3" might still show a price left over from a sale that ended in June. The description reads "lightweight comfort built for everyday miles," which sounds fine to someone skimming past it, but it never actually says whether the shoe is waterproof. A person could dig that detail out of the reviews further down the page. ChatGPT doesn't have reviews to dig through, so if the spec isn't sitting in the text it can read, the shoe more or less doesn't exist to it, whatever a shopper might eventually have found out for themselves.

Retailers write their pages for shoppers scanning them and for Google crawling them. The Ridge Trail 3's page, like every other one, has picked up a third reader lately, one deciding whether the product's even worth a mention.

What is ChatGPT shopping?

ChatGPT shopping covers three related experiences: product results, an ongoing shopping conversation, and shopping research. The difference between them is mostly about how much help the shopper needs to make a decision.

1. Product results

The simplest experience is a product appearing directly in an ordinary answer.

For example, someone asks, "What's a good electric kettle for one person that boils quickly and doesn't take up much counter space?" might get back a specific kettle, its price included, and a short note on why it fits.

For the shopper, it still feels like a normal question and answer. They asked for help choosing something, and a relevant product appeared in the response.

2. Ongoing conversation

That first recommendation can become the starting point for a longer shopping conversation.

The shopper might follow up with, "How does that compare with the smaller Philips kettle?" Then, "Which one is quieter?" And finally, "I care more about noise than boiling speed. Which would you pick?"

ChatGPT already knows they want something quick and compact from the first question, and the answer shifts as their priorities do.

3. Shopping research

Some decisions involve several products and several requirements pulling in different directions at once.

A shopper planning a trip might ask, "Help me choose walking boots for a seven-day hiking trip in Scotland. I want something waterproof and lightweight, but I don't want to spend more than £180."

There isn't one obvious answer to that. A lighter boot tends to offer less support, a highly waterproof one usually weighs more, and the best-reviewed pair might sit outside the budget entirely.

Shopping research can clarify things such as the terrain or preferred fit, then look into suitable options across the web, weighing price, availability, reviews and specifications together for each one. What comes back explains itself, something like: "given your trip, budget and priorities, these are the strongest options, and here are the trade-offs between them."

4. Advertising

Advertising is separate from all of this.

A retailer cannot pay to have its kettle or walking boots chosen as the organic recommendation in these answers. What gets recommended depends on the product information available and how well it matches what the shopper is asking for.

Paid placements go through their own advertising programme and show up labelled as ads. That buys visibility. Whether the same product earns an organic recommendation afterwards is a separate question, decided by the product data alone.

How does ChatGPT discover products?

ChatGPT finds products through four separate routes, and a retailer can end up in some, all, or none of them depending on where it sells and how its site is built.

1. Publicly available product information

ChatGPT can read product pages the same way it reads any other page on the open web, through a crawler called OAI-SearchBot. If a retailer's robots.txt blocks that crawler, or a product page sits behind a login wall or heavy bot protection, ChatGPT simply can't see what's on it.

Crawlability decides whether ChatGPT can read a kettle's page at all. The title, the price and the spec sitting on that page decide what happens next, whether that kettle actually gets recommended.

2. Merchant product data through ACP

The Agentic Commerce Protocol, which OpenAI built with Stripe and launched alongside Instant Checkout in September 2025, gives merchants a way to hand ChatGPT structured data directly, so it isn't only working from a web page. A feed submitted this way can carry live stock counts, current pricing and dozens of attributes per product, and update far more often than a page gets recrawled.

Access has so far gone to a small number of approved partners, and the ones named publicly, Etsy, Walmart, several Shopify merchants, are American. UK eligibility isn't confirmed publicly either way.

3. Shopify Catalog

Shopify handles this automatically for retailers on its platform. Shopify Catalog sends product data to ChatGPT without a separate submission, carrying over whatever's actually in the catalogue, a vague title or a price that's a month out of date included, exactly as it sits on the retailer's own site.

4. ChatGPT advertising feeds

The product data alone decides whether a retailer's kettle or boots show up in an ordinary answer or in shopping research, a separate question entirely from how much that retailer spends on ads.

What product data matters for AI discovery?

Every route into ChatGPT, whether through a crawled page or a merchant feed, depends on the same underlying data. This is what ChatGPT actually needs from a product record, and what tends to go wrong when it's missing.

How retailers can prepare product data for ChatGPT shopping

1. Start with what shoppers actually ask

Look at site searches, support questions and return reasons. These tell you what information shoppers are struggling to find.

For example, if a jacket is often returned because it runs small, say that clearly on the product page. Focus on the questions people need answered before they buy.

