When creating PPTs with NotebookLM, generation sometimes stalls around 20 slides. When it stops before even reaching half the intended volume, you have to split the work and start over from scratch. This video covers a structural workaround for that problem.

Create a Design Prompt First

Pick a layout from Behance or Dribbble and capture it with GoFullPage. Upload the image to Gemini or ChatGPT and they will extract the color scheme and structure as text. The key is extracting four types: title, body, data visualization, and process. Splitting by type lets you pick and choose styles per slide later.

Designate a Master Script as the Source

Feeding multiple sources at once causes the AI to lose context. First, extract a script containing slide numbers, titles, and key data for all 40 slides in the chat. Save it as a note and convert it to a NotebookLM source. Uncheck all existing sources and activate only this script – the AI will not wander off track.

Use Code-Style Commands Instead of Natural Language

Requesting all 40 slides at once causes a mid-generation cutoff. Split the request into slides 1–20 and 21–40. Use structured system-administrator-style commands instead of natural language. Include a continuation rule telling it not to stop at page 20 but to flow directly into page 21. Merge the two generated PPTX files with the keep-original-formatting option and you are done.

Key Takeaways

  • Breaking design references into four types (title, body, visualization, process) as prompts lets you select styles per slide
  • Activating only a single master script as the source keeps the AI from losing context across materials
  • Splitting generation into segments with code-style commands produces 40 consistent slides

Source