Which Shipping, Returns and Inventory Signals Do AI Agents Check, and How Should You Publish Them?
AI shopping agents check whether a product is in stock, what shipping costs, how long delivery takes, whether returns are accepted and for how long, and what the store's policies say. They read these facts from product feeds, structured data and policy pages. OpenAI says that when several merchants sell the same product, ChatGPT considers availability, price and quality, among other factors. Google's Store Quality program scores merchants on shipping and return experience. So publish these facts as explicit values in your feeds and markup, and as plain text on product and policy pages, with every source agreeing. An agent that cannot find your delivery time or return window cannot compare you fairly. It is likely to pick a store that states them.
Which signals do AI shopping systems use?
| Signal | Where systems read it | What they publish about it |
|---|---|---|
| Availability | Feed availability, markup availability, product page |
OpenAI's feed requires it: in stock, out of stock, pre-order, backorder or unknown |
| Price and shipping cost | Feed price and shipping_price, markup, checkout |
OpenAI: an omitted shipping price means unknown, not free |
| Delivery time | Markup shippingDetails with handling and transit time; Merchant Center shipping settings |
Google scores delivery time and shipping cost in its Store Quality program |
| Return acceptance and window | Feed accepts_returns and return_deadline_in_days; markup hasMerchantReturnPolicy; Merchant Center return policies |
Google scores return experience; OpenAI separates acceptance from the window and policy URL |
| Return costs | Markup returnFees; Merchant Center settings |
Google distinguishes free returns, customer-paid and return shipping fees |
| Seller identity | Feed seller_name; site |
OpenAI weighs whether a merchant is the primary seller or maker |
None of the companies publishes exact weights. What they do say is that these facts are inputs, and that missing values are treated as unknown.
Which missing pages stop AI assistants from understanding our delivery regions and cutoff times?
The most common gaps are these:
- No shipping page with a regions table. A banner saying "fast delivery" is not a fact. List each region or country you ship to, with services, prices and typical transit times.
- No cutoff time. "Orders placed by 2pm Monday to Friday ship the same day" answers "can I get it by Thursday?". Without it, an agent cannot estimate delivery.
- No page for where you do not ship. Excluded regions, such as islands, PO boxes and certain countries, are as important as covered ones. They prevent wrong recommendations.
- Holiday and peak-season delivery dates buried in a pop-up or a PDF.
- Delivery information only at checkout. If the shipping estimate appears only after an address is entered, crawlers never see it.
- No per-product exceptions. Oversized, made-to-order or hazardous items often have different delivery times. Say so on those product pages.
Give each of these a stable URL, link to them from product pages and the footer, and keep them as HTML text, not images or scripts.
How to structure shipping and return policies for AI readability
Write policies as facts first and legal text second:
- Open with a summary table. Region, service, cost, delivery time and cutoff for shipping. For returns: window, condition, who pays, how refunds are made and how long they take.
- Use one heading per question: "How long do I have to return an item?", "Who pays for return shipping?", "Which items can't be returned?".
- State numbers, not adjectives: "30 days from delivery", not "a generous window".
- List exceptions explicitly, such as personalised or made-to-order goods, and sealed hygiene items or software once the seal is broken. In the UK and EU, online shoppers keep a 14-day right to cancel that a "final sale" label does not remove, so list only the exceptions the law allows there.
- Date the policy's last change in the text itself ("Policy in effect from 1 March 2026"), so outdated copies are recognisable.
- Keep one version. Remove old policy pages or redirect them, and make sure help articles and product pages quote the current terms.
Then mirror the same values in structured data. Google's merchant listing markup supports shippingDetails, where delivery time is handling time plus transit time in days. It also supports hasMerchantReturnPolicy, with return fees described as free returns, customer responsibility or specific shipping fees. Markup must match the visible page.
Can better return terms make an AI assistant recommend my store?
They can help, but not on their own. Better terms matter in two ways:
- When an answer compares merchants for the same product. Return terms are one of the facts an agent can set side by side. A store that states "free returns within 60 days" gives a clear advantage. A store that states nothing gives the agent nothing to weigh.
- When a shopper asks about returns directly, as in "where can I buy [product] with free returns?". A clearly stated policy is the only way to qualify.
Google's Store Quality program includes a return experience category. Its Top Quality Store badge requires an "Exceptional" overall score, based partly on shipping and return practices. Google notes that policy changes can take up to 30 days to show in store quality scores.
