Every on-page tool gives you a number. Seventy-nine out of a hundred, grade B, a coloured ring. It feels like progress, it is easy to share with a client, and it is the thing we deliberately do not do.
What a score hides
To produce one number from twenty different checks you have to decide how much each is worth. Is a missing meta description worth five points or fifteen? Is a noindex tag worth forty, or is it worth everything?
The moment you answer that, you are asserting something about a page you have never seen. And whatever weights you pick, two pages in completely different trouble end up with the same mark:
- One page has a perfect title, description, headings and structured data, and a
noindexleft over from staging. It cannot appear in Google at all. - Another has a slightly long title and no alt text on two images. It ranks fine.
A score can easily put those within a few points of each other. Worse, the second page's owner will spend an afternoon chasing the last ten points, while the first page's owner sees green and ships.
What we do instead
The on-page checker counts:
- Passing: nothing to do.
- Warning: likely to cost you, worth fixing next time you edit the page.
- Failing: stops the page working in search.
Then it lists them worst first, grouped the way a page is read, and each one says what was found, in numbers, and why it matters. The summary ring shows how many checks pass out of how many were judged, which is a fact rather than an opinion.
You still get the quick glance. You cannot get a green light while a noindex is sitting in the head.
The same rule everywhere else
The monthly AI-answer check reports two numbers and refuses to blend them: how often an assistant mentions your brand, and how often it cites a page of your site as a source. Mentions can come from training data you cannot influence this quarter. Citations come from a search the assistant just ran, and content work moves them. One "AI visibility score" over both would hide exactly the half you can act on.
The engine's ranked findings work the same way. Each is priced in the clicks it can win, from your own impressions and position and a public click-through curve, not scored for severity on a scale we invented.
When a number is honest
Numbers are not the problem. Invented ones are. "1,401 of 2,181 pages are unknown to Google" is a number. "4.2 seconds to first byte" is a number. "79/100" is a committee's opinion in a trench coat.
Keyword density is the older version of the same mistake. Google publishes no ideal density, so our checker tells you how often the keyword appears and only warns when repetition is heavy enough to read as stuffing to a person.