A large and growing share of American viewers watch English-language shows with English captions running. The habit is a response to specific technical changes rather than to a decline in anyone's attention.
Production sound moved toward realism
Dialogue recording practice shifted toward capturing performances as delivered, including quiet, overlapping and naturalistic speech that earlier production styles would have reshot or looped.
The result reads as authentic in a mixing room and as inaudible in a living room, because the room the mix was judged in does not resemble the room it is heard in.
Directors have publicly defended the approach, which suggests it is a deliberate aesthetic rather than a failure anyone intends to correct.
Television speakers got worse as screens got better
Flat panels leave almost no cabinet volume for speakers, and the drivers are frequently mounted facing downward or backward into the wall behind them.
Dialogue sits in a frequency range that suffers most from small drivers, so speech is the first thing lost while music and effects survive.
External sound systems fix this, and most households do not have one, so the captions substitute for hardware nobody bought.
Mixes are made for rooms most people do not have
Theatrical and premium mixes assume multiple channels with a dedicated center speaker carrying dialogue, and that channel is where speech lives.
Collapsing that mix down to two small speakers folds the center into the same drivers reproducing everything else, and the dynamic range works against the viewer.
Some services now offer a dialogue-focused audio option, which is an acknowledgment that the default mix is not serving the typical viewer.
Viewing moved into shared and noisy spaces
People watch on phones and laptops in transit, in shared apartments and beside sleeping households, often with the volume kept deliberately low or off entirely.
Captions make that viewing possible, and once a person is used to reading along the habit persists into circumstances where the sound would have been fine.
Platforms built for silent autoplay trained this behavior at scale, and it carried over into longer-form viewing without anybody deciding it should.
Captions are not uniformly good
Quality varies considerably between services and between titles, and automated captioning handles names, accents and overlapping speech poorly.
Descriptions of non-speech sound are inconsistent too, which matters most for the viewers the captions were originally created to serve.
The habit spreading to general audiences has raised the visibility of these gaps, which is probably the most useful side effect of the whole shift.