Skip to content

The AI Article Writer

One title goes in. What comes out is a structured, sourced, reviewed article - written on your computer, by the AI account you already pay for. No API key.

  1. ResearchSources found and read
  2. Project knowledgeYour site, crawled
  3. BlueprintDecisions, per article
  4. OutlineSections and budgets
  5. DraftWritten by your AI
  6. ReviewReviewer and fact-check
  7. FinishingNo deletions allowed
  8. ExportHTML, Word, metadata

Decided in codeWritten or judged by your own AI

Most Of The Pipeline Is Not A Prompt

The stages that decide what an article should be - its type, its structure, its internal links, its coverage - are computed from your site and the brief. The model is asked to write, and to judge; it is not asked to invent the plan, and it is never asked for a number the engine can measure itself.

  • The title is the contract: the outline is closed against it.
  • Internal links come from your own crawled pages, so they point somewhere real.
  • A reviewer verdict is recorded with the article, and it is a gate rather than a note.
  • Every run is resumable for 6 days if it pauses or crashes.

The Eight Stages

  1. Research

    Sources are discovered and fetched from your machine, honouring robots.txt. What the article may later claim is limited to what this stage actually retrieved.

  2. Project knowledge

    Your site is crawled into a profile and a knowledge graph: audience, voice, entities, and the pages you already have.

  3. Blueprint

    A per-article plan decides content type, depth, structure and element mix. It is written in code, not asked of a model, so the same input gives the same shape.

  4. Outline

    Sections, their purposes and their length budgets, closed against the title so the article cannot wander off its own contract.

  5. Draft

    Your Claude or ChatGPT account writes the sections, section by section, against the outline and the project profile.

  6. Review and fact-check

    A reviewer judges the draft against the brief, and a fact-check pass judges its claims against the sources the run was given. The verdict is recorded with the article.

  7. Finishing

    A final pass tightens wording and fixes mechanics. It refuses any edit that would delete your words.

  8. Export

    Paste-ready HTML with the CMS fields beside it, a Word document using Word's own styles, and the article imported into the project so it stays self-contained.

What It Will Not Do

A short list, because it is the reason the output is usable.

  • No Invented SEO Metrics

    No search volume, keyword difficulty, CPC or traffic estimate appears anywhere in an article or a report. We do not have that data, so we do not print it.

    Target this keyword: 12,000 searches a month, difficulty 34.

    We hold no volume or difficulty data to say that from.

  • No Claims Without Evidence

    The article may not cite material the run was never given, and the fact-check verdict is part of the output rather than a hidden step.

    As Gartner noted in its 2026 report on content operations...

    No such report was among the sources the run read.

  • No Silent Deletions

    The finishing pass refuses an edit that would remove your words, so polishing cannot quietly shorten what you wrote.

    [three paragraphs removed for concision]

    The finishing pass may not delete; it is refused outright.

What SBMM Cloud Sees About An Article

The workflow, your app and engine versions, the state of the run, and a salted hash of the title. Not the title itself, not the brief, not the sources, and not a word of the article.