Can OpenAI's dot shop at your Shopify store?

A customer asks their dot to find a replacement adapter, check that it fits and prepare an order. Can your store help it make the right choice?

OpenAI’s dot can interact with websites, and OpenAI explicitly documents purchases using a card saved on a merchant website. Those purchases need the user’s approval, which can be given in advance when it specifically covers the purchase. OpenAI’s dot safety FAQ establishes that capability. Whether a dot can complete your particular Shopify journey still needs testing.

For a merchant, the useful starting point is one product decision: can a visiting dot find the maintained facts, choose the exact variant and leave the buyer with a clear order to review? This guide draws on official documentation checked on 1 October 2026 UTC / 2 October in Brisbane. It offers a test plan, not a report of a completed dot shopping session.

What dot changes for a merchant

OpenAI introduced dots on 29 September 2026. A dot is an ongoing agent in ChatGPT with its own cloud computer and browser. A shopper can delegate work rather than ask only for a recommendation.

OpenAI’s “A day with your dot” showcase illustrates choosing a DoorDash meal from options with delivery times and totals, then approving the order. It is an official shopping example, not a demonstration of your Shopify checkout or its browser tools.

That makes the product detail between “looks suitable” and “this is the right item” worth investigating. A browser reaching your store, an answer using the correct specification and a completed purchase are three separate outcomes. Check each.

Availability is still limited. OpenAI’s setup documentation describes gradual access for eligible adult Pro users outside the EEA, UK and Switzerland, Business Premium users in supported regions, and an Enterprise beta that admins must enable. A launch announcement does not establish that your customers already use dots.

Keep finding your store separate from shopping on it

Shopify’s ChatGPT channel is a discovery-focused referrer using Shopify Catalog; shoppers finish on the online-store checkout. That documentation names ChatGPT. It does not establish which sources a dot uses to choose a merchant.

The primary sources reviewed for this guide do not establish a dedicated dot merchant-listing process or criteria for being featured. Treat those as unknowns. If your question is why a product does not appear in ChatGPT, use the ChatGPT product-discovery checklist.

For the test below, give the dot your store URL. This isolates what happens after arrival. It cannot tell you whether an unprompted shopping search would have found you.

Start with a read-only product question

Choose a question your catalog team can answer from a current source. A real support question about a model, connector, measurement or variant is a useful candidate. Save the reference before starting, including its product identity, units and date.

Here is an illustrative prompt, using fictional product names. Replace the URL and facts with your own:

Visit {store URL}. Compare AX-4 and AX-4B for a device that needs a USB-C input and an adapter under 200 g. Show the product evidence for both constraints and the exact model you recommend. If a fact is missing, say what is unknown. Do not add anything to a cart or buy anything.

Inspect the answer against the source. Did the dot use the correct model? Did it distinguish input from output? Was the weight for the adapter or the packaged shipment? Did it identify a missing fact instead of inferring it from a similar product name?

Keep the full response and any accessible page or tool evidence. A linked page is useful, but you still need to check that the quoted fact supports the conclusion. If you cannot inspect how the answer was obtained, record that limit rather than inventing a tool trace.

Separate access failures from answer failures

A dot’s cloud browser has its own sessions; it does not inherit the shopper’s personal-browser login. Some sites block cloud browsers or require verification. A user may need a private sign-in flow or browser takeover. Enterprise administrators can also restrict cloud browser access.

If the page never opens, investigate that access step first. A blocked visit says nothing about the quality of your compatibility data. If the shopper retries on an authorized connected computer, record it as a different setup: the cloud session does not transfer automatically.

Test the exact variant before checkout

Once the answer is supported, try a separate, explicitly authorized cart task. For example:

Add one AX-4 in the 1 m option to the cart. Show the exact variant, quantity and current cart price. Stop before checkout; do not place an order.

Compare the visible product selection and cart line with the requested item. A correct product family with the wrong length is still a failed selection. Inspect the current session’s price and availability rather than relying on an earlier answer.

Shopify already provides native WebMCP storefront tools on Liquid storefronts and the Hydrogen developer preview, without a merchant installation step. A compatible agent can get product details and option combinations, select a variant and work with the cart. Start with that native capability before adding another tool.

Do not assume a dot uses WebMCP because it has a browser. OpenAI’s Site tools documentation describes support in specified ChatGPT Work and Codex desktop-browser configurations. It does not establish support in every dot cloud session. Ordinary browser interaction and a website-tool call are different mechanisms. Record which you actually observe. The storefront tools guide explains the native tools and inspection boundaries.

Keep the buyer’s approval visible at checkout

OpenAI’s saved-card purchase capability requires approval; it does not mean that adding an item authorizes payment. Dot action controls still apply, including built-in safeguards and decisions that require the user. Instructions placed on your website do not grant the shopper’s permission.

Shopify also has a separate checkout WebMCP tool list. For an agent using that route, Shopify requires showing the current order and total and getting permission before calling complete_checkout. A changed total requires permission again. A saved payment method or payment-ready state is not that permission. Only a completed checkout confirms an order.

For an initial merchant readiness check, stop before submitting payment. Inspect whether the buyer can review the variant, quantity, delivery selection and total, and take over when required. If checkout tools are unavailable, Shopify documents handing the buyer back to the checkout page. A useful handoff is a valid result; it need not be an autonomous purchase.

Keep an actual purchase test separate, with explicit authorization and a controlled test setup. Passing the product and cart checks does not prove that discounts, subscriptions, customization or payment challenges work. Use the checkout feature audit for those scenarios.

Use a four-stage record to choose the fix

Record the date, dot/client and browser setup, store URL, market, sign-in state, exact prompt, reference and full result. Note whether the dot used its cloud browser or delegated work to a local Work or Codex task. Then identify the step that needs attention:

This is a diagnostic record, not a readiness score. If the evidence does not identify the cause, leave it unresolved and narrow the next test.

Your merchant preparation checklist

• Make the decisive facts explicit. Publish exact model identifiers, units, compatibility qualifications and variant differences in maintained product content. Keep useful diagrams, but add the facts needed to judge them in text. This is a recommendation, not a claim that dots cannot read images.
• Check the difficult option. Test a similar model or uncommon length alongside the obvious default; verify the cart line rather than only the recommendation.
• Keep policies consistent. Make delivery and returns conditions accessible and clear for the relevant market. Separate product suitability from permission to return or order.
• Preserve a human route. Check that shoppers can review the cart, complete sign-in or payment verification and finish on the normal checkout page.
• Repeat one bounded question after a change. Keep the same reference and prompt, record the new setup, and retain unchanged or worse results. One successful test is not evidence of universal support.

StoreKnows can help with the product-answer investigation. It runs simulated checks and can offer merchant-approved, read-only website tools built from a saved catalog copy on supported theme-based stores, for compatible visiting agents that choose them. Where a supported proposal exists, try your own question in its preview and inspect the product data behind the answer.

Those tools do not operate checkout, update Shopify Catalog, promise featuring or certify dot compatibility. Saved catalog data is not live inventory. If native tools already answer the important question well, keep that result. If the maintained fact is wrong, fix the source first.

See how the StoreKnows check works, then install StoreKnows if you have a product question worth investigating. Keep a simulated result labeled as such; it is not a recording of what a customer’s dot saw.

StoreKnows is independently developed and not affiliated with, endorsed by or sponsored by Shopify or OpenAI.