{"resourceId":"kim-fgli-llm-access-adoption-20260928","versions":[{"version":"external-2aecc9ebf44bea3ca61198d763c821d17b3c60ee9b34afcb55a5ef38713481ff","resource":{"id":"kim-fgli-llm-access-adoption-20260928","title":"Accessible but not adopted: FGLI college students stay at shallow chatbot GenAI use","organization":"Harvard University (Hyungsik Kim)","sector":"Higher education / first-generation low-income student success","geography":"United States (primarily Boston and New York college students)","publishedAt":"Submitted to arXiv 28 September 2026 (cs.HC)","publicationDate":"2026-09-28","eventDate":null,"sourceName":"arXiv","sourceLabel":"Academic preprint (cs.HC)","sourceUrl":"https://arxiv.org/pdf/2609.36129","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["accessibility-workforce","knowledge-work","operating-model","governance-procurement"],"finding":"Drawing on 61 interviews (15 long-form FGLI students, 3 non-FGLI students, 3 FGLI program directors, 40 intercepts), Kim finds that while most FGLI participants use LLMs, none of 55 FGLI participants reported meaningful use beyond chatbot interfaces (ChatGPT/Claude), with narrow academic/career use cases. Barriers include low perceived career value, under-estimated self-efficacy, unclear starting points, low peer exposure, and resource/time constraints. The paper derives four inferred design principles: demonstrate user-specific (esp. career) value; clarify competency needs; scaffold from fundamentals to fluency; and proactively address resource barriers. Principles were not separately experimentally tested.","sledRelevance":"Directly challenges seat-license or ‘access equals equity’ GenAI student-success strategies for first-generation and low-income undergraduates. Campuses expanding ChatGPT Edu or ambient tutors should fund career-relevant demos, peer modeling, scaffolded fluency pathways, and cost/time relief—or risk widening a depth gap even when tools are free.","evidence":"Qualitative thematic analysis of semi-structured and intercept interviews with FGLI students and program directors; adoption conceptualized as a four-mode depth gradient (chatbot → tool-augmented → agentic → programmatic). Cross-sectional U.S. East Coast sample; preprint.","architectureImplications":"Pair institutional LLM access with onboarding that demonstrates major-specific value, competency maps, and progressive labs—not a single chatbot landing page. Instrument adoption depth (modes of use) rather than login counts. Prefer low-friction free tiers and campus-hosted options that reduce token/subscription anxiety for FGLI students.","governanceImplications":"Equity dashboards should report depth-of-use and barrier themes by first-gen/Pell status, not only license uptake. Design principles are interpretive recommendations—require local co-design with FGLI programs before mandating advanced agentic tools.","securityPrivacyImplications":"Interview study itself is low-risk research; campus interventions that track FGLI status with GenAI usage logs create sensitive education records requiring FERPA minimization, purpose limitation, and care against deficit framing in automated flags.","caveats":"East Coast concentration; mostly freshmen/sophomores; cross-sectional; tools evolving monthly; design principles inferred from themes, not RCTs; single-author preprint.","streamIds":["student-success"],"roles":{"sales":"For student-success and equity leaders who just bought campus GenAI seats, this study reframes the problem: FGLI students often already have chatbots but stay at shallow, narrow use. Sell a bounded adoption-depth engagement—career-relevant demos, peer modeling, and scaffolded fluency labs with FGLI program partners—not more ambient licenses. Bound claims to qualitative evidence (none of 55 FGLI participants beyond chatbots; five barrier themes) and refuse guarantees that access alone closes equity gaps.","engineering":"Engineering should instrument GenAI platforms for depth-of-use modes and barrier signals, then ship major-specific starter workflows with low token cost and optional human coaching hooks. Integrate identity/SIS only for consented equity analytics. Constraints include East Coast qualitative evidence and no tested intervention UI. Proof-of-value is movement from chatbot-only to tool-augmented/agentic modes