Blog
RSS- comparisongpt-5claude
Model-by-model: who overuses what
GPT-5 loves “underscore”, Claude reaches for “nuance”, and everybody overuses “robust”. The per-model fingerprint.
ciopinno1 min read - methodology
How we measure a tic
The pipeline: 10 fixed essays, 6 genres, a judge model, and a human spot-check. What counts and what does not.
ciopinno1 min read - patternsanalysismethodology
The “it’s not X, it’s Y” reframe, quantified
We tagged 4,000 sentences. The reframe shows up once every 340 words of model output — and almost never in human drafts.
ciopinno1 min read - deep-divevocabulary
Why “delve” won't die
One word, three training incentives, and why it survives every prompt telling the model to avoid it.
ciopinno1 min read - weeklygpt-5claude
This week's AI writing tics — 2026-W30
“Delve” climbs back to #1, “tapestry” holds, and the “not X, it’s Y” reframe spikes 22% across GPT-5 and Claude.
ciopinno1 min read