NotebookLM in UX Research: An Honest Assessment of a Specialized AI Tool

📌 Key Takeaways

• NotebookLM works exclusively with your own sources and backs up every statement with direct quotes—making analyses transparent.

• The expanded context window (1 million tokens) allows for the simultaneous analysis of very large amounts of transcripts.

• For manifest topics—i.e., what users explicitly say—NotebookLM performs very well. It does not reliably detect latent meanings and implicit patterns.

• The new Deep Research feature expands NotebookLM to include active web research—narrowing the gap with standard LLMs.

• Chat history is now saved—a previous point of criticism that has been addressed.

• Audio overviews are currently only available in English—a relevant point for German-speaking UX teams.

• The most effective use: NotebookLM as a source archive and analysis accelerator, combined with a flexible LLM for in-depth interpretive work.

  
  

Twenty transcripts, one question: Where do you start?

You’ve just finished a round of research. Twenty interviews, 45 minutes each, three different user groups. The transcripts are in front of you. You know that the key insights are hidden somewhere in there—but where exactly?

In my work as a UX consultant, I know this moment well. In the past, manually working through the transcripts was the only way. Today, there are AI tools that help with analysis. The only question is: Which tool is right for what?

Google’s NotebookLM has been discussed for some time as a specialized alternative to general AI assistants like Claude, ChatGPT, or Gemini. In this article, I’ll take a look at what the tool can really do—concretely, without marketing fluff, and from the perspective of UX research.

What is NotebookLM—and what makes it fundamentally different?

NotebookLM is an AI-powered notebook from Google that works exclusively with sources you upload yourself. At first, that sounds like a limitation. But it isn’t—it’s the central design principle.

The technical foundation: NotebookLM uses what’s known as RAG – Retrieval-Augmented Generation. This means the model doesn’t generate answers from its general training knowledge, but first searches through your uploaded documents and draws its answers directly from them. Every statement is backed by a source reference that you can click on and read in the original text.

This is a fundamental difference from general LLMs (Large Language Models—such as Claude, ChatGPT, or Gemini), which have been trained on massive amounts of text from the internet and synthesize answers from this collective knowledge.

Technical Basis: NotebookLM runs on Google’s Gemini model. The free version is sufficient for getting started; NotebookLM Plus and Google AI Ultra for Business offer higher limits and advanced features.

Important to understand: NotebookLM is currently evolving from a pure source archive into an active research tool. The new Deep Research feature (more on that shortly) significantly expands the original concept.

Which features are truly relevant for UX research?

Not every new feature is equally useful for day-to-day research. Here are the features that actually make a difference:

Source-based work with direct citations

This is the core benefit. When you ask NotebookLM, “What problems did users mention regarding the checkout process?”—you don’t just get a summary, but direct references to the exact text passages in your transcripts. This is invaluable when working with stakeholders: “User 7 said this” is no longer just your interpretation—it’s verifiable.

Expanded context window for large datasets

NotebookLM now supports Gemini’s full 1-million-token context window. In practice, this means you can load a large number of transcripts simultaneously and query them across the board. “Which topic appears in all three user groups?” is a question the tool can answer across the entire dataset.

Supported file formats

NotebookLM now accepts: PDFs, Google Docs, Word documents, images (with OCR—meaning handwritten notes or scanned documents are automatically converted into readable text), CSV files, and YouTube video transcripts. For UX research, this means: interview transcripts, raw survey data, workshop photos with Post-its, and support ticket exports can all be consolidated into a single notebook.

Deep Research: Active Web Research

This is the latest and most conceptually significant enhancement. With Deep Research, NotebookLM can not only search your own sources but also actively research the web and integrate new information into your notebook. For UX researchers, this means: You can combine your own study data with current market data or background research—without switching tools.

Note: The exact feature set of Deep Research varies depending on your plan. I recommend checking the current documentation on notebooklm.google before relying on this feature.

Generated Outputs

NotebookLM can automatically generate the following from your sources: summaries, FAQs, study guides, flashcards, briefing documents, and—new—slide decks and infographics. These outputs are starting points, not finished analyses. But as a first step before the actual analysis, they can save you a significant amount of time.

Audio Overviews – with honest assessments

One of the best-known features: NotebookLM generates a podcast-like dialogue between two AI voices from your sources. There are now several formats—Letter, Critique, Debate, Deep Dive—as well as an interactive mode where you can ask questions while listening.

