We talked before about product data enrichment, and how it helps AI agents to read your website.
But merchants ask: did it change anything? They fixed a wrong bottle size, added "fragrance-free" to a title, or wrote the real niacinamide percentage into a description. Then they open ChatGPT, ask one question, see their product or miss it, and decide whether the work paid off.
One question on one day tells you very little. A research paper from April 2026, "Don't Measure Once: Measuring Visibility in AI Search", makes this point clearly. AI answers change between runs, between prompts and over time, so the authors argue that visibility should be measured many times and read as a range of results.
AI commerce vendors have started publishing long research frameworks for this, with control groups and 90-day study windows. They are careful, and they suit a team with an analyst. Smaller WooCommerce skincare shops might not have an analyst. So this is the small version we use: one fixed list of questions, one spreadsheet and about 30 minutes a week.
Ask the same question three times and you can get three different shortlists. That is how these systems work, so a single good answer after your fix can be luck, and a single bad answer can be bad luck.
Timing adds a second problem. Google says crawling a changed page can take anywhere from several days to several months. No AI assistant tells you the date it will pick up your new product text. If you check two days after the fix and see no change, you have learned nothing yet.
Google also says AI Overviews and AI Mode may use different models, and that AI Overviews often do not appear at all. Treat each assistant and each Google feature as its own channel in your notes. One combined "AI score" hides the differences you need to see.
Take them from what buyers already ask you: chat logs, support emails and the search box on your own site. Strip out names and personal details. For one product fix, 8 to 12 questions is enough. I use three kinds.
Write the correct answer next to each question before you start, copied from the label. Then freeze the list. If you think of new questions later, test them in a separate list so your before and after numbers stay comparable.
Pick the assistants your buyers use, for example ChatGPT, Perplexity and Google AI Mode. Check on the same weekday, in the same language and country settings, and start a new chat for every question. Ask each question three times per assistant. Note that three is our practical choice for a small shop. It is not a statistical rule.
For every answer, record a few things in the spreadsheet:
Paste the full answer text in a last column. When a result looks strange three weeks later, you will want to read what the assistant actually said.
Start two weeks before you publish the fix. Those two weeks are your baseline. Also track two or three similar products you did not change. If they move at the same time as the fixed product, the assistant probably changed, and your fix may have had little to do with it. This way you are a bit more resilient to changes in the environment out of your control.
Keep "mentioned" and "correct" apart. A product can show up more often while the answer still says 50 ml when the bottle is 30 ml. More visibility with a wrong fact will cost you returns and support emails later.
Count instead of guessing. "The fact was right in 3 of 9 answers before the fix and 8 of 9 in the last two weeks" is a finding. "It looks better now" is a feeling. With numbers this small, only a large change that holds for several weeks means much. One good week can be noise.
A drop can be good news. If the should-say-no questions stop returning your retinol cream to someone who asked for retinol-free, the assistant now understands your product better. In skincare that is the result you want, because the wrong match often turns into a return or an unhappy customer.
If nothing moves after six to eight weeks, check the basics before you fix more products. Google says a page must be indexed and eligible to show a snippet before it can appear as a link in AI Overviews or AI Mode, that important content should be available as text, and that structured data should match the visible text on the page. Look for the old value in other places too: an image of the label, a marketplace listing, an old blog post or a product feed you forgot to update.
The best question list comes from your own buyers. Our selling agent keeps a record of what shoppers ask before they buy, and that record is where we take test questions from when we run this check for a store.
We do this for you. Want to see how AI agents read your store today? We built a free snapshot for WooCommerce shops.
Or, start immediately with analytics through CAIDEL for WooCommerce.
How long does it take for AI assistants to show a product data fix?
No assistant gives a fixed date. Google says crawling a changed page can take from several days to several months. Check weekly and expect to wait weeks before you judge the result.
How many times should I ask each question?
We ask each question three times per assistant each week. That is a practical choice for a small shop, not a statistical rule. AI answers vary between runs, so one answer is never enough.
Do I need special schema or an AI file to appear in Google AI Mode?
Google says no. There are no extra technical requirements and no special schema for AI Overviews or AI Mode. The page must be indexed and eligible for a snippet, and your structured data should match the visible text.
My product shows up less after the fix. Is that bad?
Not always. If the drop happens on questions your product does not fit, such as a retinol-free request for a retinol cream, the assistant now matches your product more accurately. That usually means fewer wrong orders.