AI production

Product Content Lab: bulk AI content work that survives being interrupted

Generating content for 4,000 products is not the hard part. Doing it without duplicating work, double-billing credits, or leaving the catalogue half-updated when a browser tab closes, that is the hard part. The Product Content Lab is built around that problem.

GoGee feature series · 26 of 33

How it's actually built

Job model: Bulk runs tracked per item, not per batch
Resumable: A stopped run continues from where it left off
Idempotent: Re-running does not redo completed items
Validation: Every generated payload validated before write
Auth: Authenticated, role-checked endpoints
Scope: Descriptions, alt text and imagery across the catalogue
Human gate: Generated content lands as proposals for approval

How the pieces connect

01Input
Job model
Bulk runs tracked per item, not per batch
02Deterministic
Resumable
A stopped run continues from where it left off
03Deterministic
Idempotent
Re-running does not redo completed items
04Deterministic
Validation
Every generated payload validated before write
05Output
Auth
Authenticated, role-checked endpoints

Product Content Lab, data flow, generated from the shared GoGee feature diagram template.

Per-item state is what makes bulk safe

Progress is tracked at the item level, so the run knows exactly which products are done. Interrupt it, come back tomorrow, and it resumes. Trigger the same run twice and completed items are skipped rather than regenerated, which protects both your credit balance and your published copy.

No duplicate charges for work already completed
A failure on one product does not abort the other 3,999
Progress is visible while the run is in flight

Nothing is written unvalidated

Model output is validated against the expected shape and field rules before it touches the catalogue. Content that does not validate is rejected instead of being written as a broken description or an empty alt attribute.

Generated content arrives as proposals. A human approves before the catalogue changes, the same discipline used by the AI product names and descriptions workflow.

Where it fits

Use the importer to get products in, the Content Lab to bring their content up to a publishable standard, the image editor for photography, and the feed generator to push the result to Google and Meta. That is a full catalogue-readiness pipeline with a human approval step in the middle of it.

Questions we get asked

What happens if a bulk run is interrupted?

It resumes. Progress is tracked per item, and completed items are not regenerated.

Does generated content publish automatically?

No. It lands as proposals for human approval, and every payload is validated before it is written.

Can it handle thousands of products?

Yes, that is what the resumable, idempotent design is for.