NemoClaw Knowledge Wiki

NotebookLM Gemini Workflow Optimizing AI Prompts for Structured Output

3 min read

NotebookLM + Gemini Workflow: Optimizing AI Prompts for Structured Output

Clip title: I Built a NotebookLM + Gemini Workflow That Makes Prompt Engineering Pointless Author / channel: Craig Does AI URL: https://www.youtube.com/watch?v=W-rtNL_Uf3I

Summary

This video introduces a practical workflow designed to optimize “messy” or vague AI prompts into clean, structured, and consistent outputs using Google’s NotebookLM and Gemini. The presenter, inspired by his wife’s preference for simplicity over complex prompting frameworks, demonstrates how to achieve reliable AI responses without needing advanced prompt engineering skills. The core problem addressed is the inconsistency and often unsatisfactory results users get from AI when prompts lack clear structure and a defined “thinking guide.”

The presenter explains that most prompts fail for two main reasons: a lack of structure and an absence of a thinking guide. To combat the lack of structure, the workflow utilizes JSON code to define the exact fields and format the AI’s output should take. This prevents the AI from “guessing” what information is important or how to present it, ensuring consistent and predictable results. For the “thinking guide,” the solution involves providing “example pairs”—raw, unoptimized inputs alongside their desired, structured outputs. These examples act as a reference point, teaching the AI what a “good” response looks like and enabling it to self-correct and reason effectively before generating its final answer.

Implementing this workflow is straightforward, requiring just two files and taking approximately 5-10 minutes. Users create a new notebook in Google’s NotebookLM and upload two text files: one containing “System Instructions” (which includes the JSON schema and reasoning guidelines) and another with “Example Pairs.” These files are then converted into “sources” within NotebookLM. Finally, the NotebookLM notebook is attached to a new chat in Google Gemini (or configured as a dedicated “Gem” for repeated use). This setup seamlessly integrates the structured instructions and examples, allowing users to input simple, conversational prompts and receive optimized, consistent outputs.

The primary takeaway is that this two-pronged approach—combining JSON for output structure with example pairs for a reasoning guide—significantly enhances the quality and consistency of AI-generated content. It empowers users to get predictable, high-quality results from AI even with minimal effort in crafting their initial prompts. The workflow works with a free basic Google account, making advanced AI capabilities more accessible and user-friendly by automating the complexities of prompt optimization.

Graph View