among FGLI cohorts plus reduced self-reported resource friction—not MAUs alone.","delivery":"Delivery should co-design with FGLI program directors a sequenced fundamentals-to-fluency curriculum, peer mentors who model deeper GenAI use, and explicit messaging about time and token costs. Ownership sits with student success and equity programs jointly with academic technology. Facilitators must support non-CS majors without deficit framing, while privacy reviews govern any FGLI-linked usage analytics. Acceptance means documented depth-mode movement for the pilot cohort and published resource-barrier mitigations—not login vanity metrics. Risks include time poverty blocking participation, subscription sticker shock, and overclaiming untested design principles as proven interventions."},"roleDetails":{"sales":{"customerProblem":"Campus GenAI access is expanding while first-gen/low-income students remain at shallow chatbot-only use.","stakeholders":["student success","FGLI/TRIO programs","teaching & learning","academic technology","career services"],"discoveryQuestions":["How do we measure GenAI adoption depth beyond logins?","Which majors see the lowest perceived career relevance?","What peer models exist for FGLI students?"],"valueHypothesis":"Career-demonstrated, scaffolded GenAI literacy can deepen FGLI adoption beyond chatbots when resource barriers are actively reduced.","potentialEngagement":"6–10 week FGLI co-designed GenAI fluency pilot with depth-of-use metrics and peer modeling.","unsupportedClaims":["Causal persistence/graduation lift","Nationwide generalizability from East Coast interviews","Validated DP1–DP4 intervention efficacy"]},"engineering":{"fit":"Campus LLM portals and literacy tooling for undergraduates with equity analytics.","architectureIntegration":"AuthN → major-tagged starter packs → usage-mode telemetry → optional mentor queue; minimize token cost defaults.","prerequisites":["FGLI program partnership","consent for equity analytics","career-relevant example bank"],"constraints":["qualitative evidence only","rapid tool churn","subscription/token cost externalities"],"security":"Treat FGLI flags and GenAI logs as sensitive education records; purpose-limit; no vendor training on student chats without contract.","proofOfValue":["Share of FGLI users beyond chatbot mode","Self-efficacy/value survey deltas","Peer-session completion rates"]},"delivery":{"work":"Recruit FGLI cohorts, run scaffolded labs, stand up peer models, monitor resource barriers.","dependencies":["FGLI advisors","career services examples","IT license/token budget"],"ownership":"Student success / FGLI program leadership with academic technology support.","skillsAdoption":"Facilitators who can demystify tools for non-CS majors without deficit framing.","governanceCheckpoints":["Co-design review with FGLI directors","mid-pilot barrier check","equity dashboard privacy review"],"acceptanceCriteria":["Documented depth-mode shift for pilot cohort","Resource-barrier mitigations published","No mandatory advanced-tool requirement without scaffolds"],"risks":["Deficit labeling","Time poverty blocking participation","Overclaiming untested design principles"]}},"retrievedAt":"2026-09-30T03:28:21Z","enrichedAt":"2026-09-30T03:31:28Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Shows GenAI equity is an accessibility-of-fluency problem: FGLI students need scaffolded pathways, peer exposure, and relief from time/money constraints—not only licenses. Workforce implication: FGLI advisors and career services become co-owners of GenAI literacy, not just IT.","procurementImplications":"RFPs should require adoption-depth metrics, major-specific onboarding, and low/no-cost student tiers. Do not accept seat counts as equity success. Prefer vendors supporting campus-hosted or capped-cost modes that address token anxiety.","operatingModelImplications":"Operate GenAI student success as a literacy and coaching program with FGLI partnership, peer modeling, and sequenced fluency—not a self-serve chatbot landing page. Track depth modes and barriers as routine ops metrics.","sourceVerification":{"openedUrl":"https://arxiv.org/pdf/2609.36129","referenceExcerpt":"none of 55 FGLI participants reported meaningful use of LLMs beyond the chatbot interface","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}