AEO agent
Does your store answer the question a shopper actually asked?
GEO and AEO get used interchangeably and no agreed definition tells them apart, so this is the job we mean by AEO: whether your store contains, and exposes, the answers a shopper asks for before they buy.
It checks whether your returns and shipping terms are stated somewhere an engine can read them, whether you have real support content, and whether your product copy answers questions instead of describing a mood.
The strongest single signal it looks for is specific numbers. A physical product described without one measurement, weight, duration or quantity gives an answer engine nothing to quote, and that is the flag worth acting on first.
Every check this agent runs
Read from the checklist the scan executes, not written for this page. A check added to the agent appears here with no marketing edit.
| Check | Reads | Category | Severity |
|---|---|---|---|
| returns_policy The store publishes no substantive returns policyReturns terms are what a shopping assistant needs to answer "can I send it back" before a purchase. OpenAI's Agentic Commerce Protocol feed spec lists return_policy as a recommended field and requires seller policies when checkout is enabled, and Google consumes MerchantReturnPolicy markup for its shopping surfaces. A two-word placeholder answers none of it, which is why presence alone does not pass this check. | Your storefront | Answers | high |
| shipping_policy The store publishes no substantive shipping policy"When does it arrive" is asked before almost every purchase, and the shipping policy is the store-level text that can answer it. OpenAI's Agentic Commerce Protocol requires seller policies when checkout is enabled, and Google consumes shipping markup for its shopping surfaces. A policy too short to state a delivery window answers nothing, which is why presence alone does not pass. | Your storefront | Answers | high |
| stated_returns_window The returns policy never states how long a buyer has"Can I send it back, and for how long" is the question returns terms exist to answer, and the window is the half a length check cannot see: a policy long enough to pass the substance floor can still name no deadline. Stated as a figure, the window is a fact an assistant can repeat exactly; left unstated, there is nothing exact to pass on. | Your storefront | Answers | medium |
| stated_delivery_window The shipping policy never states when orders arrive"When does it arrive" is asked before almost every purchase, and a shipping policy can clear the substance floor on carrier prose alone without ever stating a timeframe. A window stated as a figure, per region if they differ, is the one sentence that answers the question; without it, an assistant reading the policy has nothing to quote. | Your storefront | Answers | medium |
| policy_markup The returns and shipping terms are not published as structured dataGoogle's shopping surfaces consume MerchantReturnPolicy and OfferShippingDetails for the return and shipping details shown beside product listings, which makes this one of the few places structured data has a documented effect. It is one of two routes, not the only one: Merchant Center account settings can supply the same information, so absent markup is a gap in one route rather than a disqualification. What the markup adds is that the terms are readable at the page itself, beside the offer they govern. | Your storefront | Answers | medium |
| support_content No page on the store answers support questionsA support page is where a question meets its answer in one liftable block. The evidence for this factor is correlational and we say so: Semrush measured question-and-answer format as a positive correlate of appearing in AI answers (+25.5 percent), across other people's pages, and a correlation is not a promise about yours. What the check measures is only whether such a page exists and has substance. | Your storefront | Answers | medium |
| duplicate_descriptions Several products share one near-identical descriptionWhen several products carry the same paragraph, nothing ties an answer to one product rather than another, and whatever an engine lifts describes all of them equally, which is to say none of them. This check reads the product descriptions visible in the storefront page's structured data, so it sees a sample of the catalogue rather than all of it. | Your storefront | Answer-ready copy | medium |
| quantified_facts The description states no measurable factA physical product described without a single measurement, weight, duration or quantity gives an answer engine nothing concrete to repeat. This is scored on answerability alone: the one controlled re-test of adding statistics to retail pages measured no reliable benefit, so the value of a number here is that it can be repeated accurately, not that it moves any outcome. A price does not count, because every product has one. The automated fix may only surface or restructure numbers already in your product data; a figure that is not in the source is refused, never invented. | Your products | Answer-ready copy | high |
| answer_shaped_copy The description is one unbroken block of textAn extractor lifts a block, not an essay: a heading, a list, a table, or a paragraph short enough to stand alone. The evidence is correlational and we say so: Semrush measured section structure (+22.9 percent) and lists (+21.6 percent) as correlates of appearing in AI answers, across other people's pages. What this check measures is only whether your copy has a liftable shape. | Your products | Answer-ready copy | medium |
| unsupported_superlatives The description praises without evidenceA superlative with nothing behind it is a sentence no answer can safely repeat and no shopper can check. This measures copy quality only: it is not a compliance ruling and it makes no claim about how any engine treats the words. A claim standing next to its evidence, like a strength beside its milligram figure or a material beside its certification, passes untouched. | Your products | Answer-ready copy | medium |
We measure your store, not the assistants. We do not claim to know what ChatGPT or Gemini will recommend, and we will not sell you a number we cannot take.
See where your store stands.
Read-only, public pages only, about 2 minutes. One free audit per store, so we ask for an email and keep it with your result.
Already convinced? Install Agentible on your store