What does the check look at, in order?
An agent works through five questions before it recommends or buys anything. The check follows the same order, and each layer carries the weight the evidence supports.
| Layer | The question an agent is answering | Weight |
|---|---|---|
| Getting in | Am I allowed to fetch this page, and does it answer? | 8 |
| Reading the page | Is the product in the page that arrives, before anything runs? | 36 |
| Knowing the product | Do I know what this is well enough to compare it? | 28 |
| Trusting the store | Would I put my name behind recommending this seller? | 18 |
| Completing the sale | Can I actually buy it? | 10 |
Reading the page carries the most weight because it is the precondition for everything after it. Four of the five engines we score against never run JavaScript: Vercel's study of AI crawler logs found they fetch JavaScript files without executing them, and Gemini is the exception only because it rides Googlebot's rendering. A product that appears once the page loads is a product most agents never see. Getting in carries the least because it is mostly all or nothing, and all-or-nothing conditions are handled by caps instead.
How does the score stay honest?
Three rules, and each one is checkable.
Every weight lives in one file with a written reason and the source behind it. The five layers add up to 100, the checks inside each layer add up to that layer's weight, and the code refuses to start if that arithmetic is wrong.
Checks report six outcomes, not two. Beyond pass, partial and fail, a check can come back as not checked, not applicable, or informational. Those three leave the total rather than counting against you. A bottle is not marked down for having no clothing size. A page we could not fetch does not produce confident findings about product data we never saw. Your score is out of the points that applied, and the report says which points were set aside.
Every check shows its work: what we looked for, what we found quoted back verbatim, what it means for a shopper, and the fix. If the report says your return policy is not readable, it shows you the fields it found.
What happens when we cannot read your page?
Some conditions cap the score no matter how good the product data is. The report always shows the capped score next to the score the page earned on its own, names the cause in plain words, and quotes the evidence.
| What we found | Score is held at | Why that number |
|---|---|---|
| robots.txt turns away a shopping crawler for this path | 25 | The only condition with a vendor statement of exclusion behind it. OpenAI says a site opted out of OAI-SearchBot will not appear in ChatGPT search answers |
| The page tells search engines to leave it out | 45 | Serious, because AI answers draw on search indexes, but user-initiated fetchers and product feeds still reach you |
| We could not read the page at all | 45 | A low score here would be a claim about content we never saw, so we say that instead |
| No product information anywhere on the page | 30 | Nothing for an agent to work with, and nothing for us to score |
The cap is never the whole story. A store blocked by one line in one file usually recovers most of its score by changing that line, and the report says so.
Where does my score sit against other stores?
We scanned 235 product pages across 116 stores on July 25, 2026, covering Shopify, WooCommerce, Magento and BigCommerce, and kept the full distribution. The findings are written up in full. Your percentile is worked out against those stored scores.
| Band | Score |
|---|---|
| Bottom quarter | Below 47 |
| Median | 66 |
| Top quarter | 74 and above |
The practical read: below 47 an agent is probably skipping the page, and above 74 you are ahead of three quarters of the stores we have scanned.
How is this different from AI visibility monitoring?
Monitoring tools watch whether your brand turns up in AI answers. That tells you the outcome. It does not tell you which of your pages an agent could not read, or why.
| Monitoring tools | Acom's readiness check | |
|---|---|---|
| Tracks whether you appear in AI answers | Yes | No, it measures the page instead |
| Tells you whether an agent can reach the page | No | Yes, per crawler, for that exact path |
| Tells you what an agent sees before JavaScript runs | No | Yes, scored on the page that arrives |
| Tells you whether an agent could buy | No | Yes |
| Shows the evidence behind each finding | No | Yes, quoted from your page |
| Cost to start | Usually paid | Free, one link |
What do I need to run it?
One product link, ideally a best seller. Paste it, and we fetch the page the way an agent would, without running JavaScript, because four of the five engines never do. We check the policy pages and the catalog endpoints, and score what we found. We then email a 6-digit code to confirm it is you, and the report opens. No account, no store connection, no card.
Our scanner announces itself honestly and never pretends to be a vendor's bot. If your security settings turn us away, that is a finding in your report rather than something we work around.
What do I get at the end?
A score, the percentile behind it, and a list of findings grouped by the five layers, each with the evidence and the fix. You can work through it yourself. Nothing in the report is locked to Acom.
If you would rather not do it by hand across a whole catalog, that is what Acom is for. A free trial connects your store and applies the fixes across your products, and you approve every change before anything goes live.