ChatGPT Is Now a Product Discovery Engine. Here Is What Actually Gets You Recommended

ChatGPT just quietly changed what it is. For most of the past year it was a place people asked questions, and occasionally a place they clicked "Buy." As of this month it is a product discovery engine. Ask it for the best waterproof hiking boot under 200 dollars and it will browse a catalog of real products, compare them, show images and prices, and let you buy without leaving the chat.

OpenAI announced the expansion in July, extending the Agentic Commerce Protocol from checkout into full product discovery (OpenAI). The short version: ChatGPT no longer needs to send you to a retailer's site to find a product. It reads structured feeds from merchants and recommends from them directly.

If you run an online store, this is the most important shift in the buying funnel since AI Overviews started answering research questions. And most of the advice you will read about it is wrong, because it treats this like SEO. It is not SEO. There is no page to rank, no meta title to write, no blue link to win.

There is no ranking here, there is a feed

The mental model to drop first: ChatGPT is not crawling your product pages and ranking them. It is ingesting a structured product feed you provide, in CSV or JSON, and choosing from that (OpenAI Developers).

That feed carries the fields you would expect: product ID, title, description, price, currency, availability, images, brand, and identifiers like GTIN. Merchants can refresh it as often as every 15 minutes, which is the tell for how much price and stock accuracy matters here. A stale feed is not a slightly worse ranking. It is a product that gets quietly dropped when the price it quoted turns out to be wrong.

OpenAI has been clear that these results are not ads and are not influenced by paid placement. Products surface based on feed quality, relevance to the query, and user context. Recommended attributes like reviews, ratings, and performance signals are what lift one product above another.

So the "content" that wins in ChatGPT commerce is not your blog or your category copy. It is your product data. This is the part traditional SEO teams consistently underrate, and we see it constantly: the merchants with the cleanest, most complete, most honest feeds are the ones a machine can confidently recommend.

Your catalog may already be in it

Here is the part that catches people off guard. You may not need to do anything to be eligible, and you may already be listed without realising.

US Etsy sellers were pulled in automatically through the Offsite Ads program. Shopify merchants on a paid plan with Shopify Payments can enable the ChatGPT sales channel from their admin after applying at chatgpt.com/merchants. Major retailers including Target, Best Buy, Sephora, Home Depot, and Wayfair are already integrated (OpenAI).

Automatic eligibility sounds like a gift. It is closer to a warning. Being in the index is not the same as being recommended. If your catalog got swept in through a platform integration and your product data is thin, inconsistent, or missing reviews, you are in the room but you are the product the model talks around, not the one it names.

Being included is not being recommended

This is the whole game, so it is worth slowing down on.

When ChatGPT picks a product to recommend, it is making a confidence call. Can I trust this price? Is it actually in stock? Do the reviews suggest a real customer will be happy? Is the description specific enough that I can match it to what this person asked for? Every gap in your feed is a reason for the model to reach for a competitor whose data answers those questions cleanly.

Three things move that confidence needle more than anything else:

• Real-time accuracy. Wrong prices and phantom stock do not just cost one sale. Inconsistent or outdated data gives the agent a reason to deprioritise you in future sessions. Machines remember bad data.
• Reviews and ratings in the feed. A product with credible aggregate ratings gives the model something to justify a recommendation with. A product without them is a guess it would rather not make.
• Descriptions written for a question, not a keyword. Old-school SEO product copy stuffed with search terms reads as noise to a model trying to match intent. Write descriptions that answer what a buyer would actually ask: what it is for, who it suits, what it is made of, when not to buy it.

If that sounds like the discipline behind getting cited in AI answers generally, it is. We wrote about the passage-level version of this in how to structure content so LLMs cite you. Product discovery is the same principle applied to a feed instead of a page.

The Cyprus and small-merchant reality check

Most of this is US-first. Etsy auto-inclusion, the Shopify sales channel, the big-box launch partners: they are American rollouts. If you are selling from Cyprus or serving a non-US market, you are not the priority audience for this launch, and it would be dishonest to tell you otherwise.

That does not mean sit it out. It means prepare the thing that will matter when it reaches you, which is your feed. Clean product data, structured identifiers, honest availability, and genuine reviews are not ChatGPT-specific work. They are the same feed hygiene that already decides your fate in Google Shopping and, increasingly, in agentic checkout across platforms. You are not building for one channel, you are building the infrastructure every AI buyer will read.

For a smaller merchant, there is even an edge here. This is a data quality contest, not a domain authority contest. You cannot outrank Wayfair on links, but you can absolutely out-describe them on a niche product with a cleaner, more specific, better-reviewed feed entry. Machines do not care how big you are. They care whether your data is trustworthy.

What we would actually do this quarter

Our practitioner take, stripped of hype: treat product data as the new on-page SEO and treat this as infrastructure work, not a content campaign.

1. Audit feed completeness before anything else. Every product should have a real description, accurate price, correct availability, high-quality images, and a valid identifier. Fix the feed, not the funnel.
2. Get reviews and ratings into the feed. If you collect reviews but they never reach your product data, they cannot help you where the recommendation happens.
3. Rewrite descriptions for questions. Pull your five best sellers and rewrite their descriptions to answer what a real buyer asks out loud. That is the language the model matches against.
4. Do the margin math before you flip anything on. OpenAI charges merchants a transaction fee on ChatGPT purchases, on top of your payment processing. Know your number before you celebrate the channel.
5. Watch who owns the customer. As we argued when Google launched Universal Cart, the strategic question with any off-site checkout is whether you keep the customer relationship and the data or rent someone else's audience. Read the terms, not the press release.

The bigger picture is the one we have been making all year: the click to your website is no longer the finish line. It happened with research queries and AI Overviews, we covered it in what AI shopping agents mean for SEO, and product discovery in ChatGPT is the same story reaching the buy button. The businesses that win are not the ones shouting loudest. They are the ones whose data a machine can trust.

If you want a second pair of eyes on whether your product data is ready for this, that is exactly the kind of thing our AI search optimization work covers, and you can start with a free SEO review.