SEO content · written by AI · scored by an independent editor
§ How it works
Rankwrite is run by cooperating AI sessions, one per job, supervised by a human owner. Each article moves through the same desks, and the editor that scores it never wrote it. Below is how one article travels, who does what, and the exact rubric it is judged against.
01 One article’s journey
Illustrative example. Finch HR is a fictional company used to show how the process works. It is not a customer, and the scores below are not real results.
Day −10
Brief Writer
Intent is comparison: six sections, four entities, three internal links, two questions for the expert.
Day −8
Researcher
Eleven sources, eight primary. Pricing facts extracted with citations; the expert’s answers added.
Day −7
Writer
1,700 words in the client’s voice, citing as it goes.
Day −7 → −6
Editor
Scores 71: returned with the passages quoted. The second draft scores 88 and passes.
↺ Draft 1 returned to Write at 71
Day −6
Designer
A featured image and a comparison chart from the article’s own pricing data.
Day −6
Publisher
Pushed to WordPress as a draft, with schema and meta. Never live first.
Day −3
You
The client approves with one wording change. It goes live and is tracked.
The Managing Editor sees that ‘How to choose a payroll provider for a 20-person startup’ is due in ten days for Finch HR, and posts a brief request.
The Brief Writer pulls the search results, sees the intent is comparison, outlines six sections, lists four entities and three internal links, and flags two questions for Finch’s expert. The Account Manager emails them.
The Researcher gathers eleven sources (eight primary), extracts pricing facts with citations, adds the expert’s answers and posts the research pack.
The Writer drafts 1,700 words in Finch’s voice. The Editor scores it 71: two unsupported claims, one filler paragraph. It goes back to the Writer with the passages quoted.
The second draft scores 88. The Designer produces a featured image and a comparison chart from the pricing data. The Publisher pushes it to WordPress as a draft with schema and meta; the Account Manager tells Finch it is ready to review, with the score and rubric notes.
Finch approves with one wording change. The Publisher goes live and the Analyst registers the URL. The monthly report four weeks later shows its first impressions.
02 Two principles
Principle 1
Claims it cannot verify are cut, not softened.
The writer and the editor are separate AI sessions. The editor sees the draft, the brief, the sources and your voice document, and never the writer’s reasoning.
It checks every number, date and name against its source, runs plagiarism and AI-pattern checks, and does not round up because a deadline is close.
Principle 2
You review before anything goes live.
Publishing is draft-first by default: the article arrives in your CMS as a draft, with its score and the editor’s notes in your portal, for you to approve or send back.
Publishing live to a site that is set to drafts needs the human owner’s approval, and factual errors that reach your site are corrected within 24 hours.
03 Who does what
Owns every customer’s calendar, assigns briefs and enforces turnaround.
Builds the keyword and topic map: clusters, priorities and a 90-day calendar.
Turns each calendar item into a complete brief, ten days before it is due.
Gathers verified sources, extracts facts with citations, folds in your expert’s answers.
Writes from the brief, the research pack and your voice document, citing as it goes.
Scores against the rubric, verifies citations, runs originality checks, passes or returns.
Produces the featured image, diagrams and charts, and alt text.
Formats for your CMS with schema, meta and internal links; verifies the render.
Tracks Search Console data per URL, writes the monthly report, flags decaying posts.
Your contact: onboarding interview, expert questions, revisions and reports.
04 The public rubric
Before judging style, the editor verifies: every citation is opened, and every number, date, name or quoted figure is checked against its source and recorded as verified, mismatched or unverifiable. Every factual sentence needs a citation or must be common knowledge for the audience.
Then each weighted criterion is scored from 0 to 5, and the weighted total is normalised to 100. Hard criteria are pass or fail: one failure fails the article whatever its total.
80/100 to pass
The pass rule
An article passes only if its total is at least 80, it has no hard fail, and Accuracy, Sourcing and Intent match each score at least 3 out of 5.
A draft that fails goes back to the writer with the exact passage quoted, the criterion named and what would pass. After three failed attempts it is escalated for a new brief or new research, never passed.
Each scored 0–5. Bars show each criterion’s weight.
| # | Criterion | Weight |
|---|---|---|
| 1 | Accuracymin 3/5All checked facts verified; dated facts phrased as such. | 15 |
| 2 | Sourcingmin 3/5Claims cited; sources primary where numbers are involved; sources recent enough. | 10 |
| 3 | Intent matchmin 3/5Answers the target query as the top results’ readers expect, in the right format. | 10 |
| 4 | AngleSays something the top results do not, from the customer’s position. | 8 |
| 5 | CompletenessCovers every must-answer item and the outline’s table stakes. | 8 |
| 6 | SpecificityConcrete examples, numbers, named things; no generalities. | 10 |
| 7 | StructureHeadings true to content; logical order; scannable. | 5 |
| 8 | OpeningSituates the reader and promises the answer in the first 100 words. | 5 |
| 9 | VoicePoint of view, formality, tone, preferred terms and the feel of the customer’s sample paragraphs. | 8 |
| 10 | ReadabilitySentence variety, paragraph length, reading level as targeted. | 5 |
| 13 | Internal linksThe brief’s links are present, with natural anchors. | 3 |
| 14 | SEO hygieneTitle length; query in the title, the opening and one heading; no stuffing; meta fits. | 4 |
| 15 | Caveats and safetyRequired disclaimers and trade-offs present; nothing that reads as regulated advice beyond the sources. | 5 |
| 16 | Call to actionPresent, taken from the voice document, placed as briefed. | 2 |
| 17 | LengthWithin the briefed range, or the difference is justified. | 2 |
| 18 | FormattingClean Markdown; tables where there are comparisons; valid chart data. | 2 |
Any one of these fails the article, regardless of total.
11Banned phrases
Any hit from the banned filler-phrase list is an automatic fail, regardless of total.
12Originality
Plagiarism score above 5 % or an exact match over 12 words is a fail; an AI-pattern score above the policy threshold is a fail, with the flagged passages.
19Expert input(when expert input was briefed)
If expert answers exist, they are used and attributed; if absent, nothing is attributed to the expert.
20No AI self-reference
No AI self-reference and no first person outside the briefed point of view.
Struck through: In today’s fast-paced world, payroll is one more thing a small team has to get right.
Criterion 11 · Banned phrase. One hit fails the draft, whatever its total. It goes back with the phrase quoted.
With each article: the total score, and a score and a one-line note for each criterion. Here is what that looks like for the fictional Finch HR article. A real sample article with its full score and notes will be published here once our own first articles have been through the pipeline.
Read our blog, where every post shows its score →Finch HR · Draft 2 of 2
How to choose a payroll provider for a 20-person startup
Editor score 88/100
Pass| Number | Criterion and note | Weight | Mark |
|---|---|---|---|
| 1 | AccuracyEvery fee checked against the provider’s own page; prices dated. | ×15 | 5 out of 5 |
| 2 | SourcingEight of eleven sources primary; every number cites one. | ×10 | 5 out of 5 |
| 3 | Intent matchComparison format, as the results expect; the table could come sooner. | ×10 | 4 out of 5 |
| 6 | SpecificityNamed providers and fee structures; one paragraph still general. | ×10 | 4 out of 5 |
| 9 | VoicePlain, second-person tone from the voice document. | ×8 | 4 out of 5 |