Home / Notes

AI · LLM · Content pipeline

AI content with human-in-the-loop: 200+ guide articles from LLM drafts

With care law, AI mistakes are not acceptable. This is how I use a language model as a writing aid without handing it the responsibility.

Serge Zabolotny · 2026-10-18

• OpenAI API
• LLM
• Human-in-the-Loop
• Prompt Engineering
• Content Pipeline
• Next.js

The challenge

PflegeOasis is meant to offer 200+ guide articles on care law. By hand that is hardly doable, but fully automated content would be irresponsible on this topic: a wrongly stated entitlement or deadline can have real consequences for the people affected.

The pipeline

1. Draft offline: the OpenAI API creates article drafts in a separate step – never live for visitors.
2. Structured prompts: every draft gets sources and legal bases, instead of the model writing from memory.
3. Human review: every article is checked and corrected by a human before publication.
4. Publication: the article goes online only after the review.

What helps

• Put sources into the prompt so the statements stay verifiable
• Structured prompts instead of free text input, so results are comparable and reviewable
• Offline instead of live: errors are caught before publication, not after

Keep reading