Key Findings
The most-asked questions about AI in higher education, answered directly — one finding, one number, one source. No narrative padding: every answer below traces to an open dataset you can download and check.
01How many universities publish AI guidance?
144 of 186 audited records (77%) publish AI guidance. Among US Carnegie R1 universities the share is higher: 178 of 187 (95.2%).
SOURCE → Global University AI Policy Audit 2026 + US R1 AI Policy Audit 2026
02Which R1 university has no AI policy?
The University of Wyoming — the only confirmed NO among all 187 Carnegie R1 universities. Three searches found no adopted AI policy on uwyo.edu, only a 2023 working-group report, library guides, and commission news.
SOURCE → US R1 AI Policy Audit 2026
03Do any top-100 universities ban AI campus-wide?
No. Zero of the top 100 universities impose a campus-wide AI ban. The sector has standardized on governing AI use, not prohibiting it.
04How big is the policy gap between top-50 and Global South universities?
92% of top-50 universities publish AI guidance, versus 46% of Global South flagships — a 46-point gap. Students at unguided institutions face the same AI tools with no institutional rules.
05How do Central Asian universities compare on teaching vs governing AI?
They teach AI three times faster than they govern it: 58% of 31 leading universities teach AI programs, but only 19% publish an AI policy.
SOURCE → Central Asia AI Education Audit 2026
06What share of online degree listings hide the real total price?
79% of listings required computed estimates — the real total price was not findable on official sites. The audit covers 740 online programs at 537 US institutions, with the basis of every estimate explained per row.
SOURCE → Online Degree Tuition 2026
07How often does a small RAG tutor hallucinate?
On questions the knowledge base cannot answer, a 0.5B RAG tutor hallucinates 70% of the time with top_k 3, 50% with larger chunks, and 30% with top_k 5 — even when explicitly instructed never to invent answers. 90 generations, hand-graded.
SOURCE → RAG Hallucination Benchmark
08How does a new university get recognized by academic databases?
It climbs a ladder: swot, then Wikidata, then ROR, then OpenAlex. A global census of 16,822 universities shows most new institutions stall on the early rungs. ACLAS itself is on rung one (swot).
SOURCE → Academic Database Recognition 2026
09What is the Global AI Education Readiness Index?
A ranking of how ready the world’s universities are for the AI era, built on three dimensions with published weights and open data. The methodology preprint is public; the first edition is in progress.
SOURCE → Readiness Index methodology
10What does the SI Score test?
AI literacy in 18 questions across six dimensions: prompting, verification, hallucinations, ethics, privacy, and policy. Scored 0–100 across four levels, free, and it runs entirely in your browser — answers never leave your device. Labeled Beta.
SOURCE → SI Score diagnostic
11Are Study Intelligence datasets free to use?
Yes. All six datasets are CC-BY-4.0 — free for commercial use, adaptation, and redistribution with attribution to Study Intelligence (aclas.si). Each ships as CSV + JSON with per-row source URLs.
SOURCE → Open Data Commons
12How is the data verified?
Page by page, on official university domains only — never on news reports or aggregators. Every row carries its source URL, and the parsing scripts are published alongside the data so anyone can reproduce or challenge a finding.
SOURCE → Methodology
13Which R1 universities have unclear AI policy status?
Eight: Brown, Clemson, UT Health San Antonio, UT Southwestern, UMass-Boston, Miami, UNH, and Weill Cornell Medicine. Each has real AI activity — committee reports, program-level rules, intranet-gated pages — but no adopted, public, university-wide policy on its own domain.
SOURCE → US R1 AI Policy Audit 2026
14What do university AI policies actually require?
Honestly: this audit does not measure that yet. It measures publication — whether an adopted policy exists and is public — not content depth or enforcement. Assessing what policies contain is what the Readiness Index is being built to do.
SOURCE → Readiness Index methodology
15Who runs Study Intelligence, and does it grant degrees?
Study Intelligence is the research arm of the Atlanta College of Liberal Arts and Sciences (ACLAS). It does not grant degrees — ACLAS College (aclas.college) is the degree-granting institution.
SOURCE → About Study Intelligence
16Among R1 universities with a definitive verdict, how many publish AI guidance?
178 of 179 (99.4%). Excluding the 8 unverifiable cases, only the University of Wyoming is a confirmed NO.
SOURCE → US R1 AI Policy Audit 2026
17How does policy adoption fall across the QS sample?
Top 50: 46 of 50 (92%). Ranks 51–100: 43 of 49 (88%). Ranks 101–150: 38 of 50 (76%). Global South flagships: 17 of 37 (46%). Adoption falls as rank falls — and drops sharply outside the QS sample.
18What does UNVERIFIABLE mean in the Central Asia audit?
It means no policy could be confirmed on the public web — often because sites were down, login-walled, or crawler-blocked. It is a finding about transparency, not proof that no policy exists. 25 of 31 Central Asian universities are unverifiable on this definition.
SOURCE → Central Asia AI Education Audit 2026
19Do any Central Asian universities ban AI campus-wide?
Zero. Every verified Central Asian policy follows the global norm of delegated permission plus mandatory disclosure — no campus-wide bans were found.
SOURCE → Central Asia AI Education Audit 2026
20Which degree levels does the tuition audit cover?
299 bachelor-level, 228 MBA, 108 MCS, 67 BSCS, and 38 DBA programs — 740 in total across 537 US institutions.
SOURCE → Online Degree Tuition 2026
21Does the tuition audit include ACLAS itself?
Yes, and the rows are flagged. Five operator-owned rows: MBA $89, BBA $159, DBA $199, BSCS $199, and MCS $299 — all with verified (non-estimate) prices, listed under the same rules as every other institution.
SOURCE → Online Degree Tuition 2026
22How is recognition distributed worldwide?
Of 16,822 universities in the global census: 6,817 score 4/4 (41%), 5,801 score 3/4 (34%), 1,109 score 2/4 (7%), and 3,095 score 1/4 (18%) across swot, Wikidata, ROR, and OpenAlex.
SOURCE → Academic Database Recognition 2026
23Which recognition signal is scarcest?
swot domain coverage, at 47% of the 16,822 census universities — below ROR (75%) and OpenAlex (75%). Wikidata is the census frame, so it reads 100% by construction, not by achievement.
SOURCE → Academic Database Recognition 2026
24Why does retrieval not explain RAG tutor hallucinations?
Because retrieval worked: the correct chunk was in the top-k for all 10 answerable questions, yet the tutor still failed on generation. The embedding model is not the bottleneck — the 0.5B generator is.
SOURCE → RAG Hallucination Benchmark
25What does more retrieved context do to hallucination?
Raising top_k from 3 to 5 cuts the hallucination rate on unanswerable questions from 70% to 30%, and lifts citation accuracy on answerable questions from 40% to 60%. More context is the cheapest win available.
SOURCE → RAG Hallucination Benchmark
26Of the 37 Global South flagships, how many lack published guidance?
20: 17 confirmed with no policy and 3 unverifiable. Only 17 of 37 (46%) publish guidance — the lowest adoption of any sample in the audit.
27What is the largest dataset in the commons?
The global recognition census: 16,822 universities, each checked against swot, Wikidata, ROR, and OpenAlex. The other five datasets combined hold just over 1,200 rows.
SOURCE → Academic Database Recognition 2026
From answers to evidence
These answers are the short version. The full synthesis across all six datasets is State of AI in Higher Education 2026, and every R1 verdict above traces to one page per university. The segment dashboards behind these answers are on the AI Readiness Index Dashboard.