Higher Education & ResearchIssue 294 records
Student Success
For 2026-10-05, student-success evidence moves from yesterday's question of what shows AI support worked to a wider gap: students' AI use is outrunning institutions' ability to guide it, staff it, and certify learning. Tyton Partners' Listening to Learners 2026 (1,800+ undergraduates) finds 54% of students use agentic AI weekly or more, against 20% of advisors, while only 32% of institutions have a central AI policy. Its companion Driving Toward a Degree 2026 (1,200+ staff) reports four-year public caseloads of 339 and AI features in core student systems that most staff don't know exist. A Kent State survey experiment finds employers told that AI substitutes for student effort discount high-GPA graduates and trust in-person exams most. A Wonkhe/Studiosity UK review places the residual integrity risk in validating learning where AI is allowed. Operating takeaway (interpretation): audit and enable what is already licensed, give students consistent explained AI guidance, and design scalable validation of learning before claiming AI improves student success. No source this edition reports measured outcomes from an AI intervention.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 2 independent research1 academic research1 vendor claim
- Outcomes
- 2 mixed1 emerging1 cautionary
- Source freshness
- 4 new this fortnight
- Research completed
- 2026-10-06
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Permitting AI is outpacing institutions' ability to show that students did the learning
Students' AI use, including agentic tools, now runs ahead of institutional policy and staff adoption (Tyton: 54% of students use agentic AI weekly or more; 32% of institutions have a central policy). The UK sector review places the residual risk in validating learning where AI is allowed, since vivas don't scale and declarations feel like policing. The Kent State experiment adds the external stake: when employers are told AI substitutes for effort, they discount high-GPA graduates and are reassured mainly by in-person exams. Interpretation: validation design is becoming a credential-value question, not only an integrity question.
Operating questionFor each program that permits AI, where and how is student learning validated, and can that validation be explained to students and evidenced to employers?
Supporting evidenceTyton Partners (anchor partner Gates Foundation; supporting partner Stellic)Wonkhe with Studiosity (vendor co-author)Casey Boyd-Swan and C. Lockwood Reynolds, Department of Economics, Kent State University (EdWorkingPaper 26-1604, Annenberg Institute at Brown University)
Student-support gaps are about capacity and relevance, not missing tools, including AI tools already owned
Both Tyton surveys (same publisher, so not independent confirmation) point the same way. Degree-planning tools exist at 85% of institutions but only 7% of students use them, and most non-users doubt relevance rather than lack awareness. On the staff side, caseloads keep rising (339 per advisor at four-year publics) while 58–76% of staff don't know whether their core systems already have AI features, and active use is 2–7%. Interpretation: enabling, targeting and training on what is already licensed should come before buying new AI products.
Operating questionWhich support and AI capabilities do we already license, who uses them, and what would make them relevant to the students who need them?
Supporting evidenceTyton Partners (anchor partner Gates Foundation; supporting partner Stellic)Tyton Partners (anchor partners Lumina Foundation and Riipen; research partners UERU, NACADA, PeopleGrove, Stellic)
Evidence ledger
Full record · every source keeps its link and limitations
Tyton Listening to Learners 2026: 54% of undergraduates use agentic AI weekly or more, ahead of staff; supports go unused for relevance, not awareness
Survey of 1,800+ undergraduates (May 2026) compared with faculty, advisors and administrators. Administrators are now the most active daily AI users (43%) ahead of students (32%), advisors (28%) and faculty (25%), but 54% of students use agentic AI daily or weekly versus 32% of administrators, 29% of faculty and 20% of advisors. Students whose top challenges are workload anxiety and heavy course loads are twice as likely to say AI reduces their academic burden and 10 points less likely to say they would stop if prohibited; about 1 in 2 students would use AI even if prohibited. Daily student AI users are 12 points more likely to say college is worth the cost and 18 points more likely to say their degree prepares them for a career. Tyton cites its Time for Class 2026 figure that 32% of institutions have a central AI policy. On supports: degree-planning tools exist at 85% of institutions, but only 19% of students know of them and 7% use them; among non-users, 9% cite lack of awareness while 51% are unsure the support is relevant (institutional staff blame awareness, 45%).
Why it matters, evidence and limitations
- Why it matters
- Released today. It is the largest fresh U.S. read on how undergraduates are actually using AI, including agentic tools, and on why student supports go unused. Tyton frames daily AI use as beneficial, but the evidence is correlational self-report; read it alongside the archived UMD randomized trial, where tutor access lowered grades.
- Evidence and measured results
- Cross-sectional online survey of 1,800+ undergraduates fielded May 2026, compared with institutional-stakeholder responses; percentages and percentage-point gaps reported on public web pages. Benefits for daily users are self-reported associations, not outcomes; no causal design, weighting or margin of error shown on the opened pages.
- Limitations and uncertainty
- Self-reported survey; associations between daily AI use and confidence or persistence intentions may reflect selection (motivated or career-focused students use AI more); sponsor list includes a student-success software vendor (Stellic); full report and methodology appendix not inspected beyond the public web pages; the 32% central-policy figure comes from a separate Tyton survey.
