There are many services that claim to be able to recognize AI text with “99% accuracy” – without providing any proof. At the same time, there are services that claim to adapt AI texts in such a way that no detector can recognize them.
Both cannot be true at the same time.
Granted: I can recognize an unmodified text from GPT 3.5 – without any tool. Certain phrases are common. The structure seems always the same. I’ve seen articles here and there on the net where I was “99%” sure: This is copy-pasted 1:1 from this (now obsolete) AI.
But we don’t talk much about GTP 3.5 anymore. A tool like Claude 3 Opus has a very good, surprisingly human writing style.
This improvement in quality leads to an effect Gizmodo describes in an article worth reading: AI detectors flag text as AI-generated if, for example, it is too well-written.
Yes, really: Correct grammar and punctuation are now suspect. A professional writing style and content structure according to well-known style guides? That can only be AI.
Of course, this is complete nonsense.
The scientific study “Testing of detection tools for AI-generated text” came to a similar conclusion after examining 12 services:
“The researchers conclude that the available detection tools are neither accurate nor reliable.”
It doesn’t get any clearer than that.
Another problem: Even using helpful AI-powered tools could lead to mislabeling. Think of services that find mistakes in a text, improve wording, and so on. Are these improved and corrected texts suddenly “AI-generated”? Certainly not.
In other words, there is a high probability that AI recognition will be completely wrong. AI texts will slip through undetected while human-written texts will be falsely labeled as AI work.
So don’t rely on these vendors’ promises.
Instead, consider why the use of AI is seen as a problem in the first place. Then you can address that issue and discuss it with the content team.
Because one thing is clear: even if an AI delivers excellent copy, a human should always have the last word. It should also be clear when and for what purpose AI tools can be used, where their weaknesses lie, and what they are generally not well suited for.