Is The Wealth of Nations AI-generated? 3 of 7 AI detectors said yes
No. The Wealth of Nations is Adam Smith's 1776 book, and when we tested its opening on the division of labour, 3 of 7 free AI detectors flagged it as AI-generated, namely Fakespot, TMR and Fast-DetectGPT. Fakespot's 56.1% AI probability was only a little over its threshold. Desklib, which rarely flags human writing, gave it 0.2%.
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 | 0.2% |
| Fakespot AI text detector | AI-generated | 56.1% |
| MAGE | Human | raw score only |
| E5-small LoRA detector | Human | 41.6% |
| OpenAI RoBERTa detector | Human | 5.0% |
| TMR (Target Mining RoBERTa) | AI-generated | 97.6% |
The passage we tested
The greatest improvements in the productive powers of labour, and the greater part of the skill, dexterity, and judgment, with which it is anywhere directed, or applied, seem to have been the effects of the division of labour. The effects of the division of labour, in the general business of society, will be more easily understood, by considering in what...
Who flagged Adam Smith
Fakespot, TMR and Fast-DetectGPT called the opening of The Wealth of Nations AI-generated. Desklib, E5-small LoRA, MAGE and the OpenAI RoBERTa detector passed it.
None of the three flags is strong evidence. TMR flagged 83.3% of human texts in our main test, mostly because of paragraph breaks. Fakespot’s score of 56.1% sits just past its 50% line, and Fakespot flagged 34.9% of human texts overall. Fast-DetectGPT, with a threshold we tuned on a separate calibration set, flagged 15.9% of human texts and more than half of the famous ones, 17 of 30. Desklib, the detector with the best AUROC in our test, gave the passage a 0.2% AI probability.
An eighteenth-century explainer
Read the passage with modern eyes and its shape is familiar. Smith opens with a general claim about what drives productivity, says it is easier to understand through particular cases, explains why the division of labour is more visible in small trades than in large ones, and then announces an example: the pin-maker. That claim, explanation and example structure is close to what a chatbot produces when asked to explain an economic idea, and it is a likely reason some detectors reacted to it.
The vocabulary helps it along. The phrase “division of labour” comes back again and again, and words like “manufactures”, “workmen” and “trifling” repeat within a few sentences. Repeating one term makes text easier for a language model to predict, which is what Fast-DetectGPT measures. The book is also among the most cited works in economics, so the model has probably met this opening before.
Explanatory essays near the threshold
Students and analysts write this kind of text all the time. If the claim and example structure is what pushed Fakespot over its line here, an ordinary explanatory essay could be flagged the same way. When a detector flags one at a score near its threshold, treat it as a coin toss, and read the Fakespot results before acting on it.
Two other texts from 1776 got very different results. The Declaration of Independence drew four flags and Common Sense only one. All 30 are on the famous texts page.
Questions people ask
Why does an AI detector think Adam Smith is AI?
Probably because the passage explains a single idea the way a textbook or a chatbot would. It states a claim, repeats its main term many times and then walks through a worked example. The book is also widely copied online, so language models likely know it.
Which detectors passed The Wealth of Nations?
Desklib (0.2% AI probability), E5-small LoRA (41.6%), the OpenAI RoBERTa detector (5.0%) and MAGE.
How sure was Fakespot that the passage is AI?
Not very. It gave 56.1% against a 50% threshold. Fakespot flagged 34.9% of human texts in our main test and 13 of our 30 famous texts.