Chegg Discord
Chegg · Discovery and concept evaluationCase 2 of 5
Could academic support fit into the way students already study together on Discord, and which AI-assisted capabilities were ready to develop further?
- Contribution
- As UX Researcher II (contractor), I ran the discovery research and the concept tests.
- Timeline
- Eight weeks, 2024
- Methods
- Two rounds of unmoderated discovery interviews with 24 STEM students
- Three low-fidelity concept tests, 12 STEM students each
- Outcome Recommendations
- Different next steps for each concept. Homework help and quiz generation could move toward alpha with specific improvements; math solving needed another iteration first.
| Concept | What students raised | Next step |
|---|---|---|
| Homework help | Wanted detailed answers inside Discord, not a link elsewhere | Toward alpha |
| Math solving | Entering an equation was friction, and they expected image-to-text | Iterate first |
| Quiz generation | Wanted explanations for wrong answers, and a timer | Toward alpha |
Help had to work where students already were
Students already used Discord to study together. The question was how Chegg’s academic support could fit into that routine, and which AI-assisted capabilities were ready for further development. Interest in an AI study bot alone couldn’t answer either question.
Understanding the existing experience
I ran two rounds of unmoderated interviews with 24 STEM students recruited through UserTesting, all of whom had used Discord for studying in the previous year. The first round explored study routines and familiarity with Discord bots. The second focused on possible commands, expectations of academic support and friction in the tools students already used.
Unmoderated interviews let students describe their routines in their own time, which suited quick discovery. Splitting the rounds let the work move from understanding current behavior to examining expectations of a service that didn’t exist yet.
Evaluating the concepts
Next came three low-fidelity concept tests, with 12 STEM students per test, covering homework answers, math solving and quiz generation. Each test examined whether the commands were understandable, whether the functionality matched expectations, and what students wanted changed. Likelihood of use, perceived usefulness and expected learning value informed the recommendations.
These measures describe reactions to concepts. They don’t establish learning gains or the accuracy of any model.
What we learned
Homework help. Students wanted detailed answers inside Discord and objected to being sent somewhere else. The concept met its decision criteria, with a clear opportunity to improve how answers were delivered.
Math solving. Entering an equation was a source of friction, and students expected image-to-text support. The concept needed another iteration before alpha.
Quiz generation. Students asked for explanations when they got a question wrong, and for a timer. It could advance toward alpha with those improvements.
What the research changed
The findings showed how students’ expectations for entering questions and receiving answers affected each concept. The study gave the concepts different next steps, rather than a single verdict on “an AI study bot.”
What the evidence shows, and what it can’t
Shows
- What students expected academic help to do inside Discord.
- How expectations for entering questions and receiving answers affected each concept.
- Which concepts were ready to advance and which needed another round.
Can’t show
- A launched bot, or the outcome of an alpha.
- Any change in how much students learned.
- A single participant pool. The discovery and concept-test groups are counted separately.
Status Discovery and concept testing completed · recommendations for alpha and iteration