Is Pride and Prejudice AI-generated? 4 of 7 AI detectors said yes
No, Jane Austen wrote it in 1813, yet 4 of 7 free AI detectors flagged the novel's opening as AI-generated, a majority. Fakespot gave it a 79.0% AI probability and E5-small LoRA 77.9%. Desklib, which flagged only 6.3% of human texts in our main test, passed it at 45.8%, just under its 50% cut-off.
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 | 45.8% |
| Fakespot AI text detector | AI-generated | 79.0% |
| MAGE | Human | raw score only |
| E5-small LoRA detector | AI-generated | 77.9% |
| OpenAI RoBERTa detector | Human | 0.0% |
| TMR (Target Mining RoBERTa) | AI-generated | 98.4% |
The passage we tested
It is a truth universally acknowledged, that a single man in possession of a good fortune must be in want of a wife. However little known the feelings or views of such a man may be on his first entering a neighbourhood, this truth is so well fixed in the minds of the surrounding families, that he is considered as...
A majority of detectors got it wrong
Fakespot, TMR, E5-small LoRA and Fast-DetectGPT all flagged the opening of Pride and Prejudice. That makes it one of the 8 famous texts out of 30 that a majority of our detectors called AI-generated. TMR was the most confident at 98.4%, Fakespot gave it 79.0% and E5-small LoRA 77.9%.
Those detectors flag a lot of human writing in general. In our main test, TMR flagged 83.3% of human texts, E5-small LoRA 50.0% and Fakespot 34.9%. Fast-DetectGPT flagged 15.9%, using a threshold we tuned on a separate calibration set. Apart from TMR, which flagged every human fiction passage, they are much more restrained with novels and stories. Fakespot and E5-small LoRA each flagged 12.5% of the human fiction passages, and Fast-DetectGPT flagged none of the 24, which makes this result stand out.
Desklib, MAGE and the OpenAI RoBERTa detector passed it. Desklib was the closest call at 45.8%, a little under its 50% threshold.
A famous aphorism and short lines of dialogue
The first sentence is one of the most quoted opening lines in English fiction. A language model trained on web text has very likely read it many times, so it can predict the words that follow. Fast-DetectGPT flags text a model finds predictable, which is the likeliest reason it flagged this passage when it passed every human fiction passage in our main test.
The sentence itself is an aphorism: a confident, general statement in balanced clauses. That polished, summarizing tone is something AI writing is often accused of. After it comes rapid dialogue between Mr. and Mrs. Bennet, which in the copy we tested runs to many short paragraphs. TMR reacts to paragraph breaks more than to the words, as our guide to paragraph breaks shows.
Majority votes and polished prose
If you write in a polished style, or quote well-known lines, a detector may score you higher for reasons unrelated to AI. Running several detectors and going with the majority would have convicted Austen here. A careful detector with a known false positive rate, read as one piece of evidence, does better. Compare the very different result for Huckleberry Finn, or see the full famous texts results.
Questions people ask
Why does an AI detector flag Jane Austen?
Probably because the opening is so famous. "It is a truth universally acknowledged" is quoted constantly, so language models have likely seen the passage many times and find it predictable. The passage is also mostly short lines of dialogue, and one detector, TMR, flags almost any text split into many paragraphs.
Which detectors said Pride and Prejudice is human?
Desklib, MAGE and the OpenAI RoBERTa detector. Desklib came closest to flagging it, with a 45.8% AI probability against a 50% threshold.
Would a majority vote of detectors have caught the mistake?
No. Four of the seven detectors flagged the passage, so a majority vote would have called Austen's opening AI-generated.