The first thing a careful reader notices in good academic prose is the calibration: claims the writer will defend hard, claims they hedge, and claims they explicitly attribute to other people. Model output often flattens all three into a single level of confidence.
Two failure modes
- Under-hedging. Every claim stated as fact. "This proves", "clearly shows", "demonstrates that" attached to evidence that supports nothing of the sort. A marker reads it as overconfident.
- Over-hedging. Every claim wrapped in "may", "could", "appears to", "some scholars argue". A marker reads it as evasive — the writer never commits to anything.
What calibrated hedging looks like
Skilled essayists hedge the claims they are less sure of, and commit hard to the ones they will defend. The shape of an argument shows up in the variety of its modal verbs — "suggests" for the literature, "shows" for the data, "argues" for the writer's own position. A single flat register across all three is the AI tell.
Before
The literature clearly shows that minimum wage increases reduce employment. Card and Krueger's New Jersey study definitively proves this is not always the case. It is obvious that the relationship depends on many factors.
After
Most pre-1994 studies argued that minimum-wage rises reduce employment. Card and Krueger's New Jersey paper unsettled that consensus: comparing fast-food employment on either side of the Pennsylvania border, they found no significant fall after the rise. The relationship probably depends on the local labour-market slack — but the older claim of a clean negative effect no longer survives the evidence we have.