For years, a store won a sale in a simple order: the shopper typed a search, scanned ten links, opened a few tabs and bought from one of them. That order is changing. AI assistants now read the web for the shopper, compare the options and answer with a short list. Since late 2025, some of them can also complete the purchase.
This is agentic commerce: software agents that search, compare and buy on behalf of a person. This post explains what changed, what the agents need from a store, and what you can check today.
What changed in 2025
Three public announcements and one study mark the shift.
Checkout inside the answer
On 29 September 2025, OpenAI introduced Instant Checkout in ChatGPT. Shoppers in the United States can buy from Etsy sellers inside the chat, and OpenAI named Shopify merchants as the next group. The checkout runs on the Agentic Commerce Protocol, which OpenAI built with Stripe and released as an open standard, so other merchants and payment providers can connect to it.
Agents that can pay
On 16 September 2025, Google Cloud announced the Agent Payments Protocol (AP2), an open protocol for payments that an AI agent starts for a user. AP2 records the user's instruction and the approved cart as signed "mandates", so a merchant and a payment provider can prove what the person asked for and what they approved. Google says more than 60 organizations work on it, among them Mastercard, American Express, PayPal, Adyen, Etsy and Salesforce.
Fewer clicks when the AI answers
The shopper's path also changes before any agent pays. In a Pew Research Center study of the Google searches of 900 US adults in March 2025, users clicked a traditional result on 8% of visits when the page showed an AI summary, and on 15% of visits when it did not. They clicked a link inside the summary itself on only 1% of those visits.
More visits from AI tools
At the same time, AI tools send a growing number of shoppers to stores. Adobe Analytics reported that visits to US retail sites from generative AI tools grew 693% in the 2025 holiday season (1 November to 31 December) compared with the year before. The base was small, but the direction is clear.
How an agent picks a product
An agent cannot look at your photos the way a person does. It reads text, structured data and prices, and it compares them with other sources. Google describes how its AI Mode handles a shopping question: it "runs several simultaneous searches to figure out what makes a bag good for rainy weather and long journeys", and then uses those criteria to suggest products.
So the agent asks two questions about your product:
- Does the page state the facts that decide the purchase in this category (material, size, fit, compatibility, care, return terms)?
- Can the agent trust the page: does the price match the feed, is the product in stock, and what do buyers say?
A product that leaves these facts out is not wrong. It is invisible to the comparison, and the agent names a product that states them.
What a store needs to be picked
The work is not new, but the reader is. Four areas decide most of it.
- Product facts in plain text. State the attributes of the category on the page itself, in the HTML, not only in images or in a script that runs after the page loads. Our guide How to write a product page that AI answers can quote lists the checks.
- Structured data that matches the page. Product markup with price, availability, brand and identifiers helps machines read the offer without guessing.
- One price everywhere. The page, the feed and the markup must show the same price. A mismatch is a reason for an agent to trust another offer.
- Reviews you answer. Agents read public reviews. A store that replies to complaints shows that it resolves problems.
| What the agent needs | Where it reads it | What you can check |
|---|---|---|
| The facts of the category | Product page text | Material, size, fit, compatibility stated in words |
| A price it can trust | Page, feed, structured data | The same price in all three |
| Proof from buyers | Review sites and the page | Recent reviews and replies to complaints |
| Access | robots.txt | Answer bots such as OAI-SearchBot are allowed |
What does not change
Agentic commerce does not replace the basics of selling online. It raises the cost of skipping them.
- A good product still wins. An agent compares facts. When your product is the better fit for the need, a page that states why gives the agent a reason to name it.
- Search still matters. Google's AI features and ChatGPT's search answers read pages that their crawlers can reach and understand. The work of making a page clear for search is the same work that makes it clear for an agent.
- Trust still decides. A shopper who reads an AI answer still wants proof before they pay: reviews, return terms, delivery times. An agent that acts for them looks for the same proof.
What changes is how much a gap costs. On a page of ten links, a weak product page still got a chance to be clicked. In a short AI answer, a product that the agent cannot compare is left out.
Questions to ask your platform and your team
Before you plan any work, find out where you stand.
- Does our platform let us take part in agent checkouts as they open to more merchants, and what does it need from our product data?
- Which of our product attributes live only in images, tabs or scripts?
- Who owns the product feed, and who owns the product pages? Do they show the same price?
- Who reads and answers our public reviews, and how often?
The answers show whether the gaps are in your data, in your pages or in your process.
What to do this quarter
You do not need a new platform to prepare. Start with the products that sell most, and work down the list.
- Read three of your top product pages as a stranger would. Note every question a buyer asks your support team that the page does not answer.
- Check that your robots.txt does not block the crawlers of the answer engines you want to appear in.
- Compare the price on the page with the price in your product feed for the same products.
- Read your latest one- and two-star reviews and reply to the ones you can resolve.
How ecommerce.com helps
ecommerce.com reads your store the way these agents do, from public data only. Enter your store address and, in about two minutes, every product gets an SEO score and a GEO score with the checks behind them. You also see where you rank on real Google searches, who Google's AI names, which pages are dead, and what buyers say. The fixes that add the most come first.
Questions and answers
What is agentic commerce?
Agentic commerce is shopping that AI agents do for a person: they search, compare products and, where the platform allows it, complete the purchase.
Can ChatGPT buy products today?
Yes, in some cases. OpenAI introduced Instant Checkout in ChatGPT on 29 September 2025 for US shoppers, starting with Etsy sellers, with Shopify merchants named as next.
What should a store fix first?
Start with the product facts of your best sellers: state them in plain text on the page, keep the price the same in the page, the feed and the markup, and reply to recent complaints.
