To work around those rules, the Humanizer skill tells Claude to switch inflated language with plain facts and offers this instance transformation:
Before: “The Statistical Institute of Catalonia was officially established in 1989, marking a pivotal moment within the evolution of regional statistics in Spain.”
After: “The Statistical Institute of Catalonia was established in 1989 to gather and publish regional statistics.”
Claude will read that and do its best as a pattern-matching machine to create an output that matches the context of the conversation or task at hand.
An example of why AI writing detection fails
Even with such a confident algorithm crafted by Wikipedia editors, we’ve previously written about why AI writing detectors don’t work reliably: There may be nothing inherently unique about human writing that reliably differentiates it from LLM writing.
One reason is that regardless that most AI language models tend toward certain sorts of language, they will also be prompted to avoid them, as with the Humanizer skill. (Although sometimes it’s very difficult, as OpenAI present in its yearslong struggle against the em dash.)
Also, humans can write in chatbot-like ways. For instance, this text likely accommodates some “AI-written traits” that trigger AI detectors regardless that it was written by an expert author—especially if we use even a single em dash—because most LLMs picked up writing techniques from examples of skilled writing scraped from the net.
Along those lines, the Wikipedia guide has a caveat value noting: While the list points out some obvious tells of, say, unaltered ChatGPT usage, it’s still composed of observations, not ironclad rules. A 2025 preprint cited on the page found that heavy users of enormous language models accurately spot AI-generated articles about 90 percent of the time. That sounds great until you realize that 10 percent are false positives, which is sufficient to potentially throw out some quality writing in pursuit of detecting AI slop.
Taking a step back, that probably means AI detection work might must go deeper than flagging particular phrasing and delve (see what I did there?) more into the substantive factual content of the work itself.
