Temperature
Temperature is a setting that controls how much randomness an AI model uses as it writes, one reason the same prompt gets different answers.
Temperature is a setting that controls how much chance plays a part when an AI model writes. A model writes one word at a time, and at each step it has several words it could pick. At a low temperature it nearly always picks the most likely one. At a higher temperature it's more willing to pick another.
For a marketer, temperature is one of the reasons the same question can name different brands each time it's asked.
How it changes an answer
Say the model is writing "The best coffee subscriptions in the UK include..." and the most likely next name is Kestrel Roasters, with Fernhill Coffee a close second. At a low temperature, Kestrel comes first almost every time. At a higher one, Fernhill comes first some of the time, and the rest of the answer follows from there.
So the order of brands, and sometimes which brands appear at all, can shift without anything on the web changing.
Who sets it
Developers who use a model through its API can usually choose the temperature. In the chat apps people use every day, the company chooses it, and you can't see or change it.
Temperature isn't the only cause of variety. Platforms also search the web differently from one day to the next, read different pages and sometimes use different models.
Why one answer isn't enough
Because of all this, a single answer is a sample. For example, ask the same question ten times and Fernhill might be named in four. Any one of those ten answers, read alone, would give you the wrong idea. To know how often you're really named, you need lots of answers.
In AnswerOrbit
We ask in each platform's own app, so the platform decides how it answers, just as it would for anyone. That's why every number is based on seven days of answers, and why we only mark a change when luck alone would rarely make a difference that big.