Higher Education & ResearchIssue 115 records
Student Success
Five newly archived sources advance student-success evidence after 2026-09-15: a multisite GenAI feedback RCT (N=1,176) shows reflective/hybrid designs beat direct AI on delayed AI-free transfer; a semester RAG tutor RCT finds nulls on interest, self-efficacy, engagement, and tests despite high satisfaction; a disability-integrity analysis warns prohibition and detectors redistribute harm; ACC’s Digital Twin reporting shows emerging AI early-alert referral systems without outcomes yet; and a 43-university guideline study finds privacy thin and responsibility shifted to individuals. Two patterns stress agentic design over access and governed inclusion plus institutional privacy controls. Disability causal trials, advising persistence effects, and TCO remain gaps; several publisher pages stayed inaccessible.
- Evidence records
- 5
- Cross-source patterns
- 2
- Evidence classes
- 3 academic research1 independent reporting1 independent research
- Outcomes
- 3 cautionary1 mixed1 emerging
- Source freshness
- 3 recent1 older, newly relevant1 new this fortnight
- Research completed
- 2026-09-17
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
GenAI access is not durable learning—design and guidance decide transfer
The multisite hybrid/reflective feedback RCT shows delayed AI-free transfer gains when students retain evaluative agency, while the semester-long RAG tutor RCT finds nulls on interest, self-efficacy, engagement, and tests despite positive attitudes. Together they warn student-success buyers that seat access and satisfaction are weak proxies for learning.
Operating questionBefore expanding GenAI tutoring or feedback seats, which AI-free transfer (or equivalent) measures and instructional sequences are required for go-live?
Supporting evidenceKırşehir Ahi Evran University / International Journal of Educational Technology in Higher EducationUniversity of North Florida / University of Hong Kong (OSF Preprints)
Student-support GenAI needs governed inclusion and institutional privacy controls—not bans-only or individual blame
Shepherd’s disability analysis rejects prohibition/detector-first integrity as redistributing harm, while Ng et al. show university GenAI guidelines often elevate integrity and shift responsibility to individuals with thin technical privacy operationalization. ACC’s Digital Twin reporting illustrates why early-alert data fusion heightens the same governance stakes.
Operating questionWho owns assistive GenAI permissions, detector limits, and FERPA data-domain controls for student-success AI—and is liability still dumped on students?
Supporting evidenceBurnley College / International Social Sciences and Education JournalUniversity of Edinburgh / University of Illinois Urbana–ChampaignAustin Community College / Inside Higher Ed
Evidence ledger
Full record · every source keeps its link and limitations
Multisite GenAI feedback RCT: reflective and hybrid designs beat direct AI on AI-free transfer
Cluster-randomized longitudinal field experiment (N=1,176 first-year undergraduates; 48 sections; four universities; three science domains) compared peer-only, direct GenAI, reflective GenAI, and hybrid (self-evaluation + peer + GenAI) feedback. Direct GenAI improved immediate four-dimension argument-quality gain versus peer (adjusted contrast b=0.57, d=0.25). Reflective and hybrid further outperformed direct GenAI on argument-quality gain (d=0.29 and 0.40). Hybrid showed the clearest conceptual-learning advantage versus direct GenAI (b=0.25, d=0.23). On delayed AI-free transfer, reflective and hybrid significantly outperformed direct GenAI (d=0.39 and 0.48); direct GenAI did not beat peer on transfer. Process measures: reflective/hybrid raised feedback uptake and self-regulated learning versus direct GenAI.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant for U.S./international student-success and teaching-learning offices after the 2026-09-15 student-success edition flagged durable learning and FeedbackWriter post-test nulls: this previously inaccessible (Cloudflare-blocked) open article supplies multisite causal evidence that GenAI feedback design—not access alone—drives AI-independent transfer. Directly informs LMS-integrated feedback and writing-support pilots.
- Evidence and measured results
- Source findings: multilevel models with section random intercepts and institution fixed effects; baseline equivalence; high implementation fidelity (95.8%); Holm-adjusted contrasts; mediation via uptake and SRL. Distinguishes immediate product gains from delayed AI-free transfer. Author limitations: unequal time-on-task/training across conditions; peer-comment quality unmeasured; first-year science only; mediation not cycle-lagged causal proof.
- Limitations and uncertainty
- Online 2026-07-16 (before last stream run) but newly accessible after prior Cloudflare rejection on 2026-09-15. Reflective/hybrid advantages may partly reflect extra evaluative workload and GenAI training, not hybrid sequencing alone. Four institutions modeled as fixed effects; generalization beyond first-year science uncertain. Not a U.S. community-college advising trial.
