Pete Phillips

Interesting article. At UCL, we still don't have access to Pangram to check student work, which seems crazy. 'I recently reviewed 1,191 doctoral dissertations from 2025 in the fields of educational leadership and administration. My research suggests that 35 percent of those dissertations were made up of more than a negligible amount — i.e., greater than 10 percent — of AI-generated text. Ten percent were more than 50 percent AI-generated... Before you protest — “AI-detectors are unreliable”; “they’re potentially biased against nonnative speakers”; “they can be tricked” — let me tell you that I used Pangram. It is not for nothing that Pangram has become the tool of choice for much of the internet, as a slew of peer-reviewed research has found that it achieves “near-zero” false-negative and false-positive rates, even against “adversarial” and “humanizing” AI software. I have used Pangram over the last year in my undergraduate, graduate, and doctoral courses and have found it (with additional safeguards and protocols) to be near-perfect in detecting students’ inappropriate use of AI in their writing. My ongoing study used 100 dissertations published before 2020 as a control; all came out as “100 percent Human” in Pangram.'