From Vanity Metrics to Fixable Issues: A Better Website Scorecard

There is a particular kind of number that makes you feel informed without telling you anything. Your website scores 74. Out of what? Compared to whom? And most importantly: now what? I call these vanity metrics, and the web is full of tools that produce them. You paste in a URL, a gauge spins, a number appears, and you close the tab feeling vaguely uneasy but no wiser than before.

The problem isn’t that the number is wrong. It might be perfectly accurate. The problem is that a number is a summary, and summaries are only useful when you already understand the thing being summarized. A doctor who tells you “your health is a 74” has told you almost nothing. A doctor who says “your blood pressure is high, here’s why that matters, and here’s what to change” has told you something you can act on. The score without the diagnosis is theater.

When I built GazeSite, I kept running into this. The tempting thing to build is the gauge. Gauges are easy. You compute some weighted average, render a semicircle, pick a color. Users screenshot it. It looks like a product. But I noticed that when I audited websites by hand for people, nobody ever asked me for a number. They asked me what was wrong and how to fix it. The number was, at best, a way of deciding whether to keep reading.

So I ended up with a rule that I think applies to any scorecard, not just for websites: every score must be attached to a finding, and every finding must contain four things. A score, so you can prioritize. A headline, so you can scan. The issue, stated concretely enough that you could point at the pixel or the line of HTML that causes it. And the remedy, prescribed specifically enough that a developer could do it tomorrow morning without asking a follow-up question.

That last part is where most tools quit. “Improve your accessibility” is not a remedy. “Your hero image has no alt attribute, so screen readers announce it as nothing; add alt text describing what the image shows” is a remedy. The difference is the difference between a grade and a lesson. One tells you where you stand. The other changes where you stand.

There’s a subtle reason the vanity-metric version keeps winning in the market, and it’s worth being honest about. Concrete findings are falsifiable. If I tell you your contrast ratio on the pricing button fails the WCAG threshold, you can check. If I tell you your site scores 74, you can’t. Vague claims are safe claims. So tools that want to seem authoritative without doing the hard work drift toward the number, the way horoscopes drift toward statements that are true of everyone.

The hard work, it turns out, is mostly about looking at the right evidence. A useful audit has to see what a visitor sees: the rendered page, not just the source. It has to see it on a phone, because that’s where most first visits happen now, and a page that’s fine on a desktop can be unusable at 390 pixels wide. And it has to read the HTML too, because some of the most consequential problems, a missing meta description, a heading hierarchy that skips from h1 to h4, a form input with no label, are invisible in a screenshot. GazeSite captures both, on both viewports, precisely because neither view alone supports concrete findings. You can’t prescribe a remedy for something you haven’t actually observed.

Once you commit to findings instead of a single number, something interesting happens to the score itself: it becomes useful again. A score attached to a specific finding is a priority signal. If the audit says your checkout button fails on mobile and scores that finding as severe, while your footer typography scores as minor, you know what to do first. The score has gone from being the product to being the sort order. That’s its proper job.

I think there’s a general principle here about tools that evaluate things. The value of an evaluation is proportional to how directly it maps to an action. Report cards, credit scores, code linters, they all sit somewhere on this spectrum. A linter that says “code quality: C+” would be useless. A linter that says “line 41: this variable is never used, delete it” is one of the most quietly valuable tools in software. Website audits should aspire to be linters, not report cards.

So when you evaluate any audit tool, mine included, ask one question of its output: could a competent person, given only this report, make the site better this week? If the answer is yes, the score is earning its place. If the answer is “well, first I’d need to figure out what the score means,” you’ve been handed a vanity metric, and the honest move is to close the tab.

The web doesn’t need more numbers. It needs more sentences of the form: here is what’s broken, here is why it matters, here is exactly what to do. Everything else is decoration.