Skip to content

Shoppers ask AI about your returns and delivery before they buy. What should a D2C brand publish?

Shoppers ask AI about returns windows, delivery costs and warranties before they buy. How a D2C brand tests those answers and what its policy pages should say.

By Published 15 min read
A shopper with a phone asks an AI assistant, surrounded by a delivery van, a returned parcel and a policy page.

The short answer

A D2C brand should publish its returns, delivery and warranty terms as plain text on one page per policy, with the window, the cost, the countries covered and the date it changed, then make its product feed and marketplace listings say the same. Test policy questions on each engine every month, because AI answers quote old terms for months.

Key takeaways

  • Policy questions are buying questions: a shopper who asks an AI engine about your returns window or delivery cost is close to choosing, and a wrong answer can send them to a rival with clearer terms.
  • Adobe forecasts AI traffic to US retail sites up 130% year on year for 1 November to 31 December 2026, so holiday returns and delivery terms will be asked about more this season than last.
  • In Profound's data from 687 customers, the share of ChatGPT Shopping picks retrieved through integrated feeds rose from 8.26% to 61.54% on 10 July 2026, which makes the policy fields in a product feed as important as the policy page.
  • Each policy page should open with the facts an engine can quote: the window, who pays, the countries covered, the exceptions and the date the terms last changed.
  • Google applies return settings in Merchant Center or Search Console ahead of on-page markup, so a brand that updates only its page can still show the old terms in Google's shopping surfaces.

Why do policy questions matter more than product questions?

Policy questions matter because they come at the end of the decision, not the start. A shopper who asks an AI engine "can I return this if it doesn't fit?" or "how much is delivery to Australia?" has usually picked a product. The answer decides whether they buy it from you, from a marketplace or from a rival with clearer terms.

These questions also have checkable answers. Your returns window is 30 days or it is not. Delivery to India is free above a threshold or it is not. That makes them easy for an engine to answer and easy to get wrong when your pages are vague, inconsistent or out of date.

What the holiday data says about AI shoppers

AI-referred shopping is growing fastest in the season when policy questions peak. Adobe, which sells analytics software, forecasts that AI traffic to US retail sites will grow 130% year on year from 1 November to 31 December 2026. The forecast, released in September, draws on Adobe Analytics data covering more than one trillion visits to US retail sites.1

The same release includes a consumer survey of 5,000 people fielded in July 2026, with no geography stated. Among respondents who had used AI for online shopping, 69% said they were less likely to return an item.2 That is a self-reported figure, not a measured return rate. It still suggests shoppers use AI to settle doubts before they buy, and terms of sale are among those doubts.

Shoppers who use AI also check more places. In Gartner's survey of 1,004 US consumers in August 2026, shoppers using AI for gift inspiration consulted 4.48 sources on average, against 3.24 for those who did not.3 Each extra source is another chance for your terms to be stated differently.

AI shopping, 2026 holiday season

More AI visits, more checking

130%forecast growth in AI traffic to US retail sites, 1 Nov–31 Dec 2026, year on yearAdobe 1
69%of AI shoppers said they were less likely to return an item (survey, Jul 2026)Adobe 2
4.48sources consulted by US shoppers using AI for gift ideas, against 3.24 without (Aug 2026)Gartner 3
Sources: Adobe, 2026 US holiday shopping forecast, US retail sites, forecast for 1 Nov–31 Dec 2026; Adobe, consumer survey, 5,000 consumers, July 2026, geography not stated; Gartner, 1,004 US consumers, August 2026.

Research and comparison prompts carry policy conditions

Shoppers rarely ask about policy on its own; they fold it into the comparison. Profound, which sells AI visibility software, classified 200,000 commercial prompts from 20 to 26 December 2025. Product research made up 46.9% of the sample and product comparisons another 22%.4

A comparison prompt often carries a condition: "Which of these two jackets has free returns?" or "Who ships to the UAE before the end of the month?" The engine then needs a policy fact for each brand it names. If your fact is missing, the engine may skip you, guess, or quote a marketplace listing with different terms.

Some shoppers go further and ask the assistant to sort out a problem after purchase. In Gartner's survey of 3,566 B2B and B2C customers (February to March 2026, geography not stated), people were about three times as likely to use a third-party GenAI tool as a company's own chatbot for service.5 A customer asking ChatGPT how to return your product is reading your policy through ChatGPT, not your help centre.

