FeaturedCraft · 3 min read
The nine tells that make AI text obvious
Nine categories of habit that give a draft away, and what each one costs to fix. Six of them are removals. Here is the whole published taxonomy, with examples.
GenPolish · Read
Resources
Notes on editing machine text
What machine-written drafts have in common, why the usual advice about it is wrong, and what actually changes a draft. Every claim carries its source.
From the piece
In today's fast-paced world, it is important to note that a stock opener is the cheapest tell to remove.
A stock opener carries no information about the subject. Cutting it costs nothing.More in this series
Research · 6 min read
What the ChatGPT text watermark is, and what Grok has instead
OpenAI announced textGrain on 5 October 2026. Where it applies, what its detector misses, and why the narrow space in ChatGPT answers was never the mark.
Research · 6 min read
How often AI tells show up in ChatGPT, GPT-4o, Claude and o1 text
Our detection engine run over two public datasets, 85,431 human and ChatGPT answers and 300 non-fiction articles, with every figure and its limits.
Research · 6 min read
Invisible characters in AI text, and where they actually come from
Zero-width spaces survive copy and paste. Here are the eight our tools look for, what each one is for, and why none of them is a watermark.
One post a month, at most
New categories we add to the taxonomy, and what changes in the tools we publish.