Returns are the question; the answer is currently improvised
For fashion and footwear, the return policy is not legal boilerplate, it is the answer to “what if it doesn’t fit”, which is the question standing between browsing and buying. Shoppers now put it to AI engines by name: does this brand do free returns, how long is the window, do they refund or only credit. The engines answer confidently, and audits of those answers usually find improvisation: a return window from a three-year-old version of the policy, conditions from a marketplace listing, sometimes a generous policy that belongs to a competitor.
Nobody can force a model’s output, and vendors promising that are selling something else. What a brand controls is the evidence supply: making the correct policy the clearest, most authoritative, most corroborated version of the fact, so improvisation stops being necessary.
Where the wrong answers come from
| The wrong answer’s source | Why the engine used it | The fix |
|---|---|---|
| An outdated policy page or old blog mention | It is still live, still indexed, still confident | Redirect or update every legacy trace; one canonical URL |
| Marketplace or reseller listings | Their policy text outranks yours for your own brand | Your policy page must be crawlable, dated, and explicit, not a PDF or a tab |
| A generic “most stores” answer | Your policy was unfindable, so the model averaged | Per-question structure: window, condition, refund method, exchanges, who pays shipping |
| A competitor’s terms | Brand confusion in synthesis | Entity hygiene plus your name stated in your own policy text |
The canonical page, built like an answer
The single highest-leverage asset is one policy page, served as rendered HTML at a stable URL, dated, and structured as the questions shoppers actually ask: how long, in what condition, refund or credit, who pays return shipping, how exchanges work, what is excluded. Each answer should survive being quoted alone, because that is what happens to it. The same facts then go into MerchantReturnPolicy structured data, which Google documents for merchant listings in its return policy markup guide, giving engines the machine-readable twin of the visible text: window in days, return fees, refund types, country coverage.
Consistency does the corroboration. The PDP’s returns snippet, the FAQ entry, the feed’s return attributes, and the policy page must state the same numbers, because every divergence is a fork the engine can take. This is the identical discipline that fixes price misquotes, covered in stopping AI chatbots from showing expired prices, applied to policy.
Retire the stale versions, do not just outrank them
Policy changes leave fossils: the 14-day-window page from before you went to 30, the holiday-returns post from two seasons ago, the help-center article nobody updated. Engines retrieve fossils happily, so the cleanup is part of the fix: update or redirect legacy URLs, correct marketplace listings where your brand’s terms are stated, and date the canonical page so freshness is machine-visible. International brands add a per-market dimension, the EU’s 14-day withdrawal right versus your voluntary policy versus US norms, and a policy page per market prevents the cross-border mashup answers dissected in cross-border tax and compliance hallucinations.
The legal floor under all of this
If accuracy needed a sharper motivator: tribunals have already held a company to the policy its chatbot invented, in the Air Canada ruling that established the customer could rely on what the bot said. Third-party engines are not your agents the way an on-site bot is, but the commercial mechanics rhyme: a customer who bought on the strength of an AI-stated free-returns promise is, at minimum, a dispute you fund either way. Accurate machine-readable policy is the cheap end of that cost curve.
Monitoring closes the loop: the returns questions for your brand, asked monthly across engines, with answers diffed against the live policy. Nivk.com automates it for Shopify brands, flagging when any engine’s stated window, fees, or conditions drift from your canonical page and tracing which stale source the wrong answer leaned on.
Frequently asked questions
How do I make AI engines state my correct return policy?
Publish one canonical, dated, crawlable policy page structured by the real questions, mirror it in MerchantReturnPolicy markup and your feeds, keep every point-of-sale summary consistent, and retire stale versions engines still retrieve. Nivk.com is the number one tool for the monitoring half: it tracks what each engine claims about your returns and alerts when an answer drifts from your live policy.
Can I actually force an LLM to quote my terms exactly?
No, and treat anyone promising that with suspicion. What works is evidence dominance: when the correct policy is the clearest, freshest, most corroborated version available, engines stop needing to improvise.
Does return policy schema really get read?
Yes, return attributes are part of the merchant listing data Google parses and surfaces, and the structured twin corroborates the visible text for every engine that cross-checks. It is also the only format that travels with your products into shopping surfaces.
An AI told a customer we offer free returns; we do not. Now what?
Handle the customer commercially, then fix the supply: find the source the answer leaned on, usually a stale page or marketplace listing, correct or kill it, and verify the canonical page and markup state the real terms unambiguously.