Our view: treat policy prompts as a separate slice of your prompt set, not a footnote to product prompts. They have a different source mix, a different failure mode and a different owner inside the business.

Where do AI engines get your returns and delivery terms?

Engines piece together your terms from four places, and each can hold a different version. Your own policy pages are the first. Your product feed is the second, because feeds carry shipping and return fields. Marketplace listings are the third. Third-party pages, such as review sites, deal sites and old forum threads, are the fourth.

Your own pages are often the main source when the engine searches. Profound counted 11.84 billion citations on eight engines from 16 April to 16 July 2026, and about 57% pointed to company-owned websites.10 Search is not always triggered, though. Ahrefs, which sells SEO and AI visibility tools, found that 14.4% of free, logged-out ChatGPT responses in August and September 2026, across all countries, triggered a web search.11 When it does not, the engine answers from what it learned earlier, which may be last year's terms.

Wrong claims come from more than one direction. Profound's analysis of inaccurate AI claims about brands traced errors to brands' own content as well as competitors' pages and earned media, and noted that prices, offers and eligibility change often enough to go stale.6 Returns windows and delivery thresholds are exactly that kind of fact.

Where a policy answer is assembled

  1. Shopper's policy prompt
  2. Your policy pages
  3. Product feed fields
  4. Marketplace listings
  5. Third-party pages
  6. Answer, with or without a source
Illustrative. A model by the Sigzen AI editorial team; the highlighted sources are the two a brand controls directly.

The product feed now carries your policy too

For AI shopping surfaces, the feed can matter more than the page. Profound analysed ChatGPT Shopping prompts tracked by 687 of its customers between 1 July and 24 August 2026. The share of picks retrieved through integrated product feeds rose from 8.26% to 61.54% on 10 July.7

A feed is a list of products with fields, and several of those fields are policy: shipping cost, delivery time, return window and who pays for returns. Teams often fill them once at setup and never touch them again. When the returns policy changes, the page gets updated and the feed keeps the old value.

Google is explicit about which version wins in its own surfaces. Its documentation for return policy markup sets an order of precedence: return settings in the Content API for Shopping come first, then Merchant Center or Search Console settings, then product-level markup, then organisation-level markup.8 A brand that edits only the page and the markup can still show the old terms wherever Google uses the account settings.

Our view: name one owner for policy data across the page, the feed and the merchant accounts. In many D2C teams those sit with three different people, which is how a longer window on the site turns into a shorter one in a shopping answer. Our guide to the ChatGPT Shopping feed change covers who should own the feed itself.

What should a returns page say?

A returns page should open with the five facts a shopper asks about, in sentences, before any legal text. Start with the window: how many days, counted from delivery or from purchase. Say who pays for the return, and whether that differs by country. Say how the refund arrives and how long it takes. List the exceptions, such as sale items, personalised products or hygiene goods. End the block with the date the terms last changed.

Write each fact as a sentence an engine can lift whole. An invented example: "Returns are free within 30 days of delivery for orders shipped to Canada and Australia" is quotable.

Holiday terms need their own line. Many brands extend the returns window for gifts bought in November and December. If you do, state the purchase dates covered and the final return date, and remove the line when the season ends. An engine that finds an extended window with no end date may keep quoting it in March.

The five facts at the top of a returns page

  1. Window

    Days allowed, and whether they count from delivery or from purchase.

  2. Who pays

    Free returns, a fixed fee or the shopper's cost, by country.

  3. Refund

    How it is paid and how many days it takes after the item arrives.

  4. Exceptions

    Sale, personalised, hygiene or final-sale items, named plainly.

  5. Last changed

    The date the terms took effect, and any seasonal window with its end date.

Illustrative. A page structure from the Sigzen AI editorial team for D2C returns pages.

Delivery pages need the same discipline

Delivery questions are country questions, so the delivery page should be organised by country. For each market you ship to, state the cost, the free-delivery threshold if there is one, the typical delivery time, and whether duties and taxes are included at checkout. Cross-border shoppers ask about duties because a surprise charge on delivery can undo an otherwise good order.

