Blog

Writing centre demand is seasonal, concentrated and structurally impossible to staff at peak. The fix is not more appointments — it is moving the work that does not need a…

95% of students use AI. 38% say their institution provides it. The gap is not a policy problem you can close with a rule — it is a provisioning problem,…

The official enrolment figure released in November 2025 describes the 2023/24 academic year. If your strategy paper treats it as current, it is two years behind the students it describes.

94% of students use generative AI for assessed work. 12% directly include AI-generated text. Those two figures come from the same survey, and confusing them has driven a great deal…

Forty-one questions to send in writing before any demonstration, grouped by the committee that will ask them — and the four whose answers most often end an evaluation.

One product in the standard comparison set was withdrawn on 30 June 2026, and one major vendor sells both an AI detector and an AI humanizer. What an evidence-based renewal…

A step-by-step data protection review for an AI writing deployment, covering the UK DPIA route, the FERPA rights-holder rule US institutions get wrong, and the AI literacy duty on deployers.

A pilot that cannot fail is not a pilot. Eleven steps with the owner and the artefact for each, sized to one academic term and one department.

Most institutional AI policies fail on the same three components: they set one rule for every assessment, they leave disclosure undefined, and they say nothing about burden of proof.

A detector score is a probability, not evidence of intent. What a 1% false-positive rate means at institutional volume, and what a defensible integrity process looks like without one.