← Home Gemini / NotebookLM’s Debate Mode Lets Two AIs Argue…
3 min
Gemini

NotebookLM’s Debate Mode Lets Two AIs Argue Both Sides of Your Thesis

promptyze
Editor · Promptowy
07.03.2026 Date
3 min Reading time
NotebookLM's Debate Mode Lets Two AIs Argue Both Sides of Your Thesis
Two AI agents, one argument. promptowy.com

If you’ve ever wished someone would just tell you where your argument falls apart before you submit it, NotebookLM’s new Debate Mode is the closest thing to that. Google has added a feature to NotebookLM that generates two opposing AI agents, each arguing a different side of a claim or thesis — and crucially, both drawing exclusively from the documents you’ve uploaded. No generic internet knowledge, no hallucinated citations. Just your sources, turned against each other.

The feature sits inside NotebookLM’s main interface, which is already built around the idea that the AI should work with your material rather than around it. Debate Mode extends that logic: feed it a research paper, a policy brief, a collection of PDFs, and then throw a thesis at it. The two agents will construct opposing arguments grounded in whatever you gave them. Think of it as automated devil’s advocate, except it actually cites the text back at you.

Opposing agents, same source material.
Opposing agents, same source material.

How It Actually Works

The setup is straightforward. You upload your source documents to NotebookLM as usual — PDFs, Google Docs, web links, copied text — then activate Debate Mode and state the claim you want tested. The system assigns one AI agent to defend the thesis and another to attack it, both constrained to the source material in your notebook. The result is a structured back-and-forth that surfaces tensions, contradictions, and counterarguments you might have glossed over.

That source-grounding is what separates this from just asking a general-purpose chatbot to steelman your argument. A model like Gemini 2.5 Pro operating without constraints will happily generate counterarguments, but they might have nothing to do with the specific literature you’re working with. Debate Mode keeps both agents anchored to your actual evidence base, which means the weaknesses it finds are weaknesses in your sources, not abstract philosophical objections.

Arguments traced back to the source.
Arguments traced back to the source.

Who This Is Actually For

The obvious audience is researchers, students, and anyone who writes arguments for a living. A PhD candidate testing a dissertation claim before the defense. A policy analyst checking whether a recommendation holds up against the documents it cites. A journalist validating a narrative thread before publication. The common thread is: you have a claim, you have sources, and you want to know where the logic breaks down before someone else points it out.

It’s less useful if your source base is thin or one-sided. Two AI agents arguing from a collection of documents that all support the same view will produce a debate that feels like a staged event — technically happening, but not particularly illuminating. The feature rewards thorough source collections. Upload the counterarguments yourself and let the system work with the full picture.

As a prompt starting point, something like this will get you a focused debate quickly:

Debate the following thesis using only the uploaded sources: [your thesis statement]. One agent should argue in favor, the other against. Cite specific passages where possible.

Or if you want to stress-test a specific section rather than a whole argument:

Focus the debate on the methodology section of my uploaded paper. Agent 1 should defend the research design; Agent 2 should challenge its validity based on the other uploaded sources.

The Honest Limitations

NotebookLM’s Debate Mode is a brainstorming accelerator, not a peer reviewer. AI-generated counterarguments can be surface-level, particularly when the source material is dense or technical. There’s also a real risk of false confidence: if the opposing agent doesn’t land strong punches, it’s tempting to conclude your argument is airtight when the reality might be that the AI just didn’t find the right angle. Human expert review still matters — this tool doesn’t replace it, it prepares you for it.

The feature also inherits the general limitations of NotebookLM’s source handling. The quality of the debate is directly proportional to the quality and breadth of what you upload. Garbage in, polite debate out.

AI prepares the argument, you decide.
AI prepares the argument, you decide.

Why This Is Worth Paying Attention To

Google has been methodically building NotebookLM into something more than a fancy document summarizer. Audio Overviews turned your research into a podcast. Debate Mode turns it into a stress test. The direction is clear: NotebookLM wants to be the tool that sits between raw research and finished output, actively helping you think rather than just retrieve. Debate Mode is the most intellectually ambitious version of that idea yet — and for anyone who writes arguments professionally, it’s worth spending an afternoon with before your next deadline.

author avatar
promptyze
promptyze
Founder · Editor · Promptowy

Piszę o AI i automatyzacji od 3 lat. Prowadzę promptowy.com.

More →