How to Get Your Magento Catalogue Into ChatGPT Shopping

There is a lot of advice circulating that says if you add enough schema.org markup, AI assistants will start recommending your products. For ChatGPT specifically, that advice is wrong, and it is wrong in a way that wastes real money.

OpenAI does not discover your products by crawling your markup. It takes a feed. Here is what that actually means and what to do about it.

It is a feed, not a crawl

OpenAI’s commerce documentation is explicit. The integration starts by sharing a structured product feed with OpenAI, which is what gives their shopping experience the catalogue data it needs to index products and present accurate information. Delivery is by file upload over SFTP, or through their API. Promotions are API only.

Read the whole of their commerce documentation set and you will not find a single mention of schema.org, JSON-LD, microdata, structured data, or crawling. Not one. The word Merchant Center does not appear either. There is no markup you can add to a Magento product page that puts you into ChatGPT shopping.

Onboarding is also gated. Product feed onboarding is currently open to approved partners, and you apply through their merchant programme rather than switching it on yourself. Plan for an application, not an afternoon.

So there are two separate jobs

This is the mental model worth adopting, because conflating them is what causes the wasted effort.

Being readable. This is about crawlers and answer engines citing you in a conversation. It depends on your content being in the raw HTML and your robots.txt allowing the right agents. Markup helps here for Google and Bing, and does nothing documented for anyone else.

Being sellable. This is about a machine being able to list your product with a correct price and buy it. That runs on feeds and commerce protocols, and it has essentially nothing to do with your page markup.

Most stores are being sold work on the first while believing they are buying the second.

The protocol underneath

The thing worth understanding, because it outlives any single vendor, is that this is not an OpenAI-proprietary pipe. The Agentic Commerce Protocol is published as an open specification with a public repository, and OpenAI’s implementation sits on top of it. You can read it at agenticcommerce.dev.

That matters for how you invest. Building a clean, well-structured product feed with accurate prices, availability, identifiers and shipping data is not a bet on one assistant. It is the same asset Google Merchant Center already wants, and the same asset every agentic buying surface is converging on. The feed is the durable part. The destination will keep changing.

What to do, in this order

1. Fix your Google Merchant Center feed first

Same data, a channel that already exists, and requirements that are properly documented. If your Merchant Center account has disapproved items, price mismatches or missing identifiers today, you are not ready for a second feed consumer. You would just be replicating the errors.

The most common Magento cause is a feed that drifts out of sync with the site, which is exactly what Google’s automatic item updates exist to prevent. Those updates require price, priceCurrency, availability and condition to be present in your on-page markup. So markup does matter here, just not for the reason people think: it keeps your feed honest rather than getting you into an AI answer.

2. Get your identifiers right

GTIN, MPN, brand and SKU. Feeds are matched and deduplicated on identifiers, and a catalogue with inconsistent or absent GTINs will be poorly represented wherever it lands. This is unglamorous data hygiene in the Magento admin, and it is the single highest-value preparation you can do.

3. Check your robots.txt

Separate from the feed, ChatGPT also answers from search. OpenAI states that sites opted out of OAI-SearchBot will not appear in ChatGPT search answers. A surprising number of stores blanket-blocked every AI agent in 2024, never revisited it, and are now invisible. Allowing the search crawler is a different decision from allowing the training crawler, and you can make them separately.

4. Put your price in the HTML

Independent testing over the past eighteen months consistently finds that the major AI crawlers fetch JavaScript files but do not execute them. If your price or stock status is filled in after page load, which is common on Magento when private content or a third-party availability widget is involved, it is invisible to them. Fetch a product page with curl and read what comes back. We go into this in more depth in our guide to answer engine optimisation for Magento.

5. Then apply

With a clean feed, correct identifiers and a sane robots.txt, applying to the merchant programme is a short step. Doing it before those things are true just means being rejected or, worse, accepted with bad data.

What the feed actually has to contain

People imagine this is exotic. It is the same commerce vocabulary you already know, and if you have ever built a Merchant Center feed you have built most of it.

• Identity. A stable product ID, title, description, brand, and the identifiers above. The ID must not change between exports, which sounds obvious until someone regenerates it from a URL that then gets rewritten.
• Commercial state. Price with currency, sale price where it applies, availability, and condition. This is the part that goes stale fastest and causes the most rejections.
• Variants. The relationship between a configurable product and its children, expressed so a machine can tell that four rows are one product in four sizes rather than four products. Magento models this cleanly, and feed exports frequently flatten it badly.
• Fulfilment. Shipping cost and delivery estimate by destination, and your return policy. Increasingly this is what decides whether an agent will actually complete a purchase rather than just list you.
• Media. Image URLs that resolve without a session, without a watermark overlay, and at a size worth showing.

The failure mode is almost never a missing field. It is a field that is present and wrong on eight percent of the catalogue, which is enough to get the whole feed distrusted.

Where the feed comes from in Magento

Three realistic options, and the right one depends on how much your catalogue moves.

A feed extension. The common choice, and fine for most stores. The thing to check is generation frequency against how often your prices change. A nightly feed on a store running hourly repricing will be wrong most of the day, and no amount of markup fixes that.

A custom export. Worth it when your product data needs real transformation, for example when the sellable unit differs from the catalogue unit, or when bundles have to be decomposed. Build it as a scheduled job with a validation step that refuses to publish a feed that fails its own checks.

Content API rather than a file. For large, fast-moving catalogues, pushing changes beats regenerating everything. More engineering, far less drift.

Whichever you pick, add monitoring on the feed itself. The most expensive feed problem is not a broken export, because someone notices that. It is a feed that keeps publishing successfully with last week’s prices.

What not to buy

Two things are being sold hard right now and neither does what the pitch says.

• An AEO schema extension. Product and Offer markup is genuinely worth having, for Google rich results and for Merchant Center. It is not a route into ChatGPT shopping, and any extension marketed on that basis is describing something that does not exist.
• An llms.txt file. Google has stated these files are not needed for Search and will neither help nor hurt rankings. No vendor documents using one for ranking. It costs nothing, so add one if you like, but it is not a strategy and it should not be a line item.

The honest expectation

Agentic shopping is early. The volume is not there yet, and published network measurements of AI crawler traffic show enormous crawl-to-referral ratios, meaning these systems read far more than they send. Anyone promising you a traffic channel is guessing.

What is worth doing now is the part that was always worth doing: a correct, complete, well-identified product feed. It earns its keep in Merchant Center today, it is the entry requirement for agentic commerce tomorrow, and unlike a markup plugin it does not stop working when the fashion changes.

Book a free strategy call

Want to know whether your catalogue is actually ready? We will audit your feed quality, your product identifiers, what is in your raw HTML before JavaScript runs, and which agents your robots.txt is turning away. Book a free strategy call.