uprank
Sign inGet started

Guide

AI content approval workflow: review before you publish

An AI content approval workflow is a fixed sequence an article must pass before it goes live: draft from sources, review the evidence and claims, run quality checks that can block, have a named person approve one specific version, publish to a draft first, and keep an audit trail. Its purpose is that no AI-written article reaches readers on the strength of fluent wording alone. uprank builds this in: it scores each draft against 13 hard gates, lets only an Editor or Owner approve a version that has passed them, sends edits back for re-scoring, and records approvals in an audit log.

Updated

The workflow, step by step

Works for any team and any tool that writes the draft.

  1. Step 1

    Write from sources, and save every draft as a version

    Have the AI draft from sources you gathered first, and keep each draft and each edit as its own saved version. Reviewers then read a fixed piece of text, and you can always see what changed from one version to the next.

  2. Step 2

    Review the evidence and the claims first

    Before anyone reads for style, check that each factual claim can be matched to a specific passage of a source, high-risk claims first. Fixing the facts first avoids polishing sentences that will be cut.

  3. Step 3

    Run quality checks, and make them able to block

    Check the draft against written criteria such as sourcing, originality, usefulness, brand fit, readability and risk. Decide which failures stop the article outright and which are only advice. A check that cannot stop publishing tends to be ignored.

  4. Step 4

    Have a named person approve one specific version

    A human reads the version and approves it, and the approval is recorded against that exact version and that person. Where your team is large enough, the person who approves is not the person who wrote the prompt.

  5. Step 5

    Send every edit back through the checks

    An edit after the checks, even a small one, makes a new version that needs checking again. An approval of the earlier text does not cover the new text.

  6. Step 6

    Publish to a draft first, then go live after a final look

    Send the approved version to your CMS as a draft, check how it renders there, then publish. Keep going live behind an explicit action so that nothing is published by accident.

  7. Step 7

    Keep an audit trail

    Record who approved which version and when, who edited, and who published. When a reader or a colleague questions a claim later, you can show what was checked and by whom.

How uprank implements each step

13 hard gates apply to every article. The full list, with thresholds, is on the quality gates page.

  1. 1. Write from sources, and save every draft as a version

    Sources are gathered into an Evidence Pack before the draft is written, and the draft is built from that pack. Every draft and every edit is saved as a numbered version. The article page lists the versions with their time and canonical or approved badges, and compares any two side by side.

  2. 2. Review the evidence and the claims first

    The Evidence and Claims tabs on each article show the sources linked to each claim and the result of the claim check. Each extracted claim gets one of four verdicts (directly supported, partly supported, contradicted, unsupported), and a high-risk claim that the sources do not back, or that they contradict, stops the article from passing.

  3. 3. Run quality checks, and make them able to block

    When a version is scored, it is checked against 13 hard gates: 6 score gates and 7 integrity checks. A draft the pipeline scores lands in Needs Review for a person, whether it passed or not, so a failed gate never reads as success. The Quality tab shows the result.

  4. 4. Have a named person approve one specific version

    Each version that is not yet approved has its own Approve button, and it stays disabled until that version’s current assessment has cleared every hard gate. The server checks the same rule before it approves. Approving needs the Editor role or higher, so Viewers cannot, and the approval records which version was approved. The MCP server has no approve tool, so approval is something a person does in the web app.

  5. 5. Send every edit back through the checks

    Editing saves a new version and marks the earlier quality assessments stale, and an approved or published article goes back to Needs Review. A version with no current assessment is refused when you try to approve it, push it to the CMS or publish it, until it is scored again.

  6. 6. Publish to a draft first, then go live after a final look

    The Publishing tab checks the same gate and then pushes the chosen version to Strapi as a draft, and going live is done in Strapi once you have reviewed it there. Over MCP, publication_publish only records a live publication request: it needs a credential with the publication:live scope, the exact confirmation text PUBLISH LIVE and a passing assessment, and nothing is sent to the CMS.

  7. 7. Keep an audit trail

    Settings has an Audit log of privileged actions, each with its actor and timestamp. Approvals, edits, status changes and publication events are recorded with the actor, which is a user, an MCP credential or the system. An approval’s record also stores which version was approved.

What uprank does not do

  • Require a second reviewer. Any Editor or Owner can approve a version that has passed the gates, including one they edited themselves, so separating the writer from the approver is something your team arranges.
  • Make approval a precondition of pushing a draft to the CMS or of the MCP live publication request. Those steps check the hard gates; approval records a person’s decision on a version.
  • Replace a human reading the article. The gates catch what they test for, and a person still decides whether the piece is right to publish.

Common questions

Why review AI-written content before publishing it?
Fluent text says nothing about whether a claim is true, and an AI draft can contain facts, quotes or citations that do not hold up. A review step with real authority to stop publishing is the control that catches them before readers do.
Who should approve AI-generated articles?
Someone with the subject knowledge and the authority to stop publication, and where possible not the person who wrote the prompt. In uprank an Editor or Owner can approve a version and a Viewer cannot. uprank does not require a second reviewer, so that separation is up to your team.
What happens in uprank if I edit an article after approving it?
The edit is saved as a new version, the earlier quality scores are marked stale, and an approved or published article goes back to Needs Review. The new version has to be scored again before it can be approved, pushed to your CMS or recorded as a live publication request.
Does uprank publish an article automatically once it passes?
No. Passing the gates makes a version eligible to be approved, pushed to Strapi as a draft, or recorded as a live publication request over MCP; it does not take anything live. The Publishing tab creates drafts in Strapi, and going live happens in Strapi after review. Over MCP, publication_publish records a live request and needs the publication:live scope, a scope that is off unless someone grants it by hand.

Keep reading

Create your workspace