# Do AI detectors think the Gettysburg Address is AI? 1 of 7 said yes

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

> Abraham Lincoln gave the Gettysburg Address in November 1863, so no. We ran the full 285-word speech through seven free AI detectors, and only 1 of 7 flagged it as AI-generated, TMR at a 97.7% AI probability. Desklib gave it 32.7% and Fakespot 3.3%. MAGE, Fast-DetectGPT, E5-small LoRA and the OpenAI RoBERTa detector also passed it.

The Gettysburg Address, Abraham Lincoln, 1863. 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: "Four score and seven years ago, our fathers brought forth upon this continent a new nation: conceived in liberty, and dedicated to the proposition that all men are created equal. Now we are engaged in a great civil war. . .testing whether that nation, or any nation so conceived and so dedicated. . . can long endure. We are met..."

## The speech uses the patterns people call AI tells

Read the Gettysburg Address with a list of AI writing signs in hand and it fails several checks. It is built on threes: "we cannot dedicate, we cannot consecrate, we cannot hallow this ground", and "government of the people, by the people, for the people". It repeats its central word, a form of "dedicate" coming back again and again in a speech of 285 words. It speaks in large abstractions, a new nation "conceived in liberty", a "new birth of freedom", and closes on a ringing summary line, all habits editors now strip out of chatbot prose.

Most of the detectors did not react to any of that. Only TMR flagged the speech, and TMR flagged 83.3% of the human texts in our main test, mostly because of paragraph breaks. The copy we tested is split into short paragraphs, which is enough for TMR.

## How the other detectors read it

Of the classifiers that report a probability, Desklib was the most suspicious after TMR, at 32.7%, but that is well below its 50% threshold, and Desklib flagged only 6.3% of human texts overall. E5-small LoRA gave it 22.1%, Fakespot 3.3% and the OpenAI RoBERTa detector 0.0%. MAGE passed it, and so did Fast-DetectGPT, which measures how predictable a text is to a language model.

That last result is a little surprising. The speech is one of the most memorized texts in American history, and Fast-DetectGPT flagged other very familiar passages, including [Genesis 1](/famous-texts/genesis-1/). One possible reason is the source we used. The Project Gutenberg version breaks the speech up with runs of spaced dots, an unusual punctuation pattern that a language model would probably not predict, and that may have made the text look less familiar to it.

## Style is weak evidence

People who judge writing by feel often point to the devices Lincoln used, and on this speech those signals would have pointed the wrong way. Rhetorical devices like these are old, and a writer who uses them has shown nothing about AI. The [false positives page](/false-positives/) shows how often detectors misjudge ordinary human writing, and the [famous texts page](/famous-texts/) lists every passage we tested.

## Questions people ask

### Is the Gettysburg Address written like AI?

Some readers might think so, because it relies on triads and repetition, such as "we cannot dedicate, we cannot consecrate, we cannot hallow", which are now called signs of AI writing. Most detectors disagreed. Only TMR flagged it.

### Which detector came closest to flagging the Gettysburg Address?

Apart from TMR, which flagged it, Desklib gave the highest AI probability of the classifiers that report one, 32.7%, well under its 50% threshold. E5-small LoRA gave 22.1% and Fakespot 3.3%.

### Why did TMR flag the Gettysburg Address?

TMR flags almost every text with more than one paragraph. It flagged 83.3% of the human texts in our main test, so its verdict here says little about Lincoln.