The quality score
How every draft is scored out of 100: the automatic SEO checks, the Editor's five judgments, the one revision a weak draft gets, and what the score decides.
On this page
Every draft reaches you with a quality score out of 100, shown as a ring on the review queue and explained in the draft's Quality card. It's made of two checks that catch different things.
| Part | Weight | What it is |
|---|---|---|
| SEO checks | 45% | Automatic, exact checks: length, headings, the keyword in the right places, meta tags, and that the article isn't a copy of one you have |
| The Editor | 55% | An AI reviewer that reads the article against its sources and judges five things |
The SEO checks are narrow but certain; the Editor is broad but a judgment, so it carries slightly more weight. If the Editor couldn't review a draft, the score is the SEO checks alone, and autopilot never publishes that draft on its own.
Reading the score
| Score | Label | Means |
|---|---|---|
| 85 to 100 | Strong draft | Ready to publish as it is, in most cases |
| 70 to 84 | Good, with notes | Worth reading the Editor's notes before you approve |
| Below 70 | Needs work | Read it closely, or reject it with a reason |
Under the label, the card shows the two parts ("Editor 82 · SEO checks 94"), whether the draft was revised once, and how many broken links were removed. Below that: a bar for each of the Editor's five judgments, its summary, and its notes.
The Editor
The Editor reads the finished article alongside the material it was written from and scores five things from 0 to 100. Its overall score is a judgment, not an average: a serious accuracy problem caps it at 40 however good the rest is.
| Judgment | What it asks |
|---|---|
| Accuracy | Is every claim grounded in the sources? Invented numbers, quotes and features are penalized |
| Depth | Does each section deliver specifics (numbers, named tools, worked examples) or only general advice? |
| Originality | Does it add something the pages already ranking don't: first-hand facts about your product, a real angle? |
| Structure | A logical flow of sections, easy to scan, a real opening, a useful FAQ |
| Readability | Clear, direct sentences, no throat-clearing, and paragraphs that earn their place |
It checks claims about your own product against your site knowledge, and treats your team's notes as first-hand experience rather than unsourced claims. When an article has no transcript or thread behind it, the Editor checks it against the pages the article cites instead of judging it blind.
The SEO checks
Each check either passes or doesn't. The score starts at 100; each failed hard check costs 25 points and each warning 6.
| Check | Hard |
|---|---|
| Long enough for the search (at least 85% of the length the brief asked for) | Yes |
| Has a body, and enough H2 sections for its length | Yes |
| No H1 in the body (the title is the H1) | Yes |
| Meta title 40 to 60 characters | Yes |
| Meta description 120 to 155 characters | Yes |
| Focus keyword set | Yes |
| No placeholder text | Yes |
| Doesn't mention its own research material ("according to the supplied excerpt") | Yes |
| Doesn't claim to have checked a source on a date it invented | Yes |
| Not a near-copy of an article already on your site | Yes |
| Keyword in the title, in the first 100 words, and in an H2 | |
| A direct answer before the first section | |
| Uses lists; a comparison includes a table | |
| An FAQ section, when the brief asked for one | |
| Not wildly over length | |
| Caveats said once, not repeated in every section | |
| Table cells state facts, not "check their site" | |
| Doesn't borrow long stretches, or the same outline, from one of your articles | |
| No giveaway AI phrases | |
| A slug of seven words or fewer; three to six tags |
One revision before you see it
If the Editor scores a draft below 70, or a hard check fails, Seovyn sends it back to the writer once, with everything that failed. The revision replaces the draft only if it fixes a hard failure or scores at least as well: a revision that makes things worse is thrown away. Either way, the draft is then scored again and reaches you.
Your own articles, compared
Each new article is compared with your closest existing articles. A near-copy (a fifth or more of its text shared with one of them) fails a hard check. Smaller overlaps, such as borrowed passages or the same outline, are flagged as warnings. This checks for copied text and repeated structure; a different article on the same subject passes.
What the score decides
- Autopilot. A draft must score 75 or more to approve itself. See Autopilot.
- Earning autopilot. Your last five approvals must each have scored 70 or more.
- Keeping autopilot. If recent articles average below 65, Seovyn moves you back to supervised.