Fast, nuanced answers matter when institutions decide policy. Between December 2024 and February 2025 a survey of 634 Middlebury College students (over 20% of the student body) found that more than 80% use generative AI for coursework and that 61% report using it to augment learning rather than fully automate tasks. This post translates those findings into practical guidance for researchers, UX teams, and policy analysts on how to run a rigorous qualitative analysis of AI use in education and turn noisy self-reports into defensible policy recommendations. Follow the workflow below to reproduce the study’s core checks (triangulation with usage logs, thematic coding for “augmentation” vs “automation, ” and cross-segment comparisons) using Evidano (www.evidano.com) so you can move from survey to stakeholder-ready evidence in days, not months. The original coverage of the survey is on Fast Company: www.fastcompany.com/91387634/middlebury-college-students-ai-use-enhance-learning-research.
Findings Snapshot
| Date / Metric | Value | Source | Implication |
|---|---|---|---|
| Survey period | Dec 2024 – Feb 2025 | Fast Company / working paper (arXiv) | Rapid adoption window after ChatGPT launch |
| Respondents | 634 students (~20% of Middlebury) | Fast Company / arXiv | Large, single-campus sample |
| Reported AI adoption | >80% use generative AI for coursework | Fast Company / arXiv | High early adoption among students |
| Primary use: augmentation | 61% of AI users | Fast Company / arXiv | AI used as on-demand tutor & summarizer |
| Automation uses | 42% of AI users (e.g., drafts, code) | Fast Company / arXiv | Often for low-stakes or crunch periods |
| Cross-site corroboration | Data from 130+ universities, 50+ countries | Meta-datasets / Anthropic logs | Global pattern: augmentation > automation |
What happened (plain English)
Researchers surveyed 634 Middlebury students between December 2024 and February 2025 and classified reported AI activities into two buckets: augmentation (explain concepts, summarize, tutoring) and automation (writing essays, auto-coding).
To test self-report bias, the authors compared survey responses with external usage signals, including aggregated reports such as Anthropic’s education data and a broader dataset spanning 130+ universities across more than 50 countries; both supported the finding that students largely use AI to augment learning.
- Survey size: 634 students; period: Dec 2024–Feb 2025
- Augmentation reported by 61% of AI users; automation by 42%
- Authors recommend targeted policies (not blanket bans) while noting limited causal evidence on learning outcomes
Implications for researchers, UX teams, and administrators
For qualitative researchers
Treat self-reported AI use as directional, not definitive: triangulate with logs or platform telemetry when available.
Code for intent (help-seeking vs. substitution) and context (exam week, low-stakes tasks) to avoid conflating frequent use with harmful automation.
For UX and EdTech teams
Design features that support augmentation (explainers, step-through feedback) rather than enabling unmonitored automation.
Segment users by need (late-night help-seekers, course-level differences, and socioeconomic lines) to prioritize interventions.
For policy & academic leadership
Avoid one-size-fits-all bans: targeted rules and guidance preserve legitimate tutoring benefits while curbing misuse.
Commission short pilots that measure learning outcomes tied to usage types before scaling policy changes.
Do more, faster with Evidano (mapping problems to solutions)
Problem: Self-reports are noisy; Solution: Triangulation & thematic coding
Use Evidano to ingest survey spreadsheets and platform logs together, then run thematic and frequency analyses to compare reported behaviors with actual usage patterns.
Feature fit: cross-segment analysis, frequency counts, and AI chat over your corpus to probe ambiguous responses.
Problem: Inconsistent coding of 'augmentation' vs 'automation'; Solution: Reproducible codebooks
Upload a codebook once and apply AI-assisted coding to the full corpus; review suggested codes and lock decisions for reproducibility.
Feature fit: hierarchical codes → subcodes and clickable quotes for audit trails.
Problem: Need rapid follow-ups; Solution: AI avatar interviews & targeted scrapes
Run autonomous follow-up interviews to probe why students chose automation during crunch weeks, or scrape classroom forums to capture naturalistic discourse.
Feature fit: AI avatar interviewers, website/social scraping, and transcription with custom dictionaries.
Problem: Stakeholders demand visuals; Solution: Ready-made visualizations
Produce word clouds, co-occurrence networks, and cross-segment dashboards for leadership briefings within hours, not weeks.
Security note: Evidano encrypts data and does not use customer data to train third-party models (www.evidano.com).
Checklist: Run this study in 10 days
Day 1–2: Import survey CSV and relevant logs into Evidano; define demographics and segments.
Day 3: Upload codebook draft (augmentation vs automation categories) and run automated coding pass.
Day 4–5: Review and adjudicate suggested codes; lock hierarchy and extract top quotes.
Day 6: Run frequency and cross-segment analyses (by cohort, course, socioeconomic proxy).
Day 7: Triangulate with any available third-party logs or aggregated service reports (e.g., vendor summaries).
Day 8: Generate visuals (co-occurrence network, top themes) and export a stakeholder summary.
Day 9: Run targeted AI-avatar follow-ups for ambiguous clusters or recruit short interviews.
Day 10: Finalize a 2-page decision memo for administrators with recommended targeted policies and pilot metrics.
Wrapping up & next steps
Middlebury’s survey (Dec 2024–Feb 2025) shows early, widespread AI adoption with a majority of students using tools for augmentation. That signal should change how you design research and policy: measure intent and context, triangulate, and present visually defensible results.
If you want to reproduce this analysis or run a pilot across multiple campuses, start with a secure upload of your surveys and logs and let Evidano accelerate coding, triangulation, and reporting. Learn more and request a demo at www.evidano.com, or export a sample and we’ll show you a 10-day pilot tailored to your corpus.
Keep reading
- Commentary on NewsTwo Definitions: Climate Change Acceptance for UndergradsHow a PLoS One Delphi study (Aug 25, 2026) defined climate change acceptance for undergraduate science students, and how AI-enabled qualitative analysis applies it.
- Commentary on NewsResearcher-in-the-loop: AI-enabled UX researchHow the researcher-in-the-loop model governs AI-enabled UX research. Learn practical governance, stats from the August 2026 piece, and how Evidano supports this workflow.
- Commentary on NewsResearcher-in-the-Loop: Governance for AI UX ResearchGovern AI in qualitative UX research with the researcher-in-the-loop model from Jennifer L. Bowie (Aug 25, 2026): practical rules, risks, and tool mappings.