Website SEO score meaning: 18 checks, one weighted number
A score like "82" is a real number doing a specific, fairly mechanical job — not a mystery grade handed down by an algorithm no one can see inside.
Plenty of builders show a score out of 100 and leave it there, which turns the number into blind faith ("82 must be fine") or blind panic ("why is it 79, I didn't touch that page"). Here is the mechanism end to end, using Olmira's own scoring engine as the worked example — the one we can show completely, check by check, rather than describe in the abstract. Came to fix things rather than understand the maths? The on-page SEO checklist is the same 18 checks written as a to-do list.
Three tiers, three outcomes
Every check carries one of three severities, and severity sets its weight. Critical checks — a working title, exactly one main heading, not being accidentally hidden from search — are worth ten points each. Important checks sit in the middle, four to eight points. Nice-to-haves — description length, the slug's tidiness, a social share image — round a page out without being able to sink it.
Each check returns pass (full weight), partial (half) or fail (nothing). The critical checks are genuinely binary — there's no such thing as half an H1 — but plenty of the rest use the middle state for real: a meta description that's present but a little too long is a partial, not a fail, because it beats having none at all.
The full list of checks
From 18 outcomes to one number
Run all 18 checks
Each check inspects one specific thing about the page — the rendered title's character count, whether the focus keyword shows up where it should, how many content images are missing alt text — and returns pass, partial, fail, or "not applicable" when the check has nothing to measure (a page with no focus keyword set, for instance, skips both keyword checks rather than failing them).
Convert each outcome to points
A pass earns that check's full weight; a partial earns half; a fail earns zero. A "not applicable" check earns nothing, but — this is the part people miss — it's also removed from the total the page is being measured against, not counted as a missed opportunity.
Divide earned by possible
Add up the points actually earned, divide by the total weight of every check that applied to this page, multiply by 100, and round. Because "not applicable" checks drop out of both sides of that fraction, a page with no focus keyword set isn't punished for it — the two keyword checks simply drop out, and the maximum the page is being judged against shrinks along with what's actually relevant to it, rather than the missing keyword counting against it as a fail.
Band it into a letter
The final 0–100 number bands into a grade: A from 90 up, B from 75, C from 60, D from 40, and F below that. The pages on this site are held to the same engine and target 95 or higher on it, which is a genuinely tight bar once you've seen how unforgiving the critical checks are.
A few things that surprise people
No. The number is entirely internal — a platform's own weighting of things search engines genuinely do read directly off a page's HTML, like the title tag, the meta description and your headings. Nothing about the 0–100 figure is transmitted anywhere.
No — 98 is the most any page can score, and it's worth knowing why rather than chasing the last two points. The social-sharing-image check pays full marks only when the image's dimensions are known to be 1200×630 or larger, and the builder doesn't measure them, so that check settles at half its four points once an image is set. On a page where all eighteen checks apply, those are two points out of a hundred, which is where the 98 comes from. On a page where some checks come back "not applicable", they leave the total the page is measured against — as Step 3 above explains — so the cap is worked out against a smaller total rather than always being a hundred minus two. Either way it lands at 98 or below, never a flat 100. Everything else is genuinely earnable, and a perfect run isn't required: a page in the low 90s with a couple of nice-to-have checks on partial is already in strong shape.
Eight points, the same tier as meta description presence — because word count on its own is a weak signal. A page can clear 300 words and still be thin, or say everything worth saying in 250; the check exists as a floor against near-empty pages, not a reward for length.
Almost always the duplicate-title or duplicate-description check: something changed on a different page that now matches this one's title or description. Check what else was published recently before assuming the tool is wrong.