Semester-long RAG GenAI tutor RCT finds nulls on interest, self-efficacy, engagement, and tests
Semester-long randomized controlled field experiment (~450 undergraduates; in-person and asynchronous online modalities) of a custom retrieval-augmented generation (RAG) GenAI tutor in introductory marketing found no statistically significant impact on individual interest, self-efficacy, engagement, or course test performance after covariates (gender, age, race, job hours, delivery mode). Students still reported enjoying the tutor and wanting it in other classes, illustrating satisfaction–learning decoupling. Authors urge investigating pedagogical value before large investments or long-term contracts.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant after recent student-success editions archived CSUN non-generative chatbot nulls and FeedbackWriter revision-without-post-test gains: this GenAI RAG tutor null was not in the archive and supplies Pass B scrutiny on semester-scale tutoring claims. Why now: complements 2026-09-15 durable-outcome gaps with a full-term GenAI tutor achievement null.
- Evidence and measured results
- Source findings: difference-in-differences style tests for treatment×time; Table 2 shows insufficient evidence to reject nulls across interest/self-efficacy/engagement/performance hypotheses; positive student attitude items coexist with null learning metrics. Limitations: single course/instructor (n-of-one); marketing vocabulary-heavy domain; tutor lacked cross-session memory; preprint not journal version (CHBR HTML inaccessible this run).
- Limitations and uncertainty
- Preprint dated April 2025 (older than last run); included for newly relevant GenAI-tutor null not previously archived. Single-course generalizability limited; journal CHBR page could not be opened (404/access error). Satisfaction metrics must not be sold as learning evidence.
Against prohibition: GenAI bans can harm disabled students; govern instead of detect-and-ban
Conceptual analysis (social model of disability, Capability Approach, CDF-AIED) argues outright GenAI prohibition redistributes harm toward disabled students by removing assistive affordances (executive function scaffolding, format transformation, communication support, reduced disclosure dependence) while AI-detection enforcement produces discriminatory false positives. Author contends biases justify institutional governance and co-design—not exclusion—because banned systems cannot be audited or redesigned from within. Specifies conditions for governed inclusion including participatory design and disaggregated evaluation.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant to Pass B disability/integrity gaps after 2026-09-15 archived a narrative disability access review: this July 2026 open article adds the prohibition/detection discrimination and assistive-affordance argument not previously archived for student-success.
- Evidence and measured results
- Source is conceptual/normative analysis citing literatures; not a causal trial of learning or accommodation outcomes. Distinguishes claims: documented detection false-positive risks and assistive use patterns are argued from cited research; recommended governance model is interpretation/proposal.
- Limitations and uncertainty
- Not empirical RCT; UK FE/HE framing; journal is open-access niche—treat normative claims as argumentation. Does not measure learning gains from GenAI for disabled students. Published July 2026 (pre-last-run) but newly archived for integrity×disability gap.
ACC Digital Twin aims to speed AI-assisted referrals to advising and tutoring; outcomes unmeasured
Reporting describes Austin Community College’s Digital Twin Initiative (Trellis Foundation $875,000 grant): integrate cross-office student data and AI/virtual assistant features to shorten time from early alerts to tutoring, advising, financial aid, mental health, and basic-needs referrals. Soft launch targeting ~46,000 active credit students; leaders emphasize advisor dashboards for caseloads of ~500–600 and insist AI should not replace human conversations. EAB survey context: most student-success leaders want AI for routine tasks while keeping staff as the norm. No evaluated effects on persistence, GPA, or equity outcomes are reported.
Why it matters, evidence and limitations
- Why it matters
- Published 2026-09-11 and absent from archive; fills advising/early-alert implementation gap left open after CSUN admin-task nulls—shows what large community colleges are deploying now without yet producing causal student-success results.
- Evidence and measured results
- Independent reporting with on-record ACC leaders and cited EAB survey percentages. Distinguishes institutional claims/plans from measured outcomes (none yet). Not a trial.
- Limitations and uncertainty
- Reporting only; soft launch timing; no published impact metrics; vendor/platform internals not independently audited in the article. Grant-funded initiative may not generalize.
Analysis of 43 universities’ GenAI guidelines: integrity-heavy, privacy thin, responsibility shifted to individuals
Qualitative content analysis of university-level GenAI guidelines from 43 universities spanning 12 countries finds institutions recognize privacy/security concerns but often prioritize academic integrity framings; technical safeguards are hard to adapt; guidelines frequently shift responsibility onto individuals; IT sometimes operationalizes risk assessment/monitoring. FERPA/GDPR context discussed; recommendations include clearer stakeholder engagement and stronger institutional (not only individual) controls.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant Pass B privacy/governance scrutiny for student-success AI advising and tutoring deployments (e.g., ACC-style data twins): not previously archived; why now alongside emerging campus AI student-support rollouts that aggregate sensitive data.
- Evidence and measured results
- Document analysis of published guidelines (not enforcement audits or measured breach rates). Sampling via QS rankings may bias toward research-intensive institutions. Distinguishes observed guideline text patterns from actual campus practice.
- Limitations and uncertainty
- Guideline text ≠ implementation; QS-top skew; arXiv/PETS formatting with placeholder venue fields; July 2026 date predates last run but fills unarchived privacy-guideline gap.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Independent research
- Research conducted outside the implementing organization.