# Is A Scandal in Bohemia AI-generated? Only 1 of 7 AI detectors said yes

Source: https://toolbenchmarks.com/famous-texts/sherlock-holmes-scandal-in-bohemia/  
Tested: 30 October 2026. Publisher: Tool Benchmarks. Data licence: CC BY 4.0.

> No. Arthur Conan Doyle published A Scandal in Bohemia, the first Sherlock Holmes short story, in 1891. Only 1 of the 7 AI detectors we tested flagged its opening, TMR, with a 96.6% AI probability, and TMR called 83.3% of human texts AI in our main test. Desklib gave it 0.2%, and MAGE and Fast-DetectGPT passed it.

A Scandal in Bohemia (opening), Arthur Conan Doyle, 1891. 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: "To Sherlock Holmes she is always _the_ woman. I have seldom heard him mention her under any other name. In his eyes she eclipses and predominates the whole of her sex. It was not that he felt any emotion akin to love for Irene Adler. All emotions, and that one particularly, were abhorrent to his cold, precise but admirably balanced..."

## A single flag from the least reliable detector

TMR gave Watson's opening paragraphs a 96.6% AI probability. Every other detector disagreed: Desklib (0.2%), the OpenAI RoBERTa detector (0.2%), Fakespot (26.6%), the E5-small LoRA detector (35.8%), MAGE and Fast-DetectGPT.

TMR's verdict tells you almost nothing about this text. It flagged 28 of the 30 famous texts we ran, and in our main test it flagged all 101 human texts that had more than one paragraph. When we removed the line breaks from those same human texts, it flagged 25.7% of them. The excerpt here is two paragraphs, so TMR's flag most likely reflects the formatting. Our guide on [paragraph breaks](/guides/paragraph-breaks-change-ai-detector-results/) has the full test.

## Why the other detectors saw a human writer

Most of the usual reasons a classic gets flagged are absent here. The passage is famous, but Fast-DetectGPT, which reacts to wording a language model finds familiar, did not flag it. Nor is it formal or list-like. Watson writes with opinions and odd, specific choices: Holmes "eclipses and predominates the whole of her sex", a strong emotion in him would be like "grit in a sensitive instrument", and Holmes alternates "from week to week between cocaine and ambition". The italic stress on "_the_ woman" and the jab at "the late Irene Adler, of dubious and questionable memory" are a narrator's personal judgments.

## What to do with a lone flag

If you run a piece of writing through several detectors and only one says AI, check which one it was before worrying. A lone flag from TMR on multi-paragraph text is the expected result for human writing. A lone flag from Desklib, which flagged 6.3% of human texts in our main test, would deserve a closer look, though even then it is no proof. [Frankenstein](/famous-texts/frankenstein/) was also flagged by TMR alone. The [TMR page](/ai-detectors/tmr/) explains its paragraph problem, and the [famous texts overview](/famous-texts/) lists every result.

## Questions people ask

### Why did TMR flag the Sherlock Holmes story?

Most likely because the passage has two paragraphs. TMR flagged every multi-paragraph human text in our main test, and with the line breaks removed it flagged only 25.7% of them.

### Do AI detectors think Sherlock Holmes stories are AI?

Not this one. Apart from TMR, every detector passed the opening of A Scandal in Bohemia, with AI probabilities of 0.2% from Desklib, 26.6% from Fakespot and 35.8% from E5-small LoRA.