Is The Prince AI-generated? 4 of 7 AI detectors said yes
No. The passage is W. K. Marriott's 1908 English translation of Machiavelli's The Prince, written in Italian in the early sixteenth century. Even so, 4 of the 7 AI detectors we tested flagged it. Fakespot gave it a 99.8% AI probability, its highest score for any of the 30 famous texts we ran, while Desklib gave it 0.9%.
Tested
What each AI detector said
| Detector | Verdict | AI probability |
|---|---|---|
| Fast-DetectGPT (Qwen2.5-1.5B) | AI-generated | raw score only |
| Desklib AI text detector | Human | 0.9% |
| Fakespot AI text detector | AI-generated | 99.8% |
| MAGE | AI-generated | raw score only |
| E5-small LoRA detector | Human | 31.5% |
| OpenAI RoBERTa detector | Human | 0.0% |
| TMR (Target Mining RoBERTa) | AI-generated | 98.2% |
The passage we tested
All states, all powers, that have held and hold rule over men have been and are either republics or principalities. Principalities are either hereditary, in which the family has been long established; or they are new. The new are either entirely new, as was Milan to Francesco Sforza, or they are, as it were, members annexed to the hereditary state...
A majority verdict built on shaky detectors
Four detectors flagged the passage, which makes The Prince one of the 8 famous texts out of 30 that a majority of detectors called AI. Look at who voted, though. Fakespot was the most certain at 99.8%, and it flags 34.9% of human writing. TMR (98.2%) flagged 83.3% of our human test texts. MAGE flagged it, with a false positive rate of 19.0%, and Fast-DetectGPT flagged it with a threshold we set on our calibration split and a false positive rate of 15.9%.
Desklib, the most careful of the detectors that catch a real share of AI text, gave it 0.9%. The E5-small LoRA detector, which flags 50.0% of human writing, passed it at 31.5%.
Branching categories and stock connectives
The first chapter is four sentences long, and it is a classification. All states are either republics or principalities; principalities are either hereditary or new; new ones are either entirely new or annexed. Every sentence splits a category in two, much as a chatbot lays out an explanation under headings and sub-bullets, and the passage even runs into the heading of the next chapter. A classifier trained on AI answers has plausibly learned to treat that kind of branching order as a sign of a machine.
Marriott’s English adds a layer. He keeps Machiavelli’s argument moving with stock connectives such as “inasmuch as”, “I say at once” and “hence it happens that”, and most claims arrive with a reason attached. Chatbot prose tends to lean on the same kind of signposting. The book is also standard reading in politics courses and widely quoted, so the language model behind Fast-DetectGPT has likely met these lines before.
The lesson for analytical writing
Essays and reports that open by dividing a topic into types are ordinary human writing, and this result suggests Fakespot in particular may punish them. The disagreement between detectors is the other warning. On The Art of War, a translation of about the same age, Fakespot gave 2.5% and Desklib flagged it, the reverse of what happened here. When two detectors swap verdicts on texts this similar, neither verdict tells you much about authorship. See the Fakespot page for its full error rates and the famous texts overview for the other results.
Questions people ask
Why does Fakespot think The Prince was written by AI?
We can only offer a likely explanation. The first chapter sorts states into types and subtypes in a tidy branching outline, and Marriott's translation links each step with formal connectives. That orderly, explanatory structure is common in chatbot answers.
Which detectors said The Prince is human?
Desklib (0.9%), the E5-small LoRA detector (31.5%) and the OpenAI RoBERTa detector (0.0%). Desklib had the fewest false positives of the detectors that catch a real share of AI text, flagging 6.3% of human texts in our main test.
Is a 99.8% AI score proof that a text is AI-generated?
No. Fakespot gave this passage, written centuries before AI, a 99.8% AI probability. In our main test it flagged 34.9% of human texts, so a high score from it is weak evidence.