Guide · Website checks
Check product information for AI search
Check the exact product, variant and availability across the visible page, product data and the AI answer you recorded.
Pick one product prompt
Start with a product the business currently supplies. Keep the customer's requirement and the exact item together so similar versions are not confused.
Decide which customer decision it supports. A question about size needs a different source from one about delivery, so start with the page that should answer it.
Write the prompt the way a shopper would ask it, with the detail that matters to them. A question that names a size, a colour or a use is easier to check than a general one.
If the source covers another item, record the mismatch. Don't quietly treat a similar product as a match.
Check the product identifiers too. If the answer quotes a model number or a code, compare it with the one on the page and in your records, character by character.
Bundles and sets need the same care. An answer may treat a set and a single item as the same product, and the price and contents will differ.
Find the owning record
Identify the current reference for the product description and offer. A supplier brochure and a live offer may describe different conditions.
Don't infer stock from an old screenshot. If a supplier feed and your page disagree, ask the owner which is authoritative before you draft a correction.
Name one owner for each kind of detail. Price, stock, delivery and specifications often sit with different people, and each needs someone who can say what is current.
Write down where each record lives, such as the shop platform, a supplier sheet or a stock system. That makes the next check quicker and shows which record wins when two disagree.
Don't assume an assistant knows the latest position. Keep older observations as history, and have the owner review any change before it reaches the public product page.
Say when you last checked each changing detail. A stock level you read last month tells you little about today.
Keep a short history for each changing detail, with the old value, the new value and the date it changed. That history lets you date an answer that repeats an older value.
Read the visible page
Look for dimensions, materials, variants, care instructions and delivery limits where they matter. Record what a customer can actually read.
A photograph helps but may not explain a measurement. Put unclear descriptions on the review list and ask the owner to confirm them. Don't fill gaps yourself.
Check that the important details are written as text, not only shown in an image or a size chart picture. Google's guidance for AI Overviews asks for important content in textual form.
Read the page on a phone as well as a computer. A detail hidden behind a tab or an accordion is still on the page, but note where a customer has to look for it.
Keep the proposal short enough for the product owner to check. Don't invent reviews, ratings or demand. A good review starts with facts the business can support.
Put the most important correction first. A wrong size on a page customers buy from matters more than a vague sentence in the care notes.
Compare product data
Google documents product structured data for product search appearances. Ask your maintainer to compare those labels with the visible page, because a code label must describe the same item.
A validation result doesn't show an AI shopping recommendation. Save the validation finding apart from any answer you observe.
Check the price, currency and availability in the markup against the page on the same day. A mismatch between them is a finding even before you look at any answer.
Choose the product markup that fits the page
Google describes two main kinds of product markup, which is code that labels product details for machines. Each suits a different kind of page.
Product snippet markup is for pages where people can't buy the product directly, such as a review article. Merchant listing markup is for pages where customers can buy from you.
Google says merchant listing markup has more options for details such as apparel sizing, shipping and returns. Both kinds support product variants.
Ask your maintainer which kind the page uses and whether that matches what the page does. A shop page marked up as a review, or the reverse, describes the page wrongly.
Google's advice is to provide as much product information as you have, without concern for which display uses it. The point for this review is accuracy, not a particular display.
Keep the feed and the page in step
Google can receive product details in two ways, as markup on the page and as a Merchant Center feed. Google says some experiences combine the two when both exist.
Google gives an example. Product snippets may use pricing data from your merchant feed if it isn't present in the markup on the page.
So a stale feed can carry an old price even when the page is right. Check the feed's price and availability for the item alongside the page.
Google's guidance for AI Overviews also asks site owners to check that their Merchant Center information is current. Add the feed to your list of owning records.
When the feed and the page disagree, ask the product owner which one is right. Correct the other one, then record both dates.
Give each variant an address that loads it
Google documents product variant markup using ProductGroup, with properties for the variants and the attributes they vary by, such as size or colour.
Google's guidelines say each variant needs its own unique identifier in the markup, such as a SKU or a GTIN. The group needs an identifier too.
For a page where shoppers pick variants from menus, Google says the site must be able to preselect each variant with a distinct address.
Preselecting means showing the right image, price and availability for that variant, and letting the shopper add it to the basket. Open each variant address and check that it does.
