The Encyclopedia Paradox
Why laws and standards sound like AI, and how our algorithm distinguishes a real document from a generated fake.
If you paste a Wikipedia article, a federal law text, or a server manual into a classic AI detector, the algorithm will likely output a 99% AI probability. And it will be mathematically right, but factually wrong.
What is the problem?
There is a class of texts that by their nature must be dry, structural, and unemotional. These include:
- Encyclopedic and dictionary articles;
- Legal documents (laws, contracts, regulations);
- Technical specifications and press releases.
Neural networks were trained precisely on these texts. The knowledge base of all humanity is encyclopedias and official documents. The models absorbed this dry academic style as ideal. Therefore, when a live lawyer writes a contract, they use the exact same patterns as the algorithm.
How we solved this paradox (Fact-checking)
In older versions of the system, we tried to simply "disable percentages," but this created a vulnerability: lazy managers generated AI articles, and the detector mistook them for "documents" due to their dryness.
Now we have moved to a new Authorship Axis with a strict internet search stage (Fact-checking):
- Dry text + Source found = Editorial text (Document). We confirm that the bureaucratic language here is a requirement of the format and genre.
- Dry text + No source = Generated by AI. If the text has an encyclopedic style but is not found online, the decision is made based on the combination of structural features and the density of machine patterns.
Specifics of individual formats
The paradox of formal genres manifests itself differently depending on the scope of application. We have prepared separate analyses for the most complex categories:
Encyclopedias and Wikipedia
How the collective editing of thousands of people and scripts erases the human voice, turning the text into a sterile algorithm.
Legal documents
Bureaucratese vs. "digital clichés". How we distinguish a real contract or law from a plastic AI imitation.
Summary: We no longer try to write off formal genres as a margin of error. Thanks to dividing text into 4 types of authorship and fact-checking, official documents receive their deserved and confirmed blue status "Editorial text".