Study the Shift: Qualitative Analysis of Student AI Use

Campus teams are seeing students turn to generative chat bots for companionship and emotional support. Ashley Mowreader’s August 21, 2025 Inside Higher Ed piece (link below) frames the problem: therapy/companionship is a top social-media use of GenAI (HBR, April 2025) and a 2025 MIT/OpenAI study links high ChatGPT use with increased dependency for a minority of users. In this post you’ll get a concise, reproducible workflow for qualitative analysis of student AI use (from transcripts and platform logs to themes, segments, and risk flags) and three concrete ways Evidano (www.evidano.com) speeds that work while keeping data private.

Fast take & source

What to know in one paragraph:

Students increasingly ask chat bots for emotional support; counseling centers must identify dependency risks and design early interventions.
Key sources: Inside Higher Ed (Aug 21, 2025) reporting Joy Himmel at Old Dominion University, Harvard Business Review (Apr 2025) on top GenAI use cases, and MIT/OpenAI 2025 studies on dependency.
Use-case for researchers & student affairs: qualitatively classify how students use AI, measure prevalence by cohort, and surface quotes and risk signals to frontline staff.

Original article: www.insidehighered.com/news/student-success/health-wellness/2025/08/21/helping-college-students-emotionally-they-turn-ai

Findings snapshot

Date	Finding	Value / Detail	Source	Implication
Aug 21, 2025	Counseling center perspective	Joy Himmel (Old Dominion) reports more students seeking immediate online support	www.insidehighered.com (Aug 21, 2025)	Prioritize rapid screening + digital education
Apr 2025	Top GenAI use case	‘Therapy/companionship’ recognized as No.1 social-media use	hbr.org/2025/04/how-people-are-really-using-gen-ai-in-2025	Track emotional-use themes in corpora
2025	Usage vs dependency	MIT/OpenAI study: heavy ChatGPT users show higher correlation with loneliness/dependence (affects a minority)	OpenAI / MIT Media Lab (2025)	Identify heavy-use cohorts and behaviors
2024–25	Student comfort with AI mental-health suggestions	WGU Labs: 41% comfortable, 38% uncomfortable	wgulabs.org (2025)	Segment opinions before prescribing digital interventions

What happened, plain English

In 2025 researchers and campus clinicians noticed a behavior shift: some students use chat bots for immediate emotional support rather than waiting for in-person counseling. Reports (HBR, MIT/OpenAI, WGU Labs) document that therapy/companionship is a leading social use case and that heavy usage correlates, for a subset, with increased loneliness and dependence.

Why this matters: counseling centers are capacity-limited and students expect immediate help; unmoderated AI companionship can create attachment and social skill gaps.
Observable signals: students declining group work, preferring asynchronous digital contact, referencing AI 'friends' in interviews or logs.
Analytic goal: quantify prevalence (how many, which cohorts), characterize modalities (what prompts/users say), and flag high-risk patterns for follow-up.

Implications for researchers & student affairs teams

For counseling centers

Triage focus: develop low-friction digital screening to detect heavy AI-companion use and loneliness signals; embed education about AI limits and referral paths.

Operational call: partner with IT/analytics to export anonymized logs, peer-support transcripts, and intake notes for thematic review.

For UX/research teams

Measure differential attitudes: run cross-segment comparisons (residential vs. online students, by year, by prior mental-health history) to identify hotspots.

Design requirement: consent-first data collection, clear de-identification, and coding frameworks that distinguish companion-seeking language from clinical crises.

For policy & retention analysts

Link signals to outcomes: correlate AI-companion patterns with attendance, course withdrawal, and retention metrics over 30/60/90 days.

Mitigation: prioritize proactive outreach when a cohort shows rising dependence signals.

Do more, faster with Evidano

Problem: scattered, multilingual, messy inputs

Solution: ingest interview transcripts, counseling intake notes, peer‑support chat logs, and platform scraped text. Use Evidano transcription + translation with custom dictionaries to preserve clinical terms and campus jargon.

Problem: spotting rare but important risk patterns

Solution: run thematic and frequency analyses to surface low-frequency but high-risk phrases (e.g., “my AI is my only friend”), and use co‑occurrence networks to see which behaviors cluster with loneliness or avoidance.

Problem: comparing cohorts reliably

Solution: Evidano cross-segment analysis lets you compare freshman vs. seniors, residential vs. remote, or counseling‑engaged vs. non‑engaged students, with statistical summaries and clickable exemplar quotes for each segment.

Problem: limited staff time for follow-up

Solution: export concise dashboards, automated risk‑flag reports, and an AI chat assistant over your corpus so counselors can query themes and pull context-ready quotes for outreach.

Security & ethics

Evidano uses end-to-end encryption, offers PII redaction, and does not use customer data to train third-party models, enabling research-grade privacy for sensitive mental‑health work.

Checklist: 7-step workflow to reproduce this analysis (2–3 weeks pilot)

Step-by-step run-book:

1) Assemble inputs: intake forms, anonymized chat logs, peer-support transcripts, and a short student survey about AI use (consent required).
2) Import to Evidano: upload docs/spreadsheets, enable PII redaction and custom dictionary for campus terms.
3) Auto-transcribe/translate where needed and run initial thematic extraction (Evidano auto‑themes).
4) Create and refine a codebook: import an initial code hierarchy (companionship, dependence signals, help-seeking, avoidance) and apply AI-assisted coding across corpus.
5) Run cross-segment analysis: compare prevalence and language by cohort, visualize co‑occurrence networks, and surface top exemplar quotes per segment.
6) Configure risk flags: define rule-based or model-assisted flags (e.g., repeated ‘AI friend’ mentions + social withdrawal language) and generate a weekly alert report for counseling triage.
7) Share an action brief: export a one‑page dashboard and clickable quote set for front‑line staff and iterate the codebook after the first 2 weeks.

Quick FAQ

What is qualitative analysis of student AI use and when should I run it?

It’s thematic and contextual coding of texts (transcripts, logs, survey responses) to understand why and how students use AI for emotional support. Run it when you see behavioral shifts, rising loneliness, or when preparing digital mental‑health pilots.

How do I compare segments reliably?

Use consistent codebooks, automated coding for scale, and cross-segment frequency plus significance testing. Evidano automates these comparisons and provides exemplar quotes for validation.

How secure is AI-enabled research with sensitive student data?

Treat this as human-subjects work: get consent, de-identify records, log access, and use platforms (like Evidano) that encrypt data and don’t share it to train third-party models.

Conclusion, next steps

If your counseling center, retention team or UX researchers need a reproducible way to quantify and act on student AI‑companion behaviors, start with a focused 2‑week pilot: collect consented inputs, run the 7‑step workflow above, and iterate the codebook with frontline staff.

Try it in Evidano: upload a small corpus, auto‑extract themes, and generate a risk‑flag report for your counselors in hours, not weeks (www.evidano.com).

Ethics note: this guidance is research and programmatic in nature: not clinical advice. Ensure consent, data minimization, and clinical referral pathways when working with mental‑health information.

Keep reading

Browse all articles