Granola: bring your own priorities into AI meeting notes
Granola is an AI notepad for people who spend much of their time in meetings. It combines your brief notes with a transcript to create an editable record whose sources you can revisit.
Form your own approach first, then explore the reasoning to check trade-offs and validation you may have missed. There is no single right answer.
Help readers check a note's source before sharing
Scenario
You own the results page of an imagined meeting-notes tool. Users jot down short phrases and AI expands them using a transcript. Someone has mistaken “I suspect the customer finds it expensive” for something the customer actually said and shared it with the team. Design a reading flow that helps people check sources. You do not need to reproduce Granola's interface.
Conditions and requirements
- Three objects. Raw notes are what the user wrote at the time. The transcript is speech-recognition output and may contain errors. Enhanced notes are AI-organized results based on those two inputs. Source labels describe origin, not correctness.
- Scope. You may change the post-meeting note page, source-expansion area and pre-sharing prompt. You cannot add mandatory steps during meetings or prevent users from saving an unchecked private draft.
- States and actions. Readers must be able to locate evidence, correct a sentence, mark an unresolved point and recognize items ready to share. Do not rely on black versus gray text alone, or count opening a source as verification.
Use this case to check your design
The raw note says “price too high?” The transcript says “I don't know which services this plan includes.” AI writes “The customer considers the price too high.” A separate, correct item says “Send the revised explanation on Friday.” Walk through finding the first discrepancy, preserving the second item and separating unresolved content from material ready to share.
Deliverables
- Sketch the route from the results page to its source and back, including key labels and wording for two states.
- Explain how “viewed,” “edited” and “confirmed” differ, and how a mistaken confirmation can be undone.
- Propose two observation tasks to discover whether readers mistake an inference for a fact and whether they can find unresolved items. Specify observable success and failure; do not assume the study has already worked.
Explore the reasoning
- Separate origin from verification state. “From raw notes” and “from transcript” describe material. “Unresolved” and “checked” describe a person's decision. Keeping these axes separate avoids treating personal notes as inherently true.
- Expand evidence around the current sentence. Show relevant wording and enough context while preserving the return position. When notes and transcripts disagree, let the reader correct the conclusion or retain a question rather than forcing one permanently trusted source.
- Make confirmation visible and reversible. Opening, editing and confirming are distinct actions. Explain whether confirmation still applies after subsequent edits. Surface unresolved items before sharing while allowing private saving.
- Test a wrong item alongside a correct one. Ask participants to explain the basis of the price claim and handle the Friday item without unnecessary changes. Observe whether they find the conflict, preserve uncertainty and confuse source expansion with completed checking. The study has not been run; expected benefits are not results.