What better terms will not do is overcome weak fundamentals. An out-of-stock product, an uncompetitive price or poor reviews usually matter more. Improve return terms where they are genuinely worse than competitors', and then state them clearly.
Do AI agents use delivery dates more than price when selecting merchants?
No published evidence says either one outweighs the other. OpenAI's September 2025 Instant Checkout announcement said merchant ranking for the same product considered availability, price, quality, whether the merchant is the primary seller, and whether Instant Checkout was enabled. OpenAI has since moved away from standalone Instant Checkout, and purchases now complete on merchants' own sites, so the last factor no longer applies. Neither version mentioned delivery speed as a separate factor. Google scores delivery time and shipping cost together in its shipping category.
Shopper behaviour suggests both count. Among US shoppers who abandoned a cart for reasons other than just browsing, Baymard found 40% did so because extra costs such as shipping were too high. Another 20% said delivery was too slow. Cost weighs more often, but speed is far from trivial.
The practical conclusion: agents can only weigh what they can see. Publish delivery estimates as explicit values, and include shipping cost so the total price is clear. If you compete on speed, make the speed a fact in your feed and markup, not a slogan.
How should I publish warranty terms for AI product comparisons?
- Put the key terms on the product page as text: length, what is covered, what is excluded, who provides it (you or the manufacturer), and how to claim.
- Link to the full warranty document as HTML or a text-based PDF, not a scanned image.
- Distinguish your warranty from legal rights. In the EU, sellers are liable for faults that appear within two years of delivery under the Sale of Goods Directive. A "one-year warranty" should not imply shoppers have less.
- Meet pre-sale rules. In the US, the FTC's Pre-Sale Availability Rule requires that written warranties on consumer products costing more than $15 be available before purchase. Online sellers can meet this by showing the warranty or linking to it clearly.
- Use consistent wording across products, so comparisons line up: "2-year manufacturer warranty, parts and labour".
- Add it to structured data if you use the
warrantyproperty, matching the page.
Assistants comparing products often mention warranty length when it is stated clearly, because it is an easy fact to compare.
What return-policy fields should be included in an AI product feed?
For ChatGPT, OpenAI's product feed specification includes:
accepts_returns: true or false. Send false for final-sale items.return_deadline_in_days(also accepted asreturn_window): the number of days allowed, as a duration. Supply it only when returns are accepted.return_policy: the URL of the public policy that applies to the item.
The specification says acceptance must be stated separately. A policy URL alone does not set it, and a window alone does not enable returns.
For Google, return policies are set in Merchant Center under delivery and returns, at account level or for specific products. They cover the window, return methods, fees and restocking fees. Google uses them to show return information in listings and to assess store quality.
Also keep the shipping fields complete: shipping_price or the shipping settings, and availability on every row. Update them when stock changes. How to handle products that sell out is covered in how retailers should handle AI recommendations for out-of-stock products.
How do you check that every source agrees?
| Check | Pass condition |
|---|---|
| Product page | Stock status, delivery estimate, shipping cost and return window visible as text |
| Structured data | Same values in offers, shippingDetails and hasMerchantReturnPolicy |
| OpenAI feed | availability, shipping and return fields filled for every row |
| Merchant Center | Shipping and return policies set and approved |
| Policy pages | Summary tables with the same numbers |
| AI answers | Assistants state your delivery time and return window correctly when asked |
When the sources disagree, answers pick one, and it may be the wrong one. The same applies when a reseller's listing states your products' terms differently, as covered in why AI recommends a reseller instead of the manufacturer.
What might an illustrative case look like?
Consider an illustrative homeware store with next-day delivery and 60-day free returns, stronger terms than most competitors. Its product pages said only "fast shipping, easy returns". Its feed left the return fields empty, and its delivery estimate appeared only at checkout. ChatGPT answers comparing where to buy its best-selling lamp described its returns as "unclear". The store added a shipping table with a 2pm cutoff, filled the return fields in its feeds, and added return policy markup. It also put "Order by 2pm for next-day delivery. Free returns within 60 days." on every product page. Within six weeks, comparison answers listed its delivery and return terms alongside competitors'.
What do people ask about shipping and returns signals?
Should free shipping be stated as zero in the feed?
Yes, state it explicitly as zero, because an omitted shipping price is treated as unknown, not free.
Do agents read our returns page or the feed?
Both, depending on the system. Keep them identical.
How often should inventory be updated?
As often as stock changes. Stale availability leads to recommendations for products you cannot sell.
Do we need different policies for different countries?
If terms differ, yes. Publish them per country and set them per country in Merchant Center.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources