# Is Walden AI-generated? Only 1 of 7 AI detectors said yes

Source: https://toolbenchmarks.com/famous-texts/walden/  
Tested: 30 October 2026. Publisher: Tool Benchmarks. Data licence: CC BY 4.0.

> No. Walden is Henry David Thoreau's account of his life at Walden Pond, published in 1854. Of the seven AI detectors we tested, only TMR called its opening AI-generated, with a score of 97.3%. The other six passed it, among them Desklib (0.6%), the OpenAI RoBERTa detector (0.2%), Fakespot (18.8%) and E5-small LoRA (37.9%).

Walden (opening), Henry David Thoreau, 1854. Flagged as AI by 1 of 7 detectors.

| Detector | Verdict |
|---|---|
| Fast-DetectGPT (Qwen2.5-1.5B) | Human |
| Desklib AI text detector | Human |
| Fakespot AI text detector | Human |
| MAGE | Human |
| E5-small LoRA detector | Human |
| OpenAI RoBERTa detector | Human |
| TMR (Target Mining RoBERTa) | AI-generated |

Passage tested: "When I wrote the following pages, or rather the bulk of them, I lived alone, in the woods, a mile from any neighbor, in a house which I had built myself, on the shore of Walden Pond, in Concord, Massachusetts, and earned my living by the labor of my hands only. I lived there two years and two months. At..."

## TMR was the only one to say AI

TMR flagged Walden at 97.3%, and TMR flagged nearly everything: 83.3% of the human texts in our main benchmark and 28 of the 30 famous passages. Its verdict here most likely tracks formatting. The passage has two paragraphs, and TMR flagged every human text with more than one paragraph, as the [TMR results](/ai-detectors/tmr/) show.

The other six all passed it. Desklib, which flagged 6.3% of human writing, gave it 0.6%. Fakespot, which flagged 34.9% of human texts and gave Moby-Dick 90.7%, put Walden at 18.8%. E5-small LoRA gave it 37.9%, and MAGE and Fast-DetectGPT passed it as well.

## What makes this opening read as human

Thoreau starts with particulars. He lived "a mile from any neighbor, in a house which I had built myself, on the shore of Walden Pond, in Concord, Massachusetts", and he stayed "two years and two months". Named places and odd, exact durations are the kind of detail a model asked for a generic reflection seldom supplies, and the detectors may be responding to that.

The voice wanders too. He corrects himself in the first clause ("or rather the bulk of them"), grants that some would call his townsmen's questions "impertinent" before calling them "very natural and pertinent", and stops to explain why this book, unlike most, keeps the first person. Asides and second thoughts like these make the next word harder to predict, which is likely why Fast-DetectGPT, a method that flags predictable text, let it pass.

Thoreau's own [Civil Disobedience](/famous-texts/civil-disobedience/), published in 1849, drew three flags. That essay opens with general statements about government, and MAGE and Fast-DetectGPT, which both passed Walden, flagged it.

## Concrete writing still drew a flag

Concrete, personal writing seems to fare better with these detectors than abstract argument, but even here one detector said AI with near certainty. Running several free tools and treating any single "AI" result as a warning would have condemned Walden. The [famous texts page](/famous-texts/) shows that every one of the 30 passages we tested was flagged by at least one detector.

## Questions people ask

### Is TMR right that Walden is AI?

No. Thoreau published Walden in 1854, and the one detector that flagged it, TMR, flagged 83.3% of the human texts in our main test, so its verdict carries almost no information.

### Why did Civil Disobedience get more AI flags than Walden?

Civil Disobedience drew three flags, from MAGE, TMR and Fast-DetectGPT. Its opening is a general argument about government, while Walden opens with concrete facts about Thoreau's own life. That difference in register is the likeliest explanation.

### Which detector came closest to flagging Walden besides TMR?

E5-small LoRA, at 37.9% against a 50% threshold. It flagged 50.0% of human texts in our main test, so a pass from it is more meaningful than a flag.