Learning discovery

Automated notes: discovering the real problem

We thought students needed automated notes. I led research into their wider study workflows and found capture was rarely the difficulty. Students struggled to identify, understand, and revisit what mattered, and did not trust AI output enough to study from it. I used those findings to reframe the opportunity around active learning, shape the Study Notes direction, and treat generated notes as the beginning of review rather than the final output.

Project: Genio Notes
Role: Senior UX Designer, leading discovery, synthesis, design direction, and UI through delivery
Area: AI note-taking, learning workflows, discovery research, and Study Notes
Status: Launched July 2026

Snapshot

Problem: The business assumed students needed help taking notes. Before committing engineering investment to automated note generation, we needed to know whether capture was actually the problem and what a version that genuinely supported learning would need to do.
Users: Students managing lectures, recordings, transcripts, revision, and AI tools, with a particular focus on students who cannot independently capture notes.
Goal: Decide whether to invest in automated note generation and, if so, define a V1 direction that improved learning outcomes rather than simply producing content.
Outcome: The findings shaped the shipped Study Notes product: transcript-grounded generation, structured notes, and built-in next steps for revision. V1 launched in July 2026, with around 1,000 learners using it in the first week.

Short on time? Skip to the ending

Questions

The questions driving the project.

What I needed to understand

How do students currently capture and review information during and after class?
Where do students struggle most between capture, review, revision, and assessment preparation?
How are students already using AI tools to support learning?
What do students do after receiving generated notes?
Where does trust break down in AI-generated study materials?

What I needed to validate

Would automated notes solve a meaningful student problem or duplicate behaviour students already had?
Would automation support learning or remove useful effort from the process?
What would V1 need to do to support learning outcomes rather than become a me-too summariser?

Genio Notes interface showing Cornell Notes selected with key ideas beside generated notes.

Challenge

Business trigger

Competitor activity, customer demand, and a July launch window made automated notes a strategic question.

Three pressures converged in early 2026.

Competitors were offering lecture-to-notes generation, Sales and Customer Success were fielding direct comparisons, and customers were asking for auto-generated notes from recordings, especially as an alternative to peer note-takers.

The business also needed a direction that could support students with disabilities to independently access lecture content ahead of a key accessibility-sector conference in July.

The decision discovery needed to inform was whether we should invest in automated note generation at all and, if so, what V1 needed to do to improve learning outcomes rather than just produce content.

The risk was shipping a convenient but undifferentiated summariser.

Research

Methods

Exploratory survey with 20 university students on capture and review habits
Ballpark study with 31 UK-based students on what they wanted from AI-generated notes
Ballpark study with 30 UK-based students on whether AI notes supported learning or passive consumption
Synthesis of existing accessibility research and customer conversations
How might we session with the squad to translate findings into V1 approaches

Key insights

Students were already generating notes from lecture slides and recordings, often through ChatGPT or Gemini.
23 of 31 students were already generating notes, which showed capture was solved behaviour rather than an unmet need.
27 of 30 students said AI notes supported their thinking or helped them get started, rather than replacing the work.
19 of 30 usually or almost always edited AI notes before studying from them.
After reading generated notes, students commonly wrote questions, returned to source material, or self-tested.
Trust was the adoption barrier: 14 of 30 students had revised from AI notes and later found mistakes.
Mean trust that AI notes reflected only what was said in the session was 3.77 out of 5.
Students compensated for trust issues by cross-referencing against lecture notes or original sources.
Structure was value, not polish: students preferred bullet points, clear headings, and moderate detail over exhaustive output.

Hypotheses

If generation is grounded only in the transcript, students will have more confidence that notes reflect what happened in the session.
If notes are structured around scannable headings, key ideas, and moderate detail, students will find them easier to review.
If generated notes include an obvious next step, students will be more likely to move into active revision.
If different modes support different access needs, the product can help students who cannot capture independently without removing useful effort for everyone.