Important note for German-speaking teams: Audio Overviews are currently only available in English (as of March 2026). Please check the latest status at: https://notebooklm.google

Persistent Workspace with Saved Chat History

A previous point of criticism—that the conversation history was lost after closing the browser—has since been resolved. Your notebooks and conversation histories remain saved. This makes iterative work across multiple sessions significantly more convenient.

Interactive Mind Map

A newer feature that visually represents connections between topics in your sources. For exploratory analysis phases—when you don’t yet know which clusters are emerging—this can be a helpful starting point.

Real-world scenarios: Where NotebookLM really helps UX researchers

Transcript analysis with source attribution

This is the strongest use case. You upload all transcripts and ask specific questions: “Which users mentioned problems with navigation?” or “Show me all instances where users spoke about competing products without being asked.” Every answer can be verified by clicking on the original passage.

This has a practical implication for collaboration: Stakeholders can no longer say, “That’s just your interpretation”—they see the original quotes directly.

Triangulation Across Different Data Sources

Do you have interview transcripts, survey open-ended text fields, and support ticket exports? In NotebookLM, you can load all three sources into a single notebook and query them across the board: “Which problems appear in both interviews and support tickets?” Methodological triangulation—that is, combining different research methods—becomes significantly more practical as a result.

Iterative Usability Testing

You’re testing a prototype across three iterations. Create three source clusters (V1, V2, V3) and ask targeted comparison questions: “Which issues from V1 were no longer mentioned in V3?” or “What new points of criticism emerged in V3?” The evolution of user perception across iterations becomes visible.

Quick Stakeholder Briefings

The Product Owner wants “the three most important usability issues with quotes”—by tomorrow morning. NotebookLM can generate a structured briefing document with original quotes that links directly to the sources. No retyping, no manual quote searches.

Team Collaboration and Knowledge Sharing

A research archive notebook containing reports from the last two years. New team members can ask interactive questions: “What do we know about the 50+ target group?” or “Which navigation problems have already been investigated?” Institutional knowledge becomes accessible—not just archived.

Audio Overview as a Communication Format

A hypothetical scenario: Instead of producing five separate presentations for five quarterly studies, you create a 15-minute audio digest that summarizes the highlights. Target audience: Product owners, designers, and developers who want to “stay in the loop” but can’t attend every research meeting.

Important: Always transparently position the format as an “AI-generated summary”—not as a substitute for the full analysis.

Where NotebookLM Reaches Its Limits

No tool can do everything. These limitations are particularly relevant for UX researchers:

Misinterpretation rather than hallucination: NotebookLM doesn’t invent facts out of thin air—but it can misinterpret irony, sarcasm, or polite reserve in interviews. If a user says, “Yes, of course that’s totally intuitive,” while meaning the opposite, the tool may not recognize the irony. This is a real risk, especially in qualitative research.

No latent themes: NotebookLM finds what is explicitly stated—manifest content. What users communicate between the lines, what unspoken needs lie behind their statements—that remains a task for human interpretation.

No theory-driven coding: If you work with Grounded Theory, thematic analysis according to Braun & Clarke, or another methodologically grounded approach—NotebookLM cannot contribute to this process. It does not understand coding logic, does not develop theoretical concepts, and does not validate interpretations.

Limited output customization: The generated documents (summaries, briefings, FAQs) are a good starting point—but there’s little room for fine-tuning afterward. Format and level of detail are only partially customizable.

No external integrations: No direct data exchange with Dovetail, MAXQDA, Figma, or other tools in the UX research stack. The tool is a closed system.

Audio only in English: For German-speaking UX teams, this means the audio feature is currently only available to a limited extent. (As of March 2026 – please check the current status.)

Mobile lags behind: New features appear on desktop first and arrive on mobile devices with a delay.

NotebookLM vs. general LLMs – which one for what?

*Privacy note: Google states that uploaded sources and conversations are not used for model training. However, this may not apply equally to all plans. For sensitive research data, I recommend reviewing the current terms of use and, if necessary, conducting a data protection impact assessment—especially for GDPR-relevant data.

Decision Guide: NotebookLM, LLM, or Both?

