What if the last thing standing between you and a purchase was a conversation, not a checkout page? That idea moved from demo reels to real money in 2025, when chatbots started closing sales without ever sending users to a separate storefront. By 2027, the question is no longer whether people will buy things inside a chat window. It is whether the plumbing underneath that purchase can keep up with the promise.
How checkout moved into the chat window
For two decades, online buying followed a predictable script: browse, click, land on a product page, fill a cart, type card details, wait for a confirmation email. Chatbots broke that script in stages. Early shopping assistants were glorified search boxes that handed you a link. Then they started comparing products, checking stock, and answering questions about return policies. The final step, paying, was the hardest to crack, because payments are where trust, fraud controls, and regulation all collide.
That changed in September 2025. OpenAI launched Instant Checkout inside ChatGPT, letting US users buy from Etsy sellers directly in the chat, with Shopify merchants announced as the next wave. Google had already introduced Buy for Me in Google Shopping in the US, where its agent purchases from a retailer’s site on the user’s behalf. Suddenly the checkout page was optional. The conversation became the store.
The practical appeal is obvious. Nobody enjoys re-entering a shipping address on a site they visit once a year. A chatbot already knows what you asked for, which size you mentioned, and how much you wanted to spend. Removing four screens of friction from that moment is a genuine improvement, not a gimmick. This is the same shift happening across the assistant layer generally, as covered in this look at how AI chatbots work inside the tech that powers your conversations.
The protocols doing the quiet heavy lifting
Behind every in-chat purchase sits a standard that decides how a merchant, an AI agent, and a payment processor talk to each other. Two competing approaches dominate the conversation heading into 2027.
The Agentic Commerce Protocol (ACP) was co-developed by OpenAI and Stripe and published as an open specification in September 2025. It defines how an AI agent passes a purchase intent to a merchant and completes checkout without the user leaving the chat.
Google answered with the Agent Payments Protocol (AP2), published in September 2025. AP2 leans on cryptographic “mandates,” signed records of what the user intended to buy and what ended up in the cart. Those mandates create an audit trail showing the agent acted on real authorization rather than improvisation.
| Feature | ACP (OpenAI and Stripe) | AP2 (Google) |
|---|---|---|
| Published | September 2025 | September 2025 |
| Core idea | Agent-to-merchant checkout flow | Signed intent and cart mandates |
| Trust mechanism | Merchant integration and processor controls | Cryptographic proof of authorization |
| Interoperability by 2027 | Unconfirmed | Unconfirmed |
Here is the honest part: nobody can yet say whether these two standards converge, coexist, or one wins. Treat any confident claim of cross-protocol compatibility in 2027 as speculation. Card networks have layered their own programs on top. Visa announced Intelligent Commerce in April 2025, built around tokenized credentials and agent-specific controls. Mastercard unveiled Agent Pay the same month, including agentic tokens and a registration system for agents. Both are essentially saying the same thing: if a bot is going to spend someone’s money, the network wants to know exactly which bot, acting for which human, under which limits.
What actually works in practice (and what trips people up)
Watching this space closely, a few practical patterns separate smooth in-chat purchases from frustrating ones.
First, spending limits matter more than people expect. The most common mistake is authorizing an agent broadly, then discovering it interpreted “restock my coffee” as permission for a bulk order. Set a per-transaction cap and a monthly ceiling before you let any assistant buy on your behalf. Most agentic payment programs are built around exactly this kind of control, and skipping it is how you end up with a garage full of protein powder.
Second, confirm the merchant as well as the product. In-chat checkout hides the storefront, which is convenient but also means you may not notice you are buying from a third-party seller with a two-star reputation. Ask the assistant to name the seller before it pays.
Third, keep receipts in one place. Agentic purchases generate confirmations inside the chat, in email, and sometimes in a wallet app. Pick one destination and forward everything there. Disputes get messy when the proof of purchase is scattered across three interfaces.
Fourth, know what the agent cannot do. Returns, warranty claims, and address changes still usually require a human on the merchant side. A bot that happily sells you a jacket may have no idea how to process the exchange when it arrives in the wrong size.
Finally, remember that the assistant is only as good as the hardware and services running it. A sluggish device or an outdated browser can stall a payment mid-flow, and a stalled payment is a terrible time to discover a compatibility problem. If you are upgrading a machine this year, it is worth doing an AMD vs Intel CPU compatibility check before upgrading so a stalled checkout is never blamed on your processor.
The regulatory fog that will define 2027
Payments regulation was written for humans tapping “pay” on a screen. Agentic commerce asks a harder question: who authorized this transaction when the human only said “order more of the usual”?
In the European Union, the revised Payment Services Directive introduced strong customer authentication (SCA) rules that took effect in September 2019, requiring two-factor authentication for most electronic payments. How delegated, agent-initiated transactions fit those rules remains an open question. Regulators have not issued final guidance that settles it. That uncertainty is the single biggest brake on in-chat checkout expanding across Europe.
