All use cases
EmpathizeUse with care

Analyse interviews in NotebookLM — as a second pair of eyes

Ask NotebookLM to surface patterns and source-linked answers across interview transcripts, then keep interpretation and storytelling with the researcher.

Why

Why it earns its place

It can shorten the first synthesis pass and make claims easier to trace. It cannot decide what the research means, and the reported time saving is self-reported rather than independently measured.

How

Run it

  1. Prepare the corpus

    Transcribe the interviews with an approved tool. De-identify sensitive material, check transcript accuracy, and write the research question before uploading anything.

  2. Run a grounded first pass

    Upload the authorised transcripts to NotebookLM. Ask for per-interview summaries, recurring themes, contradictions, and a direct answer, with a source citation for every material claim.

  3. Look for what is missing

    Open the cited passages. Search deliberately for exceptions, quiet voices, conflicting evidence, and claims the notebook could not support.

  4. Interpret and decide

    Rewrite the synthesis in your own analytical frame. Keep NotebookLM as groundwork; the researcher owns meaning, implications, and the story told to stakeholders.

Ready to paste

Prompt

Act as a second pair of eyes on the uploaded interview transcripts. Use only these sources to address this research question: [insert research question].

First summarise the relevant evidence interview by interview. Then identify recurring themes, exceptions, and contradictions. Finally give a concise answer to the research question.

Cite an uploaded source for every substantive claim. Separate evidence from interpretation. Say “insufficient evidence” when the transcripts do not support a conclusion.

Return: 1) brief interview summaries, 2) themes with supporting interviews, 3) exceptions and tensions, 4) a source-grounded answer, and 5) points I should verify manually.

TipKeep stable participant IDs so every cited claim is easy to audit without exposing names.

The AIdea layer

The designer’s call

Human core
Interpretation, strategic meaning, and the final story stay with the researcher because context lives beyond transcript patterns.
The trade
Faster pattern-finding ↔ less close reading. Fix it by opening every cited passage and actively seeking exceptions.
Skip AI when
Skip AI when consent or data policy does not permit transcript upload, or the corpus cannot be safely de-identified.

Check before use

Watch out

Do not upload identifiable or confidential transcripts without consent and organisational approval. Source links reduce checking effort but do not guarantee a correct interpretation; verify every consequential claim against the authorised transcript.

Source of truth

Evidence

A named researcher reports ‘at least a 70% reduction’ in qualitative-analysis time but gives no calculation, project count, comparison set, accuracy test, or independent verification.

Jenna Davies · Streamlining User Interview Analysis with Google’s NotebookLM
Explore all use cases