A buyer asks an assistant, "Find me a running shoe in my size that arrives before Friday." The old search flow gives the buyer a page of ads and links. The assistant's job is different: narrow the options, check the details, and help the buyer finish the purchase.
I think this changes what a store must publish on the web. Buying an ad may still win attention. It will not help much if the assistant cannot tell whether the shoe is in stock, what delivery costs, or where checkout actually starts.
This is a prediction about how buying may evolve, not a claim that ad clicks have already stopped. Search and ads remain large businesses. The practical question for a developer is what information an agent needs to complete a real shopping task.
Start with the product record
For each variant, make the machine-readable record clear: product ID, size, color, price, currency, availability, seller, and a stable product URL. A shoe listed only as "sizes available" makes the assistant guess. A price without shipping terms can become a bad recommendation at checkout.
If the catalogue changes often, define when the record was last updated and give the agent a way to check live stock before promising anything. An index is useful for discovery; the live store remains the authority for a transaction.
Expose the actions, not just the catalogue
A useful flow may need these operations:
- Search products against the buyer's size and delivery constraints.
- Quote the current price, tax, shipping, and estimated delivery for a specific destination.
- Create or resume a cart with the exact selected variant.
- Hand the buyer into a secure checkout, with the final total visible.
- Return an order identifier and status only after the seller confirms the order.
Keep authentication and payment decisions with the appropriate parties. An assistant should not infer consent to buy from a request to compare shoes. If the quote changes, return the change before the buyer commits. If inventory disappears, fail clearly rather than swapping in another size.
OpenAI's Agentic Commerce Protocol documentation describes product discovery and checkout integrations for merchants. Google's Merchant API documents checkout settings, including how merchants point buyers to a checkout destination. These are examples of infrastructure moving toward agent-assisted commerce; neither source proves that every merchant can use every flow today.
Measure whether the handoff works
I would test this with tasks, not page views alone. Give a test agent an exact request: "Show two black running shoes, size 9, under a set budget, deliverable to this postcode by Thursday." Then inspect its answers against the store's live catalogue and checkout.
Where did it fail? Was the variant ambiguous? Was shipping calculated only after an address? Did checkout change the total? Did an out-of-stock item linger in the product feed? Those failures matter even if a human browsing the site could work around them.
This is also why I would not remove the human site. People still want to browse, compare photos, read returns policies, and decide whether to trust a seller. The agent-facing path should complement that experience, not hide a weak store behind an API.
What happens to the ad click?
Today, a retailer often pays to get a person from a search page or feed to its store. If more buyers start by asking an assistant, some of that discovery and comparison moves into the assistant. Ads may move there too, or sellers may pay for outcomes instead of traffic. Which model wins is still open.
The near-term work is less speculative: publish accurate product data, expose reliable actions, make the final cost clear, and verify the handoff. A store that solves those problems has a better chance of serving both the shopper who clicks and the shopper who asks an agent to do the search.
By Tej Pandya, founder of GrowEasy.ai.
Sources: OpenAI, "Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol" (https://openai.com/index/buy-it-in-chatgpt/); OpenAI, Agentic Checkout Spec (https://developers.openai.com/commerce/specs/checkout); Google, Merchant API checkout settings (https://developers.google.com/merchant/api/guides/products/checkout-settings).
Top comments (0)