2. Make product titles clear

A product title should make sense even without a photo.

"Cosy Knit Autumn Edit" doesn't say anything about the jumper itself. "Women's Wool Blend Crew Neck Jumper, Navy" does, stating the product type, the material and who it's for.

Include the product type, main material or feature, and who it's for. Save promotional language for the description.

3. Focus descriptions on useful facts

Provide shoppers with information that helps them make a decision.

"Stylish and versatile, perfect for any occasion" gives a shopper nothing to act on. "80% wool, 20% nylon. Hand wash only. Runs small, so consider sizing up" does.

Specific facts are more useful to both shoppers and ChatGPT than marketing language.

4. Treat every variant separately

If a shoe comes in six sizes and three colours, there are 18 versions someone can buy.

Each version should have the correct size, colour, price, stock status and image attached to it. They can still be grouped as one product, but the individual options need to be accurate.

5. Keep prices and stock up to date

Price and availability change constantly. Make sure there's one reliable source for this information, so an old sale price or an out-of-stock product doesn't stay visible by mistake.

6. Use reliable product and image links

Product and image links should work without logins, expired URLs or multiple redirects.

Images should also match the exact product variant. If someone is looking at the red version, they should see a photo of the red version.

7. Make important information easy to access

Details such as price, stock, materials and specifications should appear as readable text on the product page.

Your product page, structured data and product feeds should also say the same thing. Conflicting information makes it harder for ChatGPT to know what is correct.

Also check that OAI-SearchBot isn't accidentally being blocked by your robots.txt, CDN or firewall settings.

8. Make the data relevant for UK shoppers

For UK customers, use GBP, UK sizing, British spelling and clear VAT information.

Be specific about delivery, collection and returns. Include any safety or usage information required in the UK. Product data created for another market usually won't have this already.

How to measure AI-discovery readiness?

OpenAI's current public merchant guidance doesn't document a native dashboard for measuring organic product visibility. Retailers have to check this from more than one angle: how the catalogue actually looks right now, what their own analytics have quietly been recording, and how ChatGPT answers when someone just asks it directly.

1. Score the catalogue itself

Run through this list every few weeks. Prices change, stock levels shift, and small details slip out of date faster than most retailers expect, even on a catalogue that looked fine a month ago.

• Required fields are complete across every listing.

• Category-specific attributes are filled in for each product type.

• Price and availability match what's actually true right now.

• Data has been checked within the last few days.

• Every variant carries its own accurate price, stock level and image.

• Product and image URLs load cleanly, with no logins, redirects or expired links.

• Delivery and returns information is present and accurate.

2. Check the analytics

Set “utm_source=chatgpt.com” on links wherever possible, so traffic arriving from ChatGPT shows up as its own line, separate from general referral traffic. If a direct feed is in use, keep tracking parameters consistent across it too, so results from that specific feed can be identified on their own.

A third signal, assisted conversions, needs a bit more digging: purchases where a ChatGPT visit appeared earlier in the customer's journey, even if it wasn't the final click before buying. Standard last-click reporting credits whichever channel a shopper used right before checkout, so these are easy to miss even though they're often where ChatGPT's actual influence on a sale shows up.

3. Test it directly

Ask ChatGPT the kind of question a real shopper would ask, and make a habit of repeating it every so often. Watching which products actually come up, and whether that shifts as the catalogue improves, is a good way to tell whether any of this is actually working.

Product data is part of the customer experience

Earlier, a product listing only had to deal with two audiences: a shopper scrolling the page, and a search engine indexing it to help the page rank well. A shopper could understand a vague title by checking the photos or scrolling through a few reviews, and a search engine could still rank a page well even if one detail, like the sizing information, was thin or slightly out of date.

ChatGPT works differently from both of these. When someone asks a question related to buying a product, it looks at what the retailer has actually written about that product, checks whether that matches what the shopper needs, and decides whether to mention it at all. If the price is wrong, or a variant isn't listed properly, or a spec is missing, the product simply doesn't feature.

Take the running shoe example from earlier in this guide. Getting the sizes, colours and stock levels right for one pair of shoes isn't hard to do by hand. The real challenge is doing that properly for a few thousand products at once, while prices and stock levels keep changing every day. That's usually where things start to slip. Keeping that many details accurate, all the time, takes more than manual updates can reliably manage.

Fynd's AI PIM builds accurate titles and descriptions from raw data or photos, validates every listing before it publishes, and keeps the same catalogue consistent across every marketplace a retailer sells through.

Start with your best-selling category. 

Explore Fynd's AI PIM