We checked 49 online stores for AI shopping visibility. 8 in 10 hid the one signal AI trusts.
We ran 49 stores, from big DTC brands to small independent shops, through the same AI-visibility check a store owner runs. They scored 84 out of 100 on average and looked fine, yet more than 8 in 10 exposed no product rating an AI assistant could read. Here is the data.
We kept running into the same claim: most stores are not ready for AI shopping assistants like ChatGPT, Perplexity, Gemini and Google's AI. It is easy to say and hard to prove, so we measured it. We read a real product page from 49 stores the way an assistant does, in two groups on purpose: well-known DTC brands from best-of lists, and smaller independent shops from niche directories.
Here is the surprising part. These stores are not bad. They scored 84 out of 100 on average, every one used HTTPS, and every one showed a price a shopper could read. They look fine. Yet more than 8 in 10 of them hid the single signal an AI assistant leans on most to trust a product. You can look great to a human and still be invisible to the machine making the shortlist.
The one thing 8 in 10 stores got wrong
The signal almost everyone missed was the structured product rating: the review score written as data an assistant can read, not just star graphics on the page. 41 of the 49 stores, more than 8 in 10, showed star reviews to human shoppers but exposed no rating an AI assistant could read.
That matters because an assistant cannot see your stars. It reads the markup, and the rating is one of the few things that tells it your product is trusted. When that number is locked in an image or a review widget, the assistant has nothing to verify, so it leans toward a competitor it can read.
What we checked, and why it matters
An AI shopping assistant does not look at your photography or your reviews the way a person does. It fetches the page, reads the machine-readable facts, and recommends the products it can verify. So we checked the facts an assistant needs:
- Product schema (JSON-LD): the structured block that states your price, availability and rating in a format assistants trust.
- A machine-readable price: the number present in the markup, not locked in an image.
- Availability: a clear in-stock or out-of-stock state.
- A structured rating: the review score as data, not just star graphics.
A fuller explanation of each lives in our guide on Product schema for Shopify and AI search.
What we found
Across the 49 product pages:
| Signal an assistant reads | Stores missing it | Share |
|---|---|---|
| Structured product rating | 41 of 49 | more than 8 in 10 |
| Availability inside Product schema | 10 of 49 | about 1 in 5 |
| Price inside Product schema | 8 of 49 | about 1 in 6 |
| Product schema at all | 5 of 49 | 1 in 10 |
| A human-readable price somewhere | 0 of 49 | none |
The average grade was 84 out of 100. Ratings were the gap that showed up again and again. Price and availability were a softer version of the same story, missing from the structured data on roughly one in five pages, and only one in ten stores had no Product schema at all.
Big brand or small shop, the gap was the same
We split the sample on purpose to see whether this was a small-store problem. It was not. The well-known brands averaged 83 and the smaller independent shops averaged 86, so if anything the small shops did slightly better, and both groups hid ratings at about the same rate. This is not a budget or a size problem. It is a default-settings problem: most store platforms ship reviews as on-page widgets and never put the rating into the markup, so almost everyone has the same gap until they fix it.
What this means if you run a store
Two things follow. First, a gap here is normal, so it is not a sign you did something wrong. Second, it is a real opening, because the fix is markup, not a redesign. Most stores are one or two signals away from being readable to an assistant.
The single highest-leverage move for almost every store in this study would be the same: put your product rating into Product schema, then confirm price and availability are in there too. That is a small change to the page an assistant reads, and it is the difference between being verified and being skipped.
Check where your store stands
Before you change anything, see what an assistant gets when it reads your store today. Run the free ShelfGrader scan: paste your URL and you get a grade, the fixes ranked by impact, and the competitor an assistant would pick instead. Checking is free. If you want the fixes done, the $20 pack generates your Product schema for you to paste in, and the $99 per month plan keeps it correct as your products change.
Frequently asked questions
How did you measure AI shopping visibility?
For each store we opened a real product page and read it the way an AI shopping assistant does, checking for the facts an assistant needs before it will recommend a product: Product schema (JSON-LD), a machine-readable price, availability, and a structured rating. We scored each page out of 100 with the same free ShelfGrader check a store owner runs. The sample was 49 product pages in two groups: well-known direct-to-consumer brands from public best-of lists, and smaller independent stores from niche directories, across apparel, coffee, beauty and other categories.
Is 49 stores enough to draw conclusions?
It is a sample, not a census, and we are careful about that. We deliberately scanned two groups, big established brands and smaller independent shops, to test whether the gap was a small-store problem. It was not: both groups scored about the same (low-to-mid 80s) and both left ratings off at roughly the same rate. The raw results are saved so the numbers are checkable.
What was the most common gap?
Structured ratings. 41 of the 49 stores, more than 8 in 10, exposed no machine-readable product rating an assistant could read, even when the page showed star reviews to human visitors. The reviews were there as images or scripts; the number an assistant trusts was not in the markup. It was just as common among big brands as small shops.
Does a low score mean a store is badly built?
No, and that is the surprising part. These stores scored well, 84 out of 100 on average, every one used HTTPS, and every one showed a price a person could read. They are not broken. They simply leave specific facts out of the machine-readable layer that AI assistants rely on, which is a quick fix rather than a rebuild.
How do I check my own store?
Run the free ShelfGrader grade. Paste your store URL and in about a minute you get a score, the ranked fixes, and the competitor an assistant would pick instead. Checking is free; having the fixes done is the paid part.