This matters when you follow a link from an answer. A link to the base product page may open on the default variant, which isn't always the one the customer asked about.
If a variant address opens the wrong option, record it as a page finding for the maintainer. Then check whether the answer you saved linked that address or the base page.
Use availability words that mean one thing
Schema.org defines a fixed set of availability values. They include InStock, OutOfStock, PreOrder, BackOrder, MadeToOrder, LimitedAvailability, SoldOut and Discontinued.
Pick the value that matches what the page tells customers. A page that says made to order should not carry an in stock value in its markup.
Google's structured data guidelines ask for current information. They say Google won't show a rich result for time-sensitive content that is no longer relevant.
Google also warns that markup generated by JavaScript can make shopping crawls less frequent and less reliable. That matters most for details that change quickly, such as price and stock.
Ask your maintainer whether product markup is in the page's initial HTML or added by a script. Record the answer with the date of your check.
Write down the details customers ask about
The questions customers send you are the best guide to what a product page is missing. Look at recent emails, calls and returns for the product you're reviewing.
Typical gaps are measurements with units, materials, what fits with what, care, delivery exceptions and how returns work. Each one is a fact the owner can confirm or correct.
Write measurements in the units your customers use, and say which dimension is which. Width, depth and height in a stated order leave less room for a wrong reading.
Don't add reviews, ratings or popularity claims you can't support. Google's structured data guidelines say not to mark up misleading content such as fake reviews.
Give the owner the list with a source for each item. They decide what goes on the page, and you record what they publish and when.
Measure the right observation
Record whether the answer names the product, links to the page or describes it correctly. Treat these as separate observations.
A shopping placement, a website visit and an order each need their own evidence, and a page check can't supply any of them. Keep the original answer for later comparison.
Keep the prompt, the product and the variant together in every comparison. An answer about the narrow shelf can't be compared with one about the wide shelf as if they were the same item.
Handle products you no longer sell
Answers can go on describing a product after you stop selling it. Decide what the old page should say before you check any answer about it.
One choice is to keep the page and say clearly that the product is discontinued, with a link to any replacement. Schema.org has a Discontinued availability value for this case.
Another is to move customers to a replacement page. If you do, make sure the new page says it is a different product, so nobody reads the old details into it.
Record the date the product stopped and what happened to its page. An answer captured after that date that still offers the product is a finding with a clear starting point.
Check old links from your own site too. A guide or a blog post that still points to the discontinued product keeps the old details in circulation.
Apply the same care to a product that changes without a new name. A new material or a new size under the old name needs a dated note on the page and in your record.
A worked example
Fern Workshop sells a wooden shelf in two sizes. The saved answer links to the wrong size.
| Detail | Observation | Review or next step |
|---|---|---|
| Variant | The customer question names the wider shelf. | Keep the exact size with the prompt. |
| Source link | The answer links to the narrower shelf page. | Record the mismatch before you read anything into the product mention. |
The cited page describes the wrong size. Record the mismatch before treating the answer as information about the product requested.
Common questions
Does valid product data mean I'll be recommended?
No. A validation result and an observed recommendation are different evidence.
Can I call a mention a sale?
No. A sale needs an order record and its own attribution evidence.
Does valid product data mean I'll be recommended?
No. A validation result and an observed recommendation are different evidence.
Can I call a mention a sale?
No. A sale needs an order record and its own attribution evidence.
Check your understanding
The answer links to another size of the product. What should you record?
Read the answer and explanation
- A sale of the requested item.
- A source mismatch needing review.
- A correct match regardless of size.
Answer. A source mismatch needing review.
Keep the exact item and variant. Similar products aren't interchangeable evidence.
Answer and explanation
A source mismatch needing review.
Keep the exact item and variant. Similar products aren't interchangeable evidence.
Your next steps
Sources and method
A Carqo review method. Provider statements link to the relevant documentation.
- Google product data guidance Link checked 4 November 2026
- Google introduction to structured data Link checked 4 November 2026
- Google guidance on AI features Link checked 4 November 2026
- Google Search Central, product variant structured data Link checked 4 November 2026
- Schema.org, ItemAvailability Link checked 4 November 2026
- Google structured data policies Link checked 4 November 2026