What it is
When generating text, the model chooses among several possible continuations. Temperature adjusts how much it gambles: near 0 it picks the most likely one (stable answers), higher allows more diverse outputs.
Why it matters to an agency
For data and exact tasks (extracting, classifying), use a low temperature. For brainstorming or creative copy variants, raise it.
Example
Low temperature for a report with figures; high temperature to generate 10 headline ideas.