Mixture of Experts (MoE)
Mixture of Experts is a way of building AI models where each word is handled by only a few specialised parts of the model.
Mixture of Experts, or MoE, is a way of building an AI model out of many smaller parts, called experts, with a router that sends each piece of text to only a few of them. The model as a whole can be huge, but each word only wakes up a small share of it, so it runs faster and costs less.
Many of today's large models are thought to work this way, and several open ones say so. You don't need the details to do AI search well. It helps to know it's one more reason the same question can get different answers.
How it works
Picture a large hospital. Every patient doesn't see every doctor. A triage nurse sends