Content quality scores
The two scores RankX AI puts on a generated article. What each gate checks, why they are pass or fail rather than weighted, and what a null score means.
RankX AI puts two scores on a generated article. The quality score asks whether the piece is well made. The GEO score asks whether it is built to be extracted and cited by an answer engine. Both are the percentage of gates the piece passed, and each gate is a plain pass or fail rather than a weighted judgement.
A null score means not yet computed, not zero. Scores are computed at specific points in the pipeline, so a draft that is still being written has none.
Why gates rather than a weighted score
A content score built from weighted subjective factors is a number you can move without improving anything, and there is very little evidence that chasing one does anything at all.
So every gate here is something objectively checkable about the text: a count, a threshold, a presence test. Ten gates passed out of eleven is a statement about the document. It is not a prediction that the piece will rank.
Read the failed gates, not the number. The number is a summary; the failures are the edit list.
The quality gates
| Gate | What it checks |
|---|---|
| Word count | The piece meets its target, or a floor when it has none |
| Readability | Reading ease is above a threshold. Cited passages measurably skew simpler than uncited ones |
| Repetition | The text does not repeat itself past a threshold |
| Internal links | At least two links to your own indexable pages. Automatically passed when the site has none to link to |
| External citations | At least one outbound citation. Linking the source you quote is credibility, not leakage |
| Originality | Not too similar to your existing pages, and not too similar to a previous generation |
| Prohibited terminology | None of the phrases your brand has banned |
| Confidentiality | None of the phrases your brand has marked confidential |
| Keyword coverage | The primary keyword appears in the title, the H1 and the first hundred words |
Two of these read from your brand settings rather than from general practice, and they are the ones worth configuring: prohibited terminology and confidential phrases are how you stop a generator producing something you would have to retract.
The GEO gates
These are about extractability: whether an answer engine can lift a passage from the piece and use it.
| Gate | What it checks |
|---|---|
| Direct answer first | The piece answers its own question near the top, rather than building to it |
| Definitional opener | It opens with a definitional statement, the shape that gets lifted most |
| Question-phrased headings | At least one heading is phrased as a question a reader would ask |
| Structured question and answer coverage | The piece contains at least one explicit question-and-answer pair |
| A citation-worthy statistic | It contains a statistic that passed the claim check |
| Schema completeness | The structured data the piece needs was actually produced |
That last gate is stricter than it looks, and deliberately so. It used to check a hand-assembled list of schema types, and it passed for markup that was never generated: every article scored a point for structured data that did not exist on any page, in the database or on the customer's site. It now reads what was actually emitted, so it cannot pass for something absent.
The statistic gate is equally strict. It requires a statistic whose claim check resolved and verified, not merely that a number appears somewhere in the text.
Where the gates come from
The quality gates encode ordinary editorial standards. The GEO gates encode the best-evidenced structural findings about how AI answers are assembled: that information near the top of a document is retrieved far more reliably than information buried mid-way, that definitional sentences are what gets lifted verbatim, and that question-shaped headings map onto the sub-questions a query gets fanned out into.
None of that makes a piece good. It makes a good piece extractable, which is a different and narrower claim.
Reading a score
Below 100 is normal. Several gates are legitimately unachievable for some pieces: a purely instructional article may have no statistic worth citing, and a new site may have nothing to link to internally.
A gate that fails for a good reason is not a problem. The internal-links gate auto-passes when your site has no indexable pages, precisely because failing it there would be meaningless.
A failing keyword-coverage gate is always worth fixing, because it is mechanical: the primary keyword is missing from the title, the H1 or the opening.
A failing direct-answer gate is the most valuable one to fix, and it is usually an edit of one paragraph: move the answer to the top and stop building to it.
Scores on pieces you did not generate
The scores are computed on RankX AI's own generation pipeline. A page you wrote elsewhere and published is assessed by the Website Audit and AI Readiness instead, which check the published page rather than the draft.
Where to go next
- Generating articles.
- Content briefs, where the reader questions the GEO gates are about come from.
- AI Readiness, the same idea applied to your whole site.
Generating articles
Generating a draft from a researched brief in RankX AI. What it produces, why it never publishes, and the pre-spend checks that stop you paying for nothing.
Campaigns
Authority Campaigns group a canonical article and its channel variants around one subject. The three types, the journey a campaign follows, and when to use one.