Under the hood, poketto.me makes heavy use of LLMs. The podcast feature is a great example: Users can turn any web content into a podcast, but often that content isn’t well-suited for listening. LLMs are great at optimizing this—simplifying complex sentences, turning headlines into enumerations, describing images verbally, etc.
But the challenge: How do you craft a single, generic prompt that works across all types of content and runs unsupervised via the API?
My original prompt had a long list of dos and don’ts—like “don’t repeat the task description,” or “don’t explain what you did.” Still, occasionally, the model would insert lines like “Here’s the optimized version” or “This version simplifies…” (See Exhibit A. Super annoying.)
The fix? Ask the LLMs.
I handed the task of rewriting the prompt to the LLMs themselves—ChatGPT, Grok, DeepSeek. I described the goal and asked them how the prompt should be structured. Turns out, my original prompt was fine; it just wasn’t formatted right.
After switching to the improved structure (Exhibit B), those issues disappeared. And—bonus—the podcast scripts got noticeably better.
