Practical writing for the people who run academic integrity: how to read a similarity score, what AI-writing percentages actually mean, and how to build policy and workflow that survive exam season. Signals inform judgment — people decide.
Perplexity, burstiness, classifier ensembles: a plain-language tour of the science behind AI-writing detection, its known failure modes, and why the number on a report is a signal for human judgment — never a verdict.
Why a 30% score can be innocent and a 4% score can hide serious misconduct. A practical guide for examiners: quotes, bibliographies, common phrases, and the source-by-source reading that actually decides cases.
Bans don't survive contact with reality, and 'anything goes' fails your strongest students. A framework for policies that name permitted AI use per assessment, define evidence standards, and keep due process at the center.
Batch intake, policy bands, triage queues and certificates: how examination offices structure a high-volume screening season so that human attention goes exactly where the evidence says it should.
Synonym spinners, translation round-trips and 'humanizers' defeat exact-match checking — but semantic similarity models compare meaning, not words. What paraphrase detection can catch today, and where honest limits remain.
A PDF is easy to edit; a signed, publicly verifiable certificate is not. How QR-secured integrity certificates let graduate schools, employers and journals check a report's authenticity in seconds.
New articles, policy templates and workflow notes — sent once a month, nothing else. Subscribe with your work email in the footer below.
Find the subscribe form at the bottom of this page
