Poorly. Short passages give the model little to measure, which is why single paragraphs produce unstable scores. Detectors rely on patterns across sentences: rhythm, transition density, vocabulary predictability. A paragraph or two gives them too little to work with, so the same text can score very differently depending on what surrounds it. Treat flags on short passages as weak evidence. For a draft, run the whole section and look at which sentences are consistently flagged across tools rather than trusting a single number on a single paragraph.
More detail: Are AI detectors accurate? and AI detectors like Turnitin.