Open an AI-drafted essay and count the proper nouns in the first paragraph. The count is usually zero. "Recent studies have shown", "the literature suggests", "scholars have debated" — every reference is to a generic plural. Real academic prose names things: the author, the year, the institution, the dataset.
Why models avoid names
A model that doesn't have the source in front of it can't be sure of the name, the year or the exact claim. So it generalises — "recent research" is safer than "Acemoglu and Robinson (2012)" because the latter can be wrong. The result is prose that reads as authoritative until you ask what it's actually claiming.
The proper-noun test
Mark every proper noun in your draft — author names, place names, years, terms of art. Count them. If a 600-word essay has fewer than five, the essay is hiding behind genericism. Add the specifics from your actual reading, or do the reading.
Before
Recent studies have shown that institutional quality is a key driver of long-run economic growth. Scholars have debated the relative importance of various factors, with the literature broadly suggesting that institutions matter more than geography.
After
Acemoglu, Johnson and Robinson's 2001 paper on the colonial origins of comparative development used settler mortality as an instrument to argue that institutions, not geography, explain long-run growth differences. The argument has been challenged — Glaeser et al. (2004) suggested the instrument also picks up human capital — but the institutional case has held up better than the geographic alternatives across the subsequent literature.