A 30-minute product data audit for AI agents

Most audits of product data are built for a feed: does every field validate. An audit for AI agents asks a different question. Someone else’s model is going to read your listing, condense it and repeat it to a shopper, so the job is to find what it will get wrong or won’t find at all.

You can do a useful version in half an hour if you sample instead of trying to cover the store. This is the plan we’d follow, with a time box for each pass.

Before you start (3 minutes)

Pick 5 products. Take your best seller, one product that’s recently changed, one with lots of variants, one you’d be embarrassed to see described wrongly (anything a baby, a body or an ingredient list touches), and one at random. Open a blank note with the five product names down the left.

If the same gap shows up in three of the five, it’s probably in the rest of your catalog too.

Pass 1: are they eligible at all? (5 minutes)

Nothing else matters for a product the catalog can’t read. Shopify Catalog includes products from stores that meet a short list of requirements, and we walk through where to check each one in Shopify Catalog requirements: why products get left out.

For your five products, confirm four things:

• The store isn’t in private mode.
• Each product has a title, at least one image and a price above zero.
• Each is published to the online store.
• None is Unlisted or hidden from search engines.

Note any failure next to the product name and fix it first.

Pass 2: what does Shopify’s own meter say? (5 minutes)

Shopify gives you a scorecard. In Sales channels > Agentic, expand Listing insights for a product and you get a colored bar for it. Shopify’s help page says the meter measures five things: description completeness (word count), image coverage (number of images), product reviews (average rating and count from verified sources), variant and option completeness, and shop policy completeness (Managing agentic storefronts).

Two of those are quick to read across all five products: how long each description is, and how many images each has. The variant check has an odd detail worth reading twice. Shopify says it looks at “whether names contain hard-to-understand acronyms or numbers.” An option called “Style 3B” tells an agent nothing. Write down every option name that a stranger couldn’t decode.

The same page has a search preview that lets you “run search queries to review how your products might rank.” Run one search a shopper would type for each product. Use it to spot a product that never shows up, not to chase a rank.

Pass 3: is the basic data filled in? (5 minutes)

Shopify’s guidance on optimizing products lists the fields that matter: title, description, images, type, vendor, collections, tags, barcode, and variants including option names. It asks for “clear, accurate, and detailed descriptions.”

Open each product and check that every one of those fields has something in it, and note the gaps. If your real data lives in metafields, Catalog Mapping lets you point the catalog at them.

Pass 4: do your own facts agree? (7 minutes)

This is the pass that catches the most. A summarizer that finds two versions of a fact has to pick one, and it won’t tell you it guessed.

For each product, find every place that states an age range, size, count, material or ingredient. Compare the title, the description, the variants, the product details fields and any image that carries text. Write down every disagreement, with both versions.

Then check where the important facts live. Shopify recommends adding “detailed product specifications and technical details” and to “include sizing guides, material information, and care instructions as applicable” (Optimizing your store for AI). If a warning, an age limit or a care instruction exists only in a photo or a collapsed tab, plan to put it in the description text as well. We cover the writing side in How to write product descriptions for AI shopping agents.

Pass 5: what claims are you making, and could you back them? (5 minutes)

List the claims in each description that a shopper could act on: “fragrance-free,” “GOTS-certified,” “pediatrician recommended,” “fits standard cribs.” Our claim categories are a handy way to sort them, from safety and age to care and origin.

For each one, ask whether your copy states it plainly and whether you could show where it comes from. A claim you can’t support is a problem with or without AI. One you can support but state vaguely is where summaries drift.

What this audit can’t tell you

It only audits your side. Shopify’s catalog doesn’t simply copy your copy: its developer docs mark fields like description, top_features and unique_selling_points as Inferred, meaning “generated or enriched by Shopify” (Global Catalog docs), which we explain in What Shopify’s AI adds to your listings. Your data is the input. What comes out is a separate question, and How to read what Shopify’s catalog says about your product shows how to look at it yourself.

When we compared the catalog’s version of 192 top-ranked products with each brand’s own copy on Sep 24, 2026, 89% had at least one claim the brand never made (how we checked). That was one day and the top results only, so it isn’t a rate for your store. It does mean a tidy listing doesn’t settle the matter. The study has the examples, and more how-to guides cover the fixes.

What to fix first

Blockers from Pass 1 come first. Conflicts between your own facts from Pass 4 come second, because they’re the cheapest to fix and the most likely to be repeated wrongly. Thin descriptions and cryptic option names are a backlog to work through a product at a time.

Common questions

How often should I repeat this?

After a bulk edit, an import or a theme change, and whenever you add a product line.

Can I do this for the whole store?

Not in 30 minutes. Sampling finds patterns, and patterns are what you fix in bulk.

Does a good listing-quality score mean the catalog describes my product correctly?

No. The meter measures how complete your data is, not whether Shopify’s generated summary matches it. Comparing the two is a separate step. The free scan from Hearsay reads what the catalog tells AI agents about 5 of your products and puts each claim next to what you wrote.