Learn / Start here
Can ChatGPT buy from your store?
Since March 2026, a shopper who buys through ChatGPT completes the purchase on your own checkout rather than inside the chat. Whether it succeeds depends on four things in order: whether the agent can find your catalogue, read your prices and terms as text, compare you against an alternative, and drive your checkout as far as the payment step. Failing any one of them ends the purchase, and most stores fail at least one without knowing.
The question is answerable, and the answer is usually more interesting than yes or no. What matters is where the attempt stops.
What changed, and why it is good news
Retail Dive reported on 25 March 2026 that OpenAI had announced the previous day that it was retiring the first version of Instant Checkout, about six months after its 29 September 2025 launch. Retail Dive quoted OpenAI: "the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we’re allowing merchants to use their own checkout experiences while we focus our efforts on product discovery."
Shopify’s help centre states the result plainly. ChatGPT users "complete their purchase on your online store checkout in a ChatGPT in-app browser, or in a new tab when customers use ChatGPT web".
That is good news, because you keep the customer, the order data and a checkout you control. It is also a larger obligation than before. A checkout problem used to cost you a place in somebody else’s buy flow. Now it costs you the order outright, on a page that is yours to fix.
Where attempts usually stop
At the door
Bot protection is the bluntest failure. An interstitial that asks the visitor to enable JavaScript and wait is unanswerable to an agent: it will never render your storefront, so every other quality of your store becomes irrelevant. The same applies to a robots.txt rule that disallows agent user agents wholesale.
Your CDN may be about to make part of this decision for you. Cloudflare wrote on 1 July 2026 that from 15 September 2026, for new domains onboarding to Cloudflare, bots in its Training and Agent categories will be blocked by default on pages that display ads, that the Search category stays allowed, and that customers can opt out in their Security settings before that date. Most storefronts do not display ads, so most are not directly hit. The point is that the categories exist and somebody is going to set them, and it should be you.
In the in-app browser
A ChatGPT purchase now finishes in a mobile in-app browser, or in a new tab from ChatGPT on the web. That means a narrow viewport, a session with none of the cookies you were quietly relying on, and no extensions. A checkout that assumes a returning desktop customer, or that needs width for its payment step, fails there while looking perfect on your laptop.
At the product data
A page can look complete to you and be nearly empty to a parser. Price rendered by a script after load, stock status shown only as a colour, variants described in an image: all of these are invisible. So is a returns policy published as a graphic, which is common and quietly costly. An agent asked for "free returns" cannot confirm you qualify, so it moves on to a store that says so in text.
At the size picker
The most frustrating failures happen last. An agent that has chosen your product, on price and terms, then cannot select a size because the picker only responds to a hover-then-click sequence. It saw the options. It could not activate them. The cart is abandoned by software that had already decided to buy from you.
What "can it buy" really means
A responsible test stops at the payment step. Nobody should be placing real orders on your store to answer this question, and no card details need to be entered to learn that a checkout is reachable. Everything up to that point is observable without spending a penny.
The honest answer
For most stores the answer today is "partly". Discovery and evaluation tend to be adequate. Checkout tends to be where the journey ends, because checkout is the part built hardest around a mouse, and it is now the part that carries the sale. Knowing which stage stops the agent turns an abstract worry into a specific ticket for whoever maintains your theme.
Questions
How do I test this myself?
Open your best-selling product page with JavaScript disabled and read it as text. If you cannot see the price, the stock status, the variant options and the returns terms, an agent reading your page has the same problem. Then open your checkout on a phone, in a narrow window, signed out. That pair of crude tests catches a surprising amount.
Will blocking AI agents protect my store?
It will make you invisible to them, which is a legitimate choice, but it should be a deliberate one rather than the accidental result of a bot rule. Note that "AI bots" is no longer one switch. Cloudflare separates Search, Training and Agent traffic, and blocking the Agent category is what removes you from assistant-driven purchases while leaving search crawling intact. Decide the three categories separately, and if you do want to block, block by name in robots.txt so the decision stays visible and reversible.
Will I be able to tell a ChatGPT sale from any other sale?
Not reliably from the order alone. The purchase completes on your checkout in an in-app browser, so it arrives looking like ordinary mobile traffic. A referrer from the assistant is the usual signal, and it is easy to lose: a redirect chain, a consent wall or a stripped referrer policy will all erase it. Decide how you want to attribute this traffic before the volume matters, because you cannot backfill it.