Product data checks
Ten checks you can run on one product page before AI assistants and shoppers compare it. Each item says what to look for, how to check it, and how to record what you cannot confirm.
Last updated September 19, 2026 · ~10 min read
An AI assistant can only describe and compare a product using the facts it can read. This checklist is for merchants, ecommerce managers and product marketers who want to check those facts themselves, one product at a time, before looking at any particular assistant.
It is most useful for products that buyers compare on specifications, fit or evidence: electronics, outdoor gear, beauty, home goods and similar categories. It will not tell you why a specific assistant showed or skipped a product — assistants do not publish those reasons — and passing every item does not mean a product will appear. It tells you whether your own product information is complete, consistent and backed by sources.
Give every item one status. The statuses matter as much as the checks, because each one leads to a different next step.
| Status | Use it when |
|---|---|
| Present | The fact is published, correct, and the same everywhere you checked. |
| Missing | A fact buyers compare on is not published anywhere. |
| Conflicting | Two places you publish give different values. |
| Unconfirmed | A claim is published, but nothing you can point to supports it for this model. |
| Unknown | You could not check it yet, for example because the answer is with a supplier. Note who can answer. |
| Not applicable | The item does not apply to this product. Say why. |
Keep Unknown apart from Missing. Filling a gap with a guess turns an honest unknown into a wrong fact, and a wrong fact is harder to notice than a missing one.
What to check. The brand, product name, model and variant are written the same way in the page title, the main heading, the product details and any feed or marketplace listing.
How to check. Put the page title, the heading, the brand and model in the product details, and your feed or marketplace row side by side. Look for renamed models, missing model numbers, and variant names that differ between places (“Graphite” on one, “Charcoal” on another).
Record it as. Conflicting when any two differ. Missing when the model number is not stated at all.
What to check. Each variant carries the identifiers that exist for it: a GTIN (UPC, EAN or ISBN), the manufacturer part number (MPN), or your own SKU.
How to check. View the page source and search for gtin, mpn and sku in the structured data. Compare them with your feed or catalog export for two or three variants.
Record it as. Missing when an identifier exists but is not published. Not applicable, with the reason, when the product has none, such as a made-to-order item.
What to check. Each size, color, capacity or bundle is a separate, selectable variant with its own identifier, price and availability, rather than one description that blends them.
How to check. Select each variant on the page and watch whether the price, availability, images and identifiers change with it. Check that the text does not describe one variant’s specification as if it applied to all of them.
Record it as. Conflicting when the text states a specification that only some variants have, without saying which.
What to check. Dimensions, weight, capacity, power and similar numbers have units, and it is clear which variant or size each number applies to.
How to check. List every number on the page. Each needs a unit and, where it changes by variant, the variant it describes: “295 g (US men’s size 9)”, not “lightweight at 295 g”. Check that the page, the spec sheet and the feed use the same values.
Record it as. Conflicting when two places give different numbers for the same variant. Missing when a number buyers compare on is absent.
What to check. The page says what the product is for and what it is not for: compatibility, operating conditions, size and fit, skin or hair type, terrain, age range — whatever buyers in your category ask before they choose.
How to check. Write down three questions a buyer in your category asks before choosing, such as “Will it fit a 15-inch laptop?” or “Is it suitable for sensitive skin?”. Check whether the page answers each one as a fact, including a “not suitable for” where there is one.
Record it as. Missing for each question the page does not answer.
What to check. Every performance or suitability claim — waterproof, hypoallergenic, battery life, certified organic — points to a source that applies to this product and model: a test report, a certification, a standard or a spec sheet.
How to check. Mark every claim on the page. For each one, find the source and check three things: that it is about this model and not the previous one, that the conditions match (“10 hours at 50% volume”), and that it is still current.
Record it as. Unconfirmed when no matching source exists. Narrow the claim to what the source supports, or remove it. See product claims and evidence checks.
What to check. Key facts are text in the page itself — not only inside images, PDFs or tabs that load after a click — and the crawlers of the AI assistants you care about are not blocked.
How to check. View the page source, or fetch the page with scripts turned off, and search for your key specifications. Check robots.txt and any bot protection on your CDN; the major AI assistants publish the names of their crawlers. For ChatGPT, see the ChatGPT shopping checklist.
Record it as. Missing when a fact appears only in an image or a downloadable file. Unknown when you cannot see your CDN settings — note who manages them.
What to check. The product structured data (JSON-LD) agrees with what shoppers see: name, brand, price, currency, availability and identifiers.
How to check. Run the page through a structured data testing tool, or read the source, and compare each value with the visible page. Check again for a variant that is on sale or out of stock.
Record it as. Conflicting when the markup and the visible page differ. Fix the value where it is maintained, not only in the markup.
What to check. Price, currency, availability, shipping cost and time, the return window and the warranty are stated, and they match your checkout and policy pages.
How to check. Add the product to the cart and compare the price, shipping and availability with the product page. Open the returns and warranty policies the page links to and confirm the terms apply to this product.
Record it as. Conflicting for any mismatch. Missing when a term only appears once checkout has started.
What to check. Dated information is current: model year, “new” labels, prices, stock status, spec sheets and manuals. Feeds are updated when the price or stock changes.
How to check. Look for dates on the page, the spec sheet and the manual, and compare them with the current model. Check when your feed last refreshed and whether it follows price and stock changes.
Record it as. Conflicting when an old specification or price is still published anywhere. Write down the date you checked.
Illustrative example. The product, values and sources below are invented to show how to fill in the checklist. They are not a customer result.
Product: Example insulated water bottle, model HB-750, three colors.
| Check | Status | Note |
|---|---|---|
| 1. Product identity | Conflicting | The page says “HB-750”; the marketplace listing says “HB Bottle 0.75L”. |
| 2. Identifiers | Present | A GTIN is published for each of the three colors. |
| 4. Units | Conflicting | The page says 750 ml; the listing says 25 oz, which is about 739 ml. |
| 5. Use conditions | Missing | The page does not say whether it is dishwasher-safe or suitable for carbonated drinks. |
| 6. Claims | Unconfirmed | “Keeps drinks cold for 24 hours” has no test conditions or source. |
| 7. Readable as text | Missing | Care instructions exist only in a PDF. |
| 9. Commercial terms | Present | Price, shipping and the 30-day return window match checkout. |
| 10. Freshness | Unknown | The spec sheet is undated. Asked the supplier whether this year’s lid change affects capacity. |
After you change the page, run the checks again and write down the date. The record is what lets you tell later which change came first.
These checks are quick for one product and slow for a hundred. Hasmord runs product data checks on each product page you add, or on a CSV import; keeps missing, conflicting and unconfirmed information apart; matches claims to sources; and drafts improvements that a person on your team approves before anything changes. After you update a page, you can check it again and compare. See product data audit for AI discovery for what a report looks like.
To try it on one product, use the free check on the homepage. It needs no signup. Plan limits for larger catalogs are on the pricing page.
Related: product data audit · product claims and evidence checks · why isn’t AI recommending my product?
The facts buyers compare on and the evidence behind them: product identity and identifiers, variants, units, use conditions and limits, claims with sources, facts readable as text, structured data that matches the page, commercial terms, and freshness.
No. AI assistants do not publish how they choose products, and results vary by assistant, question, region and time. The checklist makes sure your own product information is complete, consistent and backed by sources.
Missing means a fact is not published anywhere. Unconfirmed means a claim is published but nothing supports it for this model. Unknown means you have not been able to check yet; note who can answer rather than guessing.
After any change to the product, its price or its listings, and on a regular schedule for the products you sell most. Write down the date you checked each item.