How scoring works

This page explains the RankScore on each article, what ready to publish needs, and what not measured means.

A score is a set of checks on your text. It is not a prediction. No score can promise a ranking, traffic or sales. Search engines decide those.

What RankScore is

RankScore is a number from 0 to 100 shown on every article. It is built from 96 checks, which the app calls metrics. The checks are grouped into six pillars. Each pillar has its own score.

Pillar	What it looks at
SEO	Where the keyword sits, headings, title and description length, schema, internal links, length against the pages that rank
Helpfulness	Whether the article matches what the searcher wants, covers the topic, answers the questions people ask, and gives the answer early
Originality	How much the article says that the top pages do not, and whether it repeats your own other articles
Trust	Whether facts have a source that supports them, whether first-hand claims are attributed, source quality, a named author, disclosures
Readability	Reading grade for your reader, long sentences, scannability, spelling for your chosen region
Human	Stock phrases, repeated sentence shapes, rhythm, vocabulary range, and how close the writing is to your voice samples

The overall score is a weighted mix of the pillar scores. Open the Scores tab on an article to see each pillar and each check. Select a check to see how it is worked out and what to change.

The grades shown against the overall score are:

Excellent: 90 or more
Ready: 80 to 89
Needs work: 70 to 79
Not ready: under 70

What "ready to publish" needs

A high score alone is not enough. An article is marked ready to publish only when all of these are true:

Every blocking check (gate) passes.
The overall score is 80 or more.
Every pillar that was scored is 70 or more.
On health, money, legal and safety topics, the Trust pillar is 75 or more.
No link or picture placeholders are left in the text. See Publishing to WordPress.

Blocking checks

These are the checks that stop an article being marked ready, whatever the score:

Every fact has a source that supports it. Figures that appear in no source, and cited sentences the source does not back up, fail this check.
First-hand claims are attributed. A sentence such as "we found" or "in our experience" must name who it comes from, using your Experience page at /app/vault. The engine does not invent personal stories.
Not copied from a page that ranks. The check compares runs of eight words against the top pages. Eight percent or more shared text fails it.
No hidden or look-alike characters.
Exactly one main heading (H1).
Reviewed by a qualified person, on health, money, legal and safety topics only. Choose a reviewer on the article. Naming a reviewer does not do the review. Only name someone who has actually reviewed it. See the AI content policy.
Has what this kind of article needs. For example, a how-to needs steps.
Other checks on risky claims, such as health claims, absolute claims and a missing affiliate disclosure, can also block publishing.

The Scores tab shows each gate with a tick or a cross and a short reason.

On the article's review panel, the "Mark ready to publish" button is switched off while any blocking check fails.

Metrics that are "not measured"

Sometimes a check cannot be worked out for an article. The score then says "not measured" and gives the reason. The check is left out of the score. The app does not fill the gap with a made-up number.

Common reasons:

No claims could be read from the ranking pages, so originality cannot be compared.
The article cites none of its sources, or the citations have not been checked yet.
The separate helpfulness review has not run, or could not run.
There are no voice samples to compare against. Add 300 or more words of your own writing to a voice.
The article is too short for the check (for example, too few sentences).
The article is not in English. Some checks use English word lists and are skipped for other languages.
On the Trial and Starter plans, some review-based checks show as locked: trust signals, intro quality, voice match, brief adherence, information gain, helpful content review and promise fulfilled. They are checked on Essential Plus and above.
A few Human checks need extra tools that the site owner has to switch on. Until then they show "not applicable".

Each pillar also shows how much of it could be measured. If less than 60 percent of a pillar's weight was measured, the pillar is marked "Partial" and is left out of the overall score. You can see the share measured under the pillar bar.

The separate review

Some checks need judgement: does this passage back up this sentence, does the intro give a clear promise, is the article helpful. A separate review makes these calls. It runs as its own request with its own instructions. It never sees the writer's instructions.

The review has to quote the article to back up each answer. Answers it cannot back with a quote are left out.

The app tells you how the review was done. It says one of these:

Checked by a different model from the one that wrote it.
Checked by a separate blind review on the site's AI model, with its own instructions and none of the writer's prompt.
Checked by a separate call on the same model, when only one model is set up.

The Human pillar

The Human pillar measures writing that readers notice: varied rhythm, plain words, no stock phrases, and your own voice. You raise it by writing better. SEO-AI does not offer to get past AI detectors and does not claim that content is undetectable. On Pro, Agency and Enterprise plans, the site may offer an outside detector as a second opinion. Treat that as a signal only. It can be wrong.

Fixing issues

Each issue points to the exact words and has a fix. Some fixes are made for you:

Free fixes swap a stock phrase for a plainer one.
Other fixes rewrite only the affected sentences.
Some issues need you, such as adding your own experience, choosing a reviewer or editing a claim.

After you fix or edit, the article is scored again. Your plan sets how many automatic fix rounds run. A round is undone if it makes things worse.

Beside the score

Panels next to the score explain the text from other angles without changing the score. They cover reading ease formulas, accessibility problems such as skipped headings, and technical checks on meta tags and schema.

The free content scorer

You can score text you wrote elsewhere at /tools/content-scorer. See Free tools. It runs automatic checks on the text you paste. It does not have your sources, your brief or a reviewer, so it cannot judge those.

What you still need to do

A score helps you find problems. It does not replace reading the article. See Sources and fact checking for what to check before you publish.