Tyton Driving Toward a Degree 2026: advisors at four-year publics carry 339 students; 21% of advisors use AI daily, and built-in AI in campus systems sits largely unused
Survey of 1,200+ administrators, advisors and frontline support staff at 750+ U.S. institutions (May–July 2026). High caseloads were the top barrier to improving advising for the sixth straight year (40%), with limited budget second (34%). Four-year public advisors average 339 students, up from 297 in 2024, and 35% of four-year publics expect support budgets to shrink. About one in five advisors (21%) use AI daily; they are the most strained (59% call their caseload unmanageable vs 46% of peers). Seven in ten staff would adopt an AI tool that improves student outcomes even if it changed their practice, and preference for a blended human–AI model rose to 42% from 32% in 2025. Across six core student-success systems (e.g., CRM, SIS, portal), 58% to 76% of respondents do not know whether their systems have built-in AI, and active use ranges from 2% to 7%. On career readiness, 84% of institutions offer job fairs but 23% rate them among the most impactful, and about 5% have a centralized employer-partnership strategy.
Why it matters, evidence and limitations
- Why it matters
- Released today. It gives the operating-capacity backdrop for AI in advising: caseloads are rising, budgets are flat or falling, and advisors adopting AI are doing so out of strain. The finding that built-in AI in already-licensed CRM/SIS/portal systems goes unknown and unused is a direct procurement signal.
- Evidence and measured results
- Cross-sectional survey of institutional staff; descriptive percentages and year-over-year comparisons from Tyton's annual series on the public release page. Staff perceptions, not student outcome data. Full report is behind a download form and was not inspected.
- Limitations and uncertainty
- Staff self-report; respondents may be skewed toward engaged institutions; sponsors and partners include vendors (Riipen, PeopleGrove, Stellic); no outcome data; the willingness-to-adopt figure is hypothetical; full report not inspected because it is form-gated.
Kent State survey experiment: when employers are told student AI use substitutes for effort, they discount high-GPA graduates and want in-person tests
About 1,750 U.S. for-profit professionals with hiring responsibilities, recruited via Prolific, were randomized to a control group, a mild framing (AI used responsibly to improve performance) or a strong framing (AI substitutes for student effort and ability) before rating two resumes. The mild framing produced no significant premium. The strong framing lowered willingness to interview and to negotiate salary, especially for high-GPA applicants: for high-GPA white male resumes the negotiated raise fell from 2.28% to 1.76% between mild and strong framings, and, for white male resumes, the high–low GPA interview gap shrank from 1.04 (control) to 0.62 (strong framing) on a 7-point scale. Effects did not differ systematically by applicant race or sex (a possible exception for white female applicants was traced to two industries). Asked what would ease concerns, both framing groups were most reassured by in-person exams; no group was very reassured by online microcredentials compared with in-person ones. The authors conclude institutions need ways to certify that students did the work themselves.
Why it matters, evidence and limitations
- Why it matters
- New this week and directly relevant to student success beyond graduation: if employers come to see student AI use as substitution, credentials, especially grades and online credentials, lose signal value. It gives campus leaders an external reason to invest in validating learning, not just permitting AI.
- Evidence and measured results
- Randomized information experiment within an online survey; respondent fixed effects; Romano–Wolf multiple-testing correction for main results; IRB approved (Kent State protocol 2423). Hypothetical hiring decisions on constructed resumes, not a field audit. Authors acknowledge power limits for subgroups.
- Limitations and uncertainty
- Hypothetical vignettes and resumes rather than real hiring; Prolific sample of for-profit employees; framing effects measure reactions to information, not actual student AI use; most respondents were not very concerned at baseline; some interview effects not significant at 95%; resumes omitted institution and major.
Wonkhe/Studiosity 'Keeping the learning real': UK sector moves from AI detection to integration, but validating learning in AI-permitted assessment remains the residual risk
Based on a desk review and three online round tables in September 2026 (two with senior institutional leaders and professional association representatives, one with student representatives), the report finds UK institutions shifting from bans and detection toward 'open' and 'closed' assessment mixes (traffic-light or Perkins-scale approaches), AI literacy, and validation of learning. It cites YouGov polling for Studiosity in which three-quarters of students reported stress over being wrongly accused of AI plagiarism. Students said institutional strategy has not always reached practice and that changes can seem arbitrary; vague briefs ('assist', 'brainstorm') burden students for whom English is not a first language and neurodivergent students. Closed assessments face wearable-device risks and inclusion concerns; vivas do not scale; declarations feel like policing when guidance is unclear. The authors place the residual risk in validating learning for open assessments. The report ends with a Studiosity product section.
Why it matters, evidence and limitations
- Why it matters
- Published this weekend, it is the sharpest current sector account of where AI assessment policy is stuck: permitting AI is easier than proving learning happened. It complements the Kent State employer experiment, which finds employers reassured mainly by in-person tests.
- Evidence and measured results
- Qualitative: desk review plus three round tables; no sample sizes, institution counts, or systematic coding reported. Quantitative claim (three-quarters stressed) is from separate YouGov polling commissioned by Studiosity and not detailed here.
- Limitations and uncertainty
- Co-produced with Studiosity, which promotes its own validation product in the report; qualitative round tables with unknown participant numbers; UK regulatory and assessment context (vivas, module structures) differs from U.S. practice; the YouGov statistic is vendor-commissioned.
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.
- Independent research
- Research conducted outside the implementing organization.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.