When an AI generates a long answer there are two ways to show it: wait until it is complete and display it all at once, or show it progressively, word by word, as it is generated. The second is called streaming, and it has become the standard in modern conversational interfaces.
Why streaming improves perception
Even when the total time to complete the answer is identical, seeing text appear immediately gives the sense that the system is actively working, reducing the feeling of waiting. An empty five-second wait feels longer than five seconds in which something is visibly happening.
A practical advantage too, not only psychological
With streaming, the user can start reading and judging the beginning of an answer before it is finished, and in some cases stop it on realising it is heading the wrong way — a concrete saving of time, not merely a sense of speed.
Where caution is needed
For content where order and final coherence matter — a calculation, a precise data structure — showing partial information during generation can mislead if the user acts on something incomplete. In those cases it is worth judging whether streaming is appropriate or whether it is better to wait for the finished result.