Product Direction

Support understanding and revision, not just note generation.
Treat notes as the start of a learning workflow.
Ground generation in the transcript to make trust a product commitment.
Use structure as a design lever: headings, bullet points, cue columns, and scannable hierarchy.
Preserve useful effort through guided notes where students complete parts themselves.
Build next steps into the notes experience through quiz, print, and recall pathways.
Respect institutional AI policy through admin-controlled feature management.

Core Flows

Transcript-grounded generation

Students liked the convenience of generated notes but worried about accuracy, hallucinations, missing information, and generic filler.

Made transcript-grounded generation the trust guarantee behind the product direction.
Positioned no inference or invented content as a core differentiator from generic AI tools.
Focused the experience on reducing verification effort rather than asking students to blindly trust the output.

Structured notes as a product lever

Students valued structure because it made generated notes easier to understand, scan, and absorb.

Used bullet points, headings, moderate detail, and a document-style layout to make notes easier to review.
Designed the Cornell structure so key ideas and cues were visible beside the main notes rather than buried in prose.
Treated layout and hierarchy as part of the learning experience, not presentation polish.

Support without removing useful effort

Research showed students used AI to support their thinking, not avoid thinking, so the product needed to reduce barriers without bypassing active processing.

Defined provided notes for students who cannot capture independently.
Designed guided Cornell notes as a follow-up mode where students complete sections themselves.
Protected the rewriting and reconstruction moments that help students learn.

Built-in next steps

Generated notes risk becoming another static artefact unless they help students move into review, recall, and revision.

Connected notes to QuizMe so students could turn generated notes into a self-testing step.
Included print and annotation pathways for students who revise away from the screen.
Explored recall reconstruction so notes could become a prompt for active retrieval rather than only a document to read.

Admin-controlled launch

Institutions had different AI policies, and control over availability was a purchase and adoption condition.

Launched Study Notes behind admin-controlled feature management.
Allowed institutions to enable or disable the feature according to their AI policies.
Helped the team balance learner demand with institutional governance concerns.

Iteration

What did not work

Starting with comprehensive generated notes made the brief too focused on capture.
Treating notes as the final output risked creating a static artefact instead of a study workflow.
Optimising for convenience alone missed trust, verification, and active learning.

What changed

Shifted the brief from generate notes to support the study process.
Defined transcript-grounded output as the trust guarantee.
Made Cornell structure the flagship layout direction.
Added next steps from notes into quiz, print, and recall pathways.
Launched with admin-controlled feature management to support institutional AI policy.
Continued as design lead through the Cornell UI, document-style layout, and post-launch UX polish.

Impact

Helped the team avoid committing early to undifferentiated note generation.
Changed the product brief from generate notes to support the study process.
Defined the transcript-grounded trust guarantee that now leads the product positioning.
Aligned Product, Engineering, and PMM around one evidence-backed narrative.
V1 launched in July 2026, with around 1,000 learners using it in the first week.
Around 80% of learners who opened Study Notes generated a set, and nearly half of those copied the output to use elsewhere.

Outcome

Study Notes was adopted into the team’s quarterly goals
V1 provided bullet notes shipped in July 2026
Cornell guided notes shipped days later ahead of the accessibility-sector conference deadline
Success measures were set from the research: user sentiment, notes accuracy, and adoption among students with access to the feature
Post-launch exploration continued around feedback, sentiment polling, and comparing provided vs guided preferences

The discovery moved the opportunity from the capture moment, which was crowded and increasingly solved, to the study stage, where students still needed help with trust, structure, review, and active use.
It gave V1 a defensible product direction and helped the team ship Study Notes as part of a learning workflow rather than a simple summariser.

Reflection

The biggest lesson was the importance of separating a user problem from a proposed solution.
Automated notes sounded obvious, but research showed the higher-value opportunity was trust, structure, and what students did next.
The strongest product decision was treating generated notes as the start of study, not the end of the loop.
The layout of notes became part of the learning experience, not just the interface around it.
Good discovery can protect a team from optimising the wrong outcome.
In the end, the work enabled students to turn recorded information into notes they could trust, review, question, and use as a starting point for revision.

Back to portfolio