Say where you do not ship. An engine asked "does this brand deliver to Saudi Arabia?" will often answer from nothing if your page is silent. A line that says you do not ship there yet saves a shopper a failed checkout and saves you a confused support ticket.

Warranty and guarantee pages close the loop

Warranty questions come later in ownership but feed back into buying. "Does this brand repair or replace a faulty product after a year?" is a comparison prompt in disguise. State the length of the warranty, what it covers, what it excludes and how to make a claim, again in text near the top.

Should you add structured data for return policies?

Yes for Google's shopping surfaces, but do not expect markup to fix AI answers on its own. Google supports MerchantReturnPolicy markup at the organisation level, as a store-wide standard, and at the offer level for products that differ.8 It also tells merchants who find markup hard to keep current to set return policies in Merchant Center or Search Console instead.

The evidence on markup and AI citations is thin. Ahrefs added JSON-LD schema to 1,885 already heavily cited pages between August 2025 and March 2026 and found no reliable lift in citations.9 That study was about citation counts, not policy accuracy, but it is a reason to treat markup as housekeeping rather than a growth lever.

Our view: keep the markup, the merchant settings and the visible page in step, and spend most of your effort on the visible text. The text is what every engine can read; the markup is what one company's surfaces use. Mismatches between them are worse than missing markup, because they give two sources two different answers.

A figure builds a policy page from three blocks marked with a returned parcel, a delivery van and a shield.
The parcel that comes back costs less than the sale that never happens because an engine quoted the wrong terms.

How do you test what each engine says about your policies?

You test by asking the engines the questions your shoppers and support team already hear, on a schedule. Pull the top policy questions from support tickets, chat logs and returns reasons. Turn them into a few dozen prompts per market, written the way shoppers write them, with the brand name in some and only the product type in others.

Rotate the conditions by country. A shopper in Canada may ask whether duties are included. One in Australia may ask about delivery times to regional areas. One in India may ask whether cash on delivery is available and how refunds work for it. One in the UAE may ask about delivery before a holiday weekend. Each prompt tests a different fact on a different page.

Score each answer on three things. Is your brand named where the prompt allows it? Is each policy fact right: the window, the cost, the countries, the exceptions? Which source did the engine cite, if any? Record the date, engine, country setting, full answer and sources for every run.

Two ways to read a policy answer

Mention check

  • Is the brand named?
  • Is it listed first or last?
  • One run per prompt
  • Ignores the facts stated

Fact check

  • Is each policy fact current?
  • Which source was cited?
  • Several runs per prompt
  • Each wrong fact traced to a page, feed or listing
Illustrative. A comparison by the Sigzen AI editorial team; policy prompts need the fact check.

Run each prompt more than once, because answers vary between runs. Our guide to checking your AI visibility yourself in an hour covers the method and its limits. When an engine repeats a wrong price or fee, our note on wrong prices that come from third-party sources shows how to trace it.

Marketplaces and third-party pages

A brand that sells on marketplaces has a second set of terms the engines can read. Marketplace returns policies often differ from your own, and that is fine when the listing says so. It becomes a problem when an engine quotes the marketplace's terms as yours, or yours as the marketplace's.

Make the difference explicit on your own page: "Orders placed on our site follow these terms. Orders placed through a marketplace follow that marketplace's returns policy." Then check that your marketplace listings carry the correct warranty and delivery details for the products you fulfil yourself.

Third-party pages are slower to fix. Deal sites, coupon pages and old review articles often repeat an offer or a returns window from years ago. Where you cannot, make sure your own page states the current terms with a date, so an engine that searches has a newer source to prefer.

Which teams own the fix?

Policy accuracy fails between teams. Customer service owns the wording, because it hears the questions. Ecommerce owns the feed and the merchant accounts. Marketing owns the page layout and the marketplace listings. Legal signs off on warranty language.

Agree one change process. When a returns window, a delivery threshold or a warranty term changes, the same ticket updates the page, the feed, the merchant account settings, the markup, the help centre article and the marketplace listings. That list is worth copying into your own runbook.

Where the term livesUsual ownerWhat to check after a change
Returns and delivery pagesMarketing, customer serviceFacts in text, date of change, countries listed
Product feedEcommerceReturn window, return cost, shipping fields
Merchant accountsEcommerceReturn and shipping settings match the page
Structured dataWeb teamOrganisation and offer-level policy match
Help centreCustomer serviceArticles and macros use the new terms
Marketplace listingsMarketingWarranty and delivery details for own stock

Illustrative. A change checklist from the Sigzen AI editorial team; adjust the owners to your team.

