Skip to main content

Reviewing and approving AI-generated content

How to review AI output, approve it in bulk or product by product, and turn your edits into reusable rules

Written by Alberto Barberis

Every generation produces content in a pending state: nothing reaches your storefront before you approve it. The Review and Approve page is where you compare current content with generated content and decide what goes live.

Step 1 - Open the review queue

  1. Go to Recent Activity in the left sidebar. The same button also appears in the Overview right after a job finishes.

  2. Find the completed Products Generation workflow.

  3. Click the purple Review and Approve button.

The page opens on a summary of the batch: the number of products processed, the attributes that were enhanced (shown as tag pills, such as title and description), the channels involved, and the pending / approved / rejected counts. Products are split into tabs by status - Pending approval, Approved, Rejected and Errors - and Scoring Guide explains how the quality score is calculated.

Recent Activity - the completed Products Generation job carries the Review and Approve button.

Option A - Bulk approval

Use this when the batch is large and quality is already on target.

  • Sort by Score to bring the lowest-scoring items to the top - those are the ones worth a closer look.

  • Opening a product shows the reasons behind the score, not just the number.

  • Once your spot checks look good, click Approve all. Reject all works the same way for a batch you want to discard entirely.

The batch summary, the status tabs, and Approve all above the Score column.

Option B - Review product by product

Use this for small batches, for your highest-value products, or while you are still tuning the configuration.

  • Click a product title to open its detail view: an Overall Score badge, a Before (Original) panel with your existing content field by field, and an After (New) panel with the AI content and per-attribute scores.

  • Use Edit on the After panel to adjust the copy. Titles become text inputs; descriptions open as raw HTML, so you are editing the actual markup.

  • Approve applies the AI content immediately and moves the product to the Approved tab.

  • Reject discards the AI output, keeps your current content and moves the product to the Rejected tab. A rejection can be undone if you change your mind.

Current and AI Generated side by side, with Approve, Reject and Edit at the top of the panels.

Preference Memory: your edits become rules

Every time you edit AI copy, a Save Preferences dialog asks whether Andromeda should remember the change and apply it to future generations. Each change is summarized as a mini-prompt, for example Use “Handmade” instead of “Handcrafted”, which you can accept or discard before saving.

  • Accepted mini-prompts are saved in Preference Memory as reusable rules.

  • Future generations apply those rules automatically, with no reconfiguration on your side.

The Save Preferences dialog turns an edit into a rule for future generations.

Why editing beats rejecting. A plain rejection discards the content without telling the AI what was wrong. An edit, even a small one, becomes a permanent rule. It is the fastest way to cut down review time on the batches that follow.

Did this answer your question?