AI product visibility for ecommerce: measure the shelf before you change it

A shopper does not ask an assistant for “running shoe keywords.” They ask for a waterproof trail shoe for a wide foot under $150. The answer may name three products, explain the trade-offs and cite a retailer, a review site or the manufacturer. If yours is missing, a conventional rank report will not explain why.

That is the useful meaning of AI product visibility: whether your products enter the answer when the question resembles a buying decision.

Start with questions a buyer could act on

Ten good questions beat 500 generated variations. Pick one product category and write questions across four jobs:

• choosing the best product for a specific use;
• comparing two plausible options;
• finding an alternative to a known product;
• checking a constraint such as price, fit, material or delivery.

Remove questions that could not change a purchase. “Tell me about coffee” may produce an answer, but it does not test commercial visibility. “Which decaf coffee subscription offers whole beans?” has a decision inside it.

Use the same wording every time. AI answers vary, so changing the prompt and the model together leaves you unable to tell what moved.

Record the answer as evidence, not a score

For every run, save the date, model, market, exact prompt, products named, order of appearance and cited URLs. A screenshot helps with review, but structured rows make comparisons possible.

Calculate mention share only after keeping the underlying evidence. If your brand appears in two of ten answers and a competitor appears in seven, the gap is useful. It is not a market-share figure and should never be presented as one.

The citations often contain the better clue. A competitor may appear because a respected comparison page describes its use case clearly. Another may have consistent product attributes across its page and feed. The answer tells you where to investigate; it rarely proves the cause.

Check four layers before recommending a fix

First, confirm access. Can the relevant page be fetched, indexed and read without relying on an uncertain rendering path?

Second, compare product facts. Are price, availability, variants, identifiers and specifications clear on the page? Google documents that product data can arrive through page-level structured data, a Merchant Center feed, or both. Using both improves eligibility for its shopping experiences, but does not guarantee an appearance.

Third, inspect consistency. Google’s Merchant Center guidance asks merchants to keep product titles and descriptions aligned with the landing page. A feed saying one thing while the page says another creates a data-quality problem before it becomes a visibility theory.

Fourth, review external evidence. Which publishers, reviewers and marketplaces are cited? Your own page can describe a product; independent sources can establish the context in which it is compared.

Repeat the baseline without moving it

After a change, rerun the same question set under the same recorded conditions. Report movement, no movement and confounders. A model update or a newly indexed review may change an answer without your implementation causing it.

Do not promise a result in 30 days. Thirty days is a sensible review point, not a platform deadline.

The first report should end with three actions ranked by impact, effort and testability. A precise recommendation such as “align variant identifiers between the feed and these five product pages” is useful. “Improve AI SEO” is not.

Run the ten-question ecommerce comparison if you want the baseline prepared for you.

Sources

• Google Search Central, Product structured data.
• Google Merchant Center, Tips to optimize product data.
• Google Merchant Center, Landing page requirements.