Use NotebookLM if …

• you want to analyze many documents simultaneously

• source attribution and traceability are important for stakeholders

• you work in a team and want to share insights asynchronously

• you need structured outputs quickly for briefings or documentation

• you want to perform triangulation across different data sources

  
  

Use a general LLM if …

• you conduct theory-driven, interpretive analysis

• you want to uncover latent themes and implicit meanings

• you engage in an iterative analytical dialogue that develops over many conversation steps

• you need flexible output formats or code generation

• you want to think through methodological decisions in dialogue

  
  

The most sensible approach: Hybrid

In my practice, the combination works best: NotebookLM as a source archive and analysis accelerator—for the initial overview, quote searches, and stakeholder briefings. A general-purpose LLM for in-depth interpretive work—for developing codes, refining hypotheses, and situating findings within theoretical frameworks.

Together, these tools do not replace human judgment. But they take a significant amount of time-consuming routine work off your hands.

Practical Getting-Started Tips for UX Researchers

1. Start small. One notebook, one study, five transcripts. Get to know the tool before you upload your entire research archive.

2. Ask precise questions. “What did users say about navigation?” yields better results than “What are the most important findings?” – the more specific the question, the more useful the answer.

3. Actively use the citation link. Always click on the source references and check the context. The tool can take sentences out of context – especially with long documents.

4. Separate data sources sensibly. Create separate notebooks for different studies; don’t lump everything together. This keeps the analysis clean and the results traceable.

5. Treat generated outputs as hypotheses. Summaries, FAQs, and briefings are starting points—not finished analyses. Always put them to the test.

6. Be cautious with irony and subtext. For transcripts where respondents politely evade questions or use sarcasm: Manual review is indispensable here.

   
   

FAQ: Frequently Asked Questions About NotebookLM in UX Research

Can I use NotebookLM for GDPR-compliant research?

That depends on your specific context. Google states that user data is not used for model training. For personal research data, I recommend conducting a data protection impact assessment and reviewing the current terms of service—especially if you’re working with sensitive target groups. I’m not a legal advisor, but this is something you should clarify before using the tool.

How many transcripts can I realistically analyze in a single notebook?

The 1-million-token context window technically allows for a very large number of documents. In practice, I recommend loading thematically related studies together and placing unrelated projects in separate notebooks. The quality of the responses improves when the sources share a clear common focus.

Does NotebookLM replace analysis tools like MAXQDA or Dovetail?

No. NotebookLM is not a qualitative data analysis tool in the traditional sense. It has no coding function, no category system, and no structured analysis workflows. It is a very powerful source archive with an intelligent search function—but it is not a substitute for dedicated QDA software.

Can NotebookLM also work with videos or audio files?

Not directly with audio files yet. YouTube videos can be added as sources via their URL—NotebookLM then uses the automatically generated transcript. Your own audio recordings must first be transcribed before they can be integrated.

How does NotebookLM differ from a standard Gemini chat?

In standard chat, Gemini draws on its entire training knowledge and can search the web. NotebookLM limits its answers to the sources you’ve uploaded—and directly references each statement within them. This makes answers more traceable, but less flexible.

Conclusion: A powerful tool—when used in the right context

NotebookLM isn’t a jack-of-all-trades. But it’s a very useful tool for a specific part of UX research work: quickly and transparently exploring large volumes of sources.

What it does well: Searching transcripts, revealing patterns across data sources, generating stakeholder briefings with source citations, and functioning as a persistent team archive.

What it can’t do: Identify latent themes, perform theory-driven coding, integrate with external tools, or make the interpretive decisions that define qualitative research.

My recommendation: Give it a try—but with clear expectations. Start with a completed study that you’ve already analyzed. That way, you can directly compare what the tool finds and what it overlooks. That’s the most honest way to get started.

About the author:

Tara Bosenick is a UX consultant and co-owner of Uintent. Since 1999, she has been helping companies make their products more user-friendly—using sound research methods and a clear eye for what matters most. As a speaker at conferences such as Mensch & Computer and the World Usability Congress, she shares her knowledge of UX and AI. Her workshops on UX-AI prompting and AI integration embody what makes for good UX: clear benefits, direct applicability—and enjoyment of the process.

💌 Not enough yet? Then read on—in our newsletter.

Comes out four times a year. Sticks with you longer. https://www.uintent.com/de/newsletter