Every check we run. Every fix we ship.

Search engines and AI shopping agents cannot see a photograph. Alt text is the only description of the image they can read, so a filename or a repeated line leaves them with nothing.

The meta description is the snippet under your result. Without one, the search engine picks a sentence from the page for you.

The SEO title is the clickable headline in search results. Past 60 characters it is cut off mid-phrase.

This is about passage attribution, not the local-search folklore about repeating your name and address everywhere. A generative engine retrieves a passage, not a whole page: if the passage it lifted never names the brand, there is nothing in it to attribute the product back to you, whatever the rest of the page said.

Retrieval crawlers and training crawlers are different jobs, and one robots.txt line often stops the wrong one. OpenAI states that a site blocking OAI-SearchBot will not be shown in ChatGPT search answers, while GPTBot only gathers training data. Blocking the training set is a fair choice and costs a store nothing here; blocking the retrieval set shuts an engine out of the storefront altogether.

Google requires a page to be indexed and eligible for a snippet before it can be used in its AI features. A noindex or nosnippet directive is the page asking to be left out, and it applies whether or not anyone meant it to: these tags survive theme changes and staging launches, so they are usually left over rather than chosen.

Retrieval crawlers do not reliably execute JavaScript. When the copy is written into the page after load, what the crawler fetched is an empty shell with a navigation bar in it, and the product text was never there to read.

sameAs is the list of other places the same company is, and link rot is what happens to lists nobody re-reads: a profile gets deleted or renamed, and the storefront keeps pointing at nothing, or at whoever took the handle after you. Only a definitive answer counts here: a profile that refused our request is recorded as unknown, never as gone, because social platforms wall off automated traffic routinely and a refusal is not evidence of anything.

Anything reading the storefront treats "Acme", "ACME Inc." and "acme inc" as three names until something tells it otherwise, so what one brand did gets split three ways. This is about the store being legible as one company, and we make no claim about what any engine does with it.

A 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.

An 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.

A 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.

Returns 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.

"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.

"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.

"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.

A 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.

When 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.

Blog articles are where answer content usually lives: long enough to state a real answer, stable enough to be retrieved. This factor reads your blog through an optional, read-only permission you grant separately and can revoke; until you do, we simply report that we could not look, and nothing counts against you. What it measures is the same shape the support-content check measures on pages: a question a shopper would ask, answered directly under it, in text long enough to mean something.

Every kind of store has a handful of questions shoppers ask an engine before buying, and we keep a small, published panel of them per store category: the same five questions our weekly visibility sample asks. This flag names the panel questions none of your pages or articles addresses in a heading. It is a coverage note, deliberately the lightest flag on this checklist: a question you cannot be quoted on is a question somebody else answers, and each one this names is one piece of content away from covered.

AdsBot is the crawler Google uses to check the pages your ads point at, and it plays by its own rule: it ignores a blanket 'block everything' line and obeys only a rule that names it. So a store that blocked 'all bots' has not blocked AdsBot and is fine here. This flag fires only when your robots.txt names AdsBot-Google in a Disallow, which is usually a copy-pasted blocklist, and it means Google cannot verify your landing pages: ads get disapproved for reasons that look like a mystery.

Price and availability mismatches are the two most common reasons Merchant Center disapproves a product: Google compares the page it fetches against the data it was given, and a disagreement reads as unreliable. We probe a small sample of your product pages and compare each page's own structured data against your product record. A page whose offer states a different price, or no availability at all, is what this flag names. We only ever compare a page against your own data, and a page we could not read or parse is reported as unknown, never as wrong.

The same MerchantReturnPolicy and OfferShippingDetails markup the AEO agent checks has a second, separately documented consumer: Google's shopping surfaces use it to show shipping cost and return windows on ads and free listings. Without it, Google falls back to account-level settings in Merchant Center, or estimates, and an estimate that disagrees with your checkout is a disapproval risk. This is the one place structured data has a documented commercial effect, which is why two agents check it: AEO for answers, this one for ads. Fixing it once satisfies both.