The core case
The clearest illustration in the thread: photograph someone standing in fixed light, first full length and then head and shoulders. The light has not changed, but the proportion of bright and dark area in the frame has, so the meter reads differently and the two frames do not match. In manual they do. Everything else in the thread is a variation on this.
Where it matters most
- Panorama frames, which have to stitch without visible steps in brightness.
- Time-lapse, where per-frame metering produces flicker across the sequence.
- Any set of images meant to be viewed as a group, where inconsistency reads as carelessness.
Where the meter is simply wrong
- High-contrast subjects in fixed light. One contributor meters carefully once — spot metering a white surface and opening up from that reading — then sets it manually and stops thinking about it.
- Food, where dark surfaces and bright highlights mislead automatic metering.
- Fireworks, where the meter has no stable subject to read and reacts to whatever happened last.
- The common factor: the light is not changing, so there is nothing for automation to usefully track.
What aperture priority is still better at
- Walking around, where the light changes constantly and the subject does not wait.
- Combined with exposure compensation it gives most of manual's control while still tracking changing conditions.
- Several contributors said they use aperture priority as their default and switch to manual for the specific situations above, rather than treating either as the correct choice.