A 30-day plan before the holiday peak

A D2C team can do most of this in a month, and the holiday season is the month to do it. Adobe expects the biggest jumps in AI traffic to US retail sites on Thanksgiving and Black Friday.1 Shopify said on its Q2 2026 earnings call that AI-driven traffic and orders to its stores tripled year on year in that quarter.12 Policy answers are likely to be read more in the coming weeks than at any other time this year.

Thirty days to policy answers you can stand behind

  1. Pull policy questions from support; write a few dozen prompts per market; run them and record every wrong fact.

  2. Rewrite the returns, delivery and warranty pages with the facts in text at the top and a date of change.

  3. Update feed fields, merchant settings, markup, help centre articles and marketplace listings to match.

  4. Re-run the same prompts; trace any wrong fact that remains to its source and fix or report it.

Illustrative. A plan from the Sigzen AI editorial team; holiday terms get their own check in week 4.

Clear terms are some of the cheapest AI visibility work a D2C brand can do

Policy pages are short, factual and owned in-house, which makes them quicker to fix than almost anything else that shapes AI answers. A brand that states its returns, delivery and warranty terms in plain text, keeps its feed and listings in step and tests the answers each month gives the engines no reason to guess.

The payoff is in the last step of the purchase. Adobe's July 2026 data showed AI-referred visits to US retail sites converting 60% better than other traffic.13 Those are shoppers who arrive ready to buy. Our view: the brands that keep them are the ones whose terms the engine stated correctly before the click.

Sigzen AI includes policy prompts in every D2C audit for that reason. Our note on Adobe's holiday AI traffic forecast has a wider Q4 readiness list if you want to go further this season.

Sources

  1. Adobe, 2026 US holiday shopping forecast: Adobe Analytics, 1 trillion+ visits to US retail sites, forecast for 1 Nov–31 Dec 2026 (Sep 2026)
  2. Adobe, consumer survey in its 2026 holiday forecast: 5,000 consumers, fielded Jul 2026; geography not stated (Sep 2026)
  3. Gartner, survey of 1,004 US consumers, Aug 2026 (Oct 2026)
  4. Profound, holiday shopping prompts: 200,000 prompts, 20–26 Dec 2025; geography not stated (Jul 2026)
  5. Gartner, survey of 3,566 B2B and B2C customers, Feb–Mar 2026: third-party GenAI vs company chatbots; geography not stated (Jul 2026)
  6. Profound, where inaccurate AI claims come from (2026)
  7. Profound, ChatGPT Shopping feed retrieval: 687 customers, 1 Jul–24 Aug 2026 (Sep 2026)
  8. Google Search Central, Merchant return policy (MerchantReturnPolicy) structured data (seen Nov 2026)
  9. Ahrefs, schema and AI citations test: 1,885 already heavily cited pages, Aug 2025–Mar 2026 (May 2026)
  10. Profound, where AI citations come from: 11.84B citations, 8 engines, 16 Apr–16 Jul 2026 (Jul 2026)
  11. Ahrefs, ChatGPT query fan-out study: 66.6M free, logged-out GPT-5.6 Luna responses, Aug–Sep 2026, all countries; factual subset 20,000 US prompts; citation ranks US, Mar–Sep 2026 (Oct 2026)
  12. Shopify, Q2 2026 earnings call, 5 Aug 2026: company-reported figures for Shopify stores, Q2 2026 (Aug 2026)
  13. Adobe, U.S. consumers are embracing LLMs to make travel plans (retail and travel data): US retail sites, more than 1 trillion visits, July 2026 (Aug 2026)

Questions readers ask

  • ChatGPT quotes an old returns window when it answers from what it learned before the change, or when a source it reads still carries the old terms. Product feed fields, marketplace listings, deal sites and help centre articles are the usual culprits. Update every place the term lives, put the date of change on the policy page and re-test the same prompts a month later.

Keep reading

Free. Enter your domain and Sigzen AI checks how five engines answer your buyers' questions, against up to three competitors.

No account needed. Two minutes to submit; we email the full five-engine report when the check has run.