What Should Businesses Do to Prepare for Agentic Commerce?

What is agentic commerce, and where does it stand?

Agentic commerce means AI assistants that research, compare and increasingly buy on a shopper's behalf. The main developments so far:

Two models are emerging. In one, the assistant handles discovery and sends the shopper to your checkout. In the other, the assistant completes the purchase itself through a protocol connected to your systems. Both depend on the assistant understanding your products accurately.

How can businesses prepare for AI shopping search?

1. Get your product data into the feeds assistants use. Keep Google Merchant Center complete and current. Apply to share your feed with OpenAI through its merchant programme. Where your platform offers AI shopping channels, switch them on.
2. Make each product unambiguous: a title naming the product and variant, brand, GTIN where one exists, and key attributes such as size, material, colour and compatibility. OpenAI's feed specification asks for one row per purchasable variant, with titles up to 150 characters.
3. State availability explicitly on every row, and update it when stock changes.
4. Publish shipping and return terms as values, not slogans. OpenAI's specification treats a missing shipping price as unknown, not free.
5. Keep product pages readable without JavaScript, with prices, specifications and policies in the HTML.
6. Build genuine reviews by asking every customer for an honest review in their own words, on your own site and on platforms that allow review requests.
7. Keep prices consistent between your feed, your page and your checkout.

Which operational signals agents weigh is covered in which shipping, returns and inventory signals AI agents check, and how to publish them.

What should a retailer fix first to become ready for AI assistant checkout?

In order of impact:

1. Accurate stock and price in real time. An agent that buys something you cannot ship, or at a price you no longer honour, creates a failed order and an unhappy customer. Feed updates should follow stock changes within minutes or hours, not days.
2. Complete shipping and returns data, so the agent can state total cost and delivery time before buying.
3. A checkout without friction: guest checkout, standard payment methods including digital wallets, no fees that appear only at the last step, and clear error messages. Baymard's research found 40% of US shoppers who abandoned a cart for reasons other than browsing did so because of unexpected extra costs.
4. Your platform's agentic features. If you use Shopify or another major platform, its integrations with AI assistants will usually reach you before you could build your own. Use them.
5. Post-purchase pages: order status, returns and support pages that state the process plainly, since agents will increasingly handle "where is my order?" and "how do I return this?".
6. Fraud and bot rules that allow legitimate agents. Card networks are introducing tokens that identify agent-initiated payments. Review with your payment provider how agent transactions are recognised, so real orders are not blocked as bots.

A checkout that works well for people usually works well for agents too. Agents simply expose every inconsistency faster. How to prepare a catalogue for ChatGPT specifically is covered in how to prepare an ecommerce catalog for Buy it in ChatGPT.

Does agentic commerce differ for service businesses?

The same principles apply to bookings. Assistants already book restaurant tables and are connecting to home services marketplaces. For services, the equivalent of a product feed is a booking system that shows real availability, with prices or price ranges, service areas and cancellation terms stated plainly. If your bookings run through a platform that assistants integrate with, keep that profile complete and your availability current. If they run through your own site, make sure the booking form works without scripts that block automated but legitimate use, and that confirmation and cancellation steps are clear.

What can wait?

• Building your own shopping agent. Assistants' own agents and your platform's integrations will reach shoppers first.
• Custom protocol integrations, unless you are a large retailer invited into a programme. Most merchants will connect through Merchant Center, a platform app or a payment provider.
• An MCP server for your store. It may become useful, but it is not where most agent traffic starts today, as covered in whether websites need MCP to be ready for AI agents.
• Exclusive deals with one assistant. Keep your data open to all the main channels while the market settles.

How do you measure readiness and results?

• Feed diagnostics: errors and warnings in Merchant Center and any other feed programme you join.
• AI-referred traffic and orders by source, with conversion rates compared with other channels. Adobe reported that AI-referred visits to US retail sites converted 60% better than non-AI traffic in July 2026, up from 42% in March.
• Accuracy checks: ask assistants about your products' price, stock, delivery and returns, and compare with reality.
• Test purchases through any agentic checkout you are enabled for, including returns.
• Failed or disputed agent orders, tracked separately so you can see where data or checkout breaks down.

What might an illustrative case look like?

Consider an illustrative homeware retailer that wanted to "get ready for AI shopping" and was quoted for a custom shopping agent. An audit first found that its feed updated stock once a day, that its shipping prices appeared only at checkout, and that 15% of products lacked GTINs. The retailer moved to hourly feed updates, published shipping and return terms as feed values and on product pages, and added missing identifiers. It applied to OpenAI's merchant programme and enabled its platform's AI channel integrations. Over the next quarter, AI-referred orders grew, and the questions assistants got wrong about its products fell. It shelved the custom agent.

What do people ask about agentic commerce?

Will AI agents replace my website?

Not soon. For now, most assistants still send shoppers to merchants' sites to buy, and your pages remain the source of product facts.

Do I need to support every protocol?

No. Connect through the channels your platform and payment provider support, and keep your data accurate everywhere.

Is agentic checkout safe for customers?

The payment networks and assistants are building in identity and authorisation controls. As a merchant, make sure your policies and support work for orders placed by agents.

How soon should a small shop act?

Now, on the basics: feeds, product data, policies and checkout. They pay off with or without agents.