Blog
The weekly column on AI's accent, plus longer guides built on the sourced data.
OpenAI Promises Natural Writing From GPT-6.1 Terra; Tests Are Thin
OpenAI launched GPT-6.1 Terra billed as a clearer, more natural writer, but the only public writing test we found is small and found no clear winner.
Why do AI-writing checkers miss the most obvious AI writing?
Six regular expressions, validated on thirteen of thirteen positives, found none of the thirty-seven instances a reader found in the same fifteen documents.
Study · Aug 29Do AI-writing checkers agree with each other? I tested three
Two independent rule sets, run over the same 456 documents, reproduce three findings from the corpus study. Both also contain the same measurement mistake.
Guide · Aug 28Why does AI keep apologizing and rating its own answers?
AI prose addresses an evaluator, human prose addresses a reader. Six shapes that habit takes, and a one-second test for catching it.
Study · Aug 27Do all AI models write the same way? I measured four
160 documents, four models, one pre-ChatGPT human baseline. Em dashes are a Claude habit, the delve vocabulary is a Gemini habit, and no rule describes AI writing in general.
Guide · Aug 11I had an AI build a website that catches AI writing
The build log behind this site: the weekly robot, the honesty rules, and the embarrassing parts.
Guide · Aug 11Claude now watermarks its text. Here's what that changes.
The invisible mark proves a model touched the text, and nothing more. What writers should and shouldn't worry about.
GuideHow to tell if something was written by AI
The measurable signs, ranked, and why word lists alone stopped working.
GuideThe words AI overuses, measured
Delve, underscore, meticulous, and friends, with the actual multipliers.
GuideMake ChatGPT not sound like AI
The five rules that do the heavy lifting, and what settings can't fix.
GuideWhy does AI use so many em dashes?
The per-model rates, the fix OpenAI shipped, and the one chatbot still doing it.
This week · Oct 26Google Says Gemini 4 Argon Skips Clichés; Testers Disagree
Google says its new model avoids stock phrases and hedging, but one 140-run test and some early users report the old habits, em dashes included, are still around.
This week · Oct 19Wikipedia Puts an End Date on the Em-Dash Tell
Wikipedia's AI-writing guide now dates em-dash overuse as ending in September 2026, and the tells that replaced it keep shifting from model to model.
This week · Oct 12Labs Promise Plainer Prose, and the Tells Change Shape
Anthropic and OpenAI both say their newest models write more clearly, but the measured habits have moved rather than vanished.
This week · Oct 5Em Dashes Fade While 'This Matters' Takes Their Place
Graphite's study of 13,000 AI tells says the em dash is nearly gone from top models, and that new habits like announcing importance are filling the gap.
This week · Sep 21GPT-6 Astra Brings Bullets Back, Even as ChatGPT Cuts Them
OpenAI's own docs admit its newest flagship model defaults to bulleted, Markdown-heavy answers, undoing a plainer-prose fix from weeks earlier.
This week · Sep 14Claude's New Model Trades Stock Phrases for Run-On Paragraphs
Anthropic's own guide for Fable 5.1 admits denser prose, and a viral term names the AI 'teaser' habit that keeps reappearing.
This week · Sep 7The Load-Bearing Tic, Now With a Growth Chart
A new GitHub study quantifies Claude's favorite catchphrase, while OpenAI tunes two of its own tells out of ChatGPT.
This week · Aug 31Claude Opus 5's New Tic: Everything Is "Load-Bearing"
A single overused metaphor turned into a developer meme this week, showing how AI writing tells are model-specific and short-lived.
This week · Aug 24The Em Dash Retired. Nine New Tells Took Its Place.
A viral checklist of post-em-dash writing tics shows AI-detection folklore evolving as fast as the models it tracks.
This week · Aug 17Why AI Loves the Em Dash: The Markdown Theory
New research ties AI's em-dash habit to markdown training, while Wikipedia officially names the 'not X, it's Y' tell.