Most of the tooling being built right now answers one question: did an AI assistant mention your brand? It is a fair question and it has a whole category of products behind it — run a set of prompts, count the mentions, report a visibility score.
None of it tells you what happened after the mention. An agent that hears your name goes to your page. What it can do when it arrives is a completely separate property of your site, and it is the one that decides whether the mention was worth anything.
The clearest example: two prices on one page
A product page shows $59. Its structured data — the block written specifically for machines — says 49.00, left behind by a sale that ended. Both numbers are on the page. A person only ever sees one of them.
Every validator will pass this page. The markup is well-formed, the schema is correct, the required fields are present. Nothing is broken in the sense a linter understands. The page is simply telling two different stories, and the agent believes the one addressed to it.
This is what we call a coherence failure, and it is invisible to any tool that checks structure rather than agreement.
The same shape, in other places
Once you look for disagreement between the human view and the machine view, it turns up repeatedly.
- A buy button that only exists after JavaScript runs. An agent reading the HTML your server sent finds no button. The page looks finished to you and has no purchase path to it.
- Product copy that loads client-side. The agent summarises an empty shell, then tells a buyer about you using whatever it found — which was not your page.
- Variants with no addressable URL. Asked for “the blue one in medium”, an agent cannot construct a link, so it cannot hand the buyer the thing they asked for.
- A consent wall in front of the content. The agent reports on the wall, because the wall is what it was served.
Why visibility tools cannot see any of this
A mention-tracking tool observes the assistant’s output. Everything above happens on your page, before the assistant forms an answer, and leaves no trace in the output except a wrong price or a missing recommendation that nobody attributes to a markup mismatch.
The two measurements are complementary rather than competing. Visibility tells you the mention happened. Readiness tells you whether the mention could go anywhere. Being mentioned by an assistant that then quotes a price you will not honour is worse than not being mentioned.
How to check your own page in two minutes
Open your product page, view source, and search for application/ld+json. Compare every price, name and availability value in that block against what the rendered page shows. Then disable JavaScript and reload: whatever remains is approximately what an agent gets.
The free scan does this and the rest of it automatically, and every finding names the check it came from so you can read the rule at /rubric rather than take our word for it.