Two other EU laws shape the field. The EU AI Act, in force since August 1, 2024, includes transparency obligations for AI systems that interact with people, which bears directly on whether a user must be told they are dealing with an agent rather than a human. The Digital Markets Act, in force since November 2022 with obligations applying from March 2024, imposes interoperability duties on designated gatekeepers. That matters because a world where checkout only works inside one dominant assistant is precisely the outcome the DMA was designed to prevent.
For consumers, the practical read is simple: expect clearer disclosure that a bot is handling the transaction, and expect more friction, not less, in the EU than in the US through 2027. For merchants, expect compliance work that does not exist in the current US rollout.
Where the money actually flows
The economics of in-chat checkout are still being negotiated, and the negotiations are revealing.
Merchants gain a shorter path to purchase but hand over a slice of the customer relationship. When the assistant owns the conversation, the merchant loses the chance to upsell, capture an email, or control the post-purchase experience. Some sellers will accept that trade for volume. Others will resist, particularly brands with strong direct channels.
Payment processors, meanwhile, see agentic commerce as a land grab. Stripe’s early involvement with ACP positioned it as infrastructure for the OpenAI ecosystem. Card networks are racing to make sure their rails, not a rival’s, carry the transactions. The winner in 2027 may not be the best protocol but the one with the deepest merchant integrations.
There is also a quieter cost: fraud. Agents acting on stored credentials are an attractive target. The mandate systems in AP2 and the tokenization in Visa’s and Mastercard’s programs exist largely to limit damage when something goes wrong. Whether they succeed at scale is unproven, and no verified transaction volumes are public as of this writing.
What to expect by 2027
A few forecasts are safer than others.
Expect in-chat checkout to be normal in the US and patchy in the EU, held back by unresolved SCA questions. Expect at least two major protocol ecosystems to coexist rather than merge, forcing merchants to integrate with more than one. Expect disclosure rules to make it obvious when an agent is spending your money. And expect the novelty to wear off fast, because the best payment experience is the one you stop noticing.
The comparison to earlier tech shifts is useful. Just as people learned to check CPU and GPU compatibility before a big upgrade, covered in guides like this RTX 5090 vs 4090 compatibility check before upgrading, shoppers will learn to check which assistants support which merchants before trusting one with a card. The friction does not vanish. It moves.
Conclusion
Checkout inside the conversation is not a fantasy anymore. It is a live, funded, regulated reality that grew up fast between 2025 and 2027. The technology works well enough to be useful, the standards are real but unsettled, and the rules are still catching up. For shoppers, the smart move is to use in-chat buying where it saves genuine time, set hard spending limits, and confirm the seller before the money moves. For merchants, the smart move is to treat agentic checkout as a new channel with its own costs, not a free upgrade. The conversation is now the storefront. Whether it becomes a trustworthy one depends less on the chatbots and more on the boring plumbing nobody sees.
Frequently Asked Questions
What is agentic checkout in an AI chatbot?
Agentic checkout is a purchase completed by an AI agent on a user’s behalf, usually without leaving the chat. The user states what they want, the agent confirms details like size and shipping, and the payment is processed through a protocol connecting the merchant and the payment provider. OpenAI’s Instant Checkout and Google’s Buy for Me are the two best-known examples.
Is buying things inside ChatGPT or Google safe?
It can be, provided basic controls are in place. The card networks have built agent-specific tokens and registration systems specifically to limit exposure, and Google’s AP2 uses signed mandates to prove the user authorized the purchase. The main risks are overly broad spending permissions and unclear merchant identity, both of which users can manage by setting limits and asking who the seller is.
Will the ACP and AP2 payment protocols work together by 2027?
That is unknown and should not be assumed. The Agentic Commerce Protocol from OpenAI and Stripe and Google’s Agent Payments Protocol are separate standards with overlapping goals. No confirmed interoperability exists, and anyone claiming the two will merge by 2027 is guessing. Merchants should plan for integrating with more than one protocol.
Why is in-chat checkout harder to launch in Europe?
Strong customer authentication rules under the revised Payment Services Directive require two-factor verification for most electronic payments. How delegated, agent-initiated purchases satisfy those rules has not been settled by final regulatory guidance. Until it is, European rollouts will move more slowly and cautiously than US ones.
Do I still get buyer protections when an AI agent pays for me?
Protections depend on the merchant and the payment method, not on whether a bot pressed the button. Card network chargeback rights generally still apply, but disputes get harder when proof of purchase is scattered across chat logs, email, and wallet apps. Keeping all confirmations in one place is the simplest way to protect yourself.
What is the biggest mistake people make with in-chat purchases?
Authorizing an agent too broadly. Granting open-ended permission to “buy what I need” invites misinterpretation, from bulk orders to the wrong size. Setting a per-transaction cap and a monthly ceiling, then confirming the merchant name before payment, prevents most of the problems users report.
