Shopping Agents Are Now Judged on Return Rates — and Who Gets Cited When They Explain Why

A quiet shift landed in agentic shopping pilots this quarter: the evaluation metrics moved past conversion and started including post-purchase outcomes. Assistants that recommend products are increasingly measured on whether those recommendations get returned, disputed, or refunded — not only on whether they produced a click.

That change matters for merchants because it alters which parts of your site become citation sources. When an agent is optimising for a clean outcome, a retailer's *returns page* stops being a legal formality and becomes a decision input: it is where the assistant verifies the return window, the condition requirements, the restocking fee, and who pays return shipping.

Coverage of the latest agentic commerce experiments describes a scoring model that combines several signals: whether the stated specification matched the delivered product, how often buyers initiated a return, and how frequently support disputes referenced a mismatch between what the assistant said and what the merchant's policy actually allowed. None of these signals are new to merchants — but they were previously invisible to the recommendation layer.

The practical consequence is that vague policy language now carries a cost. A returns page that says "returns accepted within a reasonable period" gives an agent nothing to verify. An explicit page — 30 days from delivery, unopened items only, buyer pays return shipping, refunds issued within 5 business days — gives the agent a fact it can state, cite, and stand behind.

Most stores treat post-purchase documentation as a compliance artefact written for humans in a dispute, not as structured data for machines. The returns policy lives in a footer link, the warranty terms live in a PDF, and the shipping SLA lives in a support article. An assistant assembling an answer has to reconcile three inconsistent sources — or decline to answer confidently.

This is the same consistency problem merchants have been solving on product pages, now moving one step later in the funnel. The stores that already synchronised price, stock, and shipping across page, feed, and schema have a head start, because the same discipline extends to returns and warranty with almost no new infrastructure.

Start by stating each post-purchase fact exactly once, in one canonical wording, then mirroring it across the returns page, the product page summary, the FAQ, and any structured data you publish. Next, add the questions buyers actually ask — "can I return an opened item?", "who pays return shipping?", "how long does a refund take?" — as direct Q&A blocks. Finally, check that your shipping, returns, and warranty statements do not contradict each other across channels, because contradiction is the fastest route to being silently skipped.

Post-purchase facts are exactly the kind of content no one audits, because they never appear in a conversion dashboard. Visora reads your store the way an assistant does — across product pages, policy text, and structured data — and points out which of those facts are missing, vague, or in conflict. A free scan at /audit will show you how much of your returns and warranty language an agent can actually verify.