History, and the quotation that was never said

The subject where fabrication is invisible

In maths a wrong answer is wrong. In history a fabricated detail looks exactly like a real one, because the whole subject is made of specifics: dates, names, quotations, archive references, the position a particular historian took in a particular book.

Those specifics are the weakly-represented category from Module 1. Which means history is the subject where the mechanism bites hardest and where the failure is hardest to see without checking.

The four fabrications to expect

Invented quotations. The most common and the most damaging. A plausible line, attributed to a real figure, in their register. Famous historical figures attract these enormously: many quotations widely attributed to Churchill, Gandhi, Lincoln or Einstein were never said by them, and that misattribution is in the training data too, so the model reproduces it confidently.

Composite events. Two similar events blended into one, with a date from one and details from the other. Reads perfectly. Wrong.

Invented historiography. Historian X argued that...: where X is real, the position is plausible for their school, and they never wrote it. Markers in history departments spot this immediately, because they know the literature.

Date drift. Confident dates that are one or two years out. Especially bad for anything before 1800, and for events outside western Europe and North America, where the training text is thinner.

The check, which is the discipline of the subject anyway

History already has the answer: go to the source. What is new is that you must apply it to a category of claim that did not previously exist: the claim that arrives already footnoted-looking.

• Every quotation gets traced to a primary source or a scholarly edition. If you cannot find where it was said, written or recorded, you do not use it. A quotation you found only in secondary repetition is not evidence.
• Every historian’s position gets traced to the book or article, with a page number, that you have opened.
• Every date gets checked against a reference work.

The free path is good here. Google Books for full-text search of published works, which is excellent for finding whether a quotation appears anywhere before a given year. Internet Archive and HathiTrust for out-of-copyright texts. JSTOR has free limited reading; your library has the rest. National archives are increasingly digitised and free: the UK National Archives, the US National Archives, and many national and state collections.

The bibliographic test for a quotation

A specific and effective move. Search the exact phrase in Google Books, restricted by date. If a quotation attributed to a nineteenth-century figure first appears in print in 1987, it is a twentieth-century invention. This is how the quotation-checking community works and you can do it in two minutes, free.

Where the model earns its place in history work

Substantially, and it is worth being clear about it:

• Explaining a historiographical debate’s shape: what the positions are, roughly why they differ. Then go and read the actual historians.
• Context for a primary source you are reading. Paste the source; ask what a reader at the time would have understood by a term. Verify anything specific.
• Translating and paraphrasing archaic or foreign-language text, which is strong, with the caution that a translation you cannot check is a claim you cannot verify.
• Arguing against your thesis, which is the essay move from Module 2 and is worth more in history than almost anywhere else.
• Making your prose clearer, at sentence level, on writing you produced.

Source criticism has not changed

The questions you were taught still apply and now apply to a new kind of text. Who produced this? For what audience? What did they gain by saying it? What would they not have known? What is missing?

Run those on a model’s output and the answers are unusual: produced by a statistical process, for you, gaining nothing, knowing whatever was in its training text, missing everything specific that was rare. That is not a source. It is a starting point with no provenance, and treating it as a source is the single mistake to avoid in this subject.

The one that gets people caught

A fabricated citation in a history essay is the most common academic-integrity trigger in humanities departments, and it usually was not intended as dishonesty: the student believed the reference was real. The intention does not change the outcome. Open every source. Ten minutes with a library catalogue is the entire defence, and it is also just doing the subject properly.