EdTech

Multi-model AI verification examples for edtech professionals.

A test that gets you into the university — but maybe not into the country

Duolingo's English Test manual treats admissions acceptance and visa acceptance as the same thing. Four frontier models, read in parallel, found the gap a student discovers too late.

Read the analysis

A reassuring story about AI in classrooms — built on numbers it never lets you interrogate

Turnitin's roundup tells students AI misuse is rare and mostly benign. Four models found the reassurance rests on a baseline-free statistic, three surveys treated as one, and adoption figures that understate its own sources.

Read the analysis

An AI-detection company's hallucination exposé — and the gaps in its own evidence

GPTZero found 100 fabricated citations at NeurIPS 2025. Its own report rests on an accuracy figure it never validates and a false-positive rate it admits but won't disclose.

Read the analysis

A faculty primer on AI — and the warnings it never reaches

McGraw Hill tells instructors "what you need to know" about AI. Four models found 13 warnings a single reviewer never thought to look for — hallucinated citations, FERPA, unreliable detectors, near-universal untrained student use.

Read the analysis

A new workforce metric — defined by the company selling the fix

LinkedIn Learning's Talent Velocity Report teaches a vendor-built framework as an industry standard. Four models found the disclosures it left out — and the numbers that don't hold up.

Read the analysis

A higher-ed outlook that forgets to say college still pays

ETS's 2026 trends briefing dwells on credential devaluation and a collapsing entry-level market — but never establishes that a degree still pays off, the one baseline a worried student needs most.

Read the analysis

What Pearson's "strong inclusion" guide left off the page

A practitioner guide to inclusion under Ofsted's new FE & Skills framework — missing the Equality Act and SEND duties that already make most of it law.

Read the analysis

A skills report built to be quoted — and what it quietly leaves out

Udemy's 2026 trends report is real data and a useful starting point — but its most persuasive numbers rest on baselines, methodology, and independence it never supplies. Four models found the gaps a single summary repeats right past.

Read the analysis

A post about "what modern learners expect" — that never mentions what could go wrong

Instructure's read on 2026 learners is reasonable on its face. Four models found the half it left unsaid: AI's accuracy risk, the learners flexibility forgets, and effectiveness claims with nothing behind them.

Read the analysis

The report celebrates a boom that policy was about to reverse

Kaplan's £37.4bn case for studying in the UK is built on a peak-year cohort — just as the dependant ban, a doubled health surcharge, and a shrinking Graduate Route turned the picture against the students reading it.

Read the analysis