Claude Code Workflow for Content Simplification

Bridging the gap between "too academic" and "too childish" is a classic prompt engineering challenge. I ran into this recently while trying to find science reading material for my 10-year-old. He's a strong reader, but professional science journals are too dense, and "junior" versions are often stripped of all the interesting nuance, making them boring. Instead of hunting for a non-existent middle ground, I decided to build a custom pipeline to rewrite complex science articles into a specific "Goldilocks" zone of readability.

My AI Workflow for Level-Appropriate Content

The goal wasn't just to "summarize" but to maintain the scientific integrity while adjusting the cognitive load. I spent a lot of time iterating on the prompt to control sentence length and information density per paragraph. If you're building a similar LLM agent for educational content, you can't just ask for "simple English"—you need constraints.

Here is the logic I implemented in my prompt engineering process:

1. Density Control: I explicitly told the model to limit each paragraph to one core concept. This prevents the "wall of text" effect that kills a child's interest.
2. Vocabulary Guardrails: Rather than removing complex terms, I instructed the AI to introduce the term and immediately follow it with a relatable analogy.
3. Flow Tuning: I iterated on the "burstiness" of the text, ensuring a mix of short and medium sentences to keep the rhythm engaging.

If you want to try this, here is a template I used to tune the output:

Act as an expert science communicator for children. 
Rewrite the following technical text for a 10-year-old who is an advanced reader.

Constraints:
- Sentence length: Max 15 words per sentence.
- Structure: One main idea per paragraph.
- Tone: Curious and engaging, not condescending.
- Vocabulary: Retain key scientific terms but define them using a "concept -> analogy" pattern.
- Avoid: Generic adjectives like "amazing" or "incredible"; use descriptive facts instead.

Input Text: [Paste scientific article here]

Real-World Results and Gotchas

The biggest hurdle was the "oversimplification trap." Early versions sounded like they were written for a toddler. To fix this, I stopped using the word "simple" in my prompts and started using "accessible but rigorous."

By treating this as a deployment of a specific AI workflow rather than a one-off chat, I was able to create a consistent style guide. Now, I can take a dense article from a source like Oyla and transform it into something my son actually enjoys reading without losing the actual science. It's a practical tutorial in how fine-tuning the "persona" and "constraints" of an LLM can solve a very specific niche problem that commercial products often miss.