Updated
Updated · XDA Developers · Jul 19
One-Line Prompt Restores NotebookLM Source Discipline After Gemini Rebrand, Tested Across 3 Versions
Updated
Updated · XDA Developers · Jul 19

One-Line Prompt Restores NotebookLM Source Discipline After Gemini Rebrand, Tested Across 3 Versions

1 articles · Updated · XDA Developers · Jul 19

Summary

  • A single appended sentence made NotebookLM answer only from provided documents, cite supporting sources, flag gaps, and stop offering web searches or extra tasks.
  • The workaround targets behavior the author says worsened around the Gemini Notebook rebrand, when the tool began suggesting web lookups, comparisons, and unsolicited outputs instead of stopping at source limits.
  • Three prompt tests shaped the final wording: one version halted out-of-scope answers, another produced a source-by-source audit, but only the combined line did both while blocking an unprompted lab-report template offer.
  • Two physics-lab tasks exposed the shift most clearly—asking about "XDA" triggered a web-search suggestion without the line, while a spherometer question showed NotebookLM could identify relevant PDFs and even flag 2 illegible files.

Insights

A user's prompt fixed a Google AI. Is natural language the new settings menu?
As Google's AI changes its own rules, is user trust becoming a casualty of innovation?
Will specialized AI tools survive the industry's push for all-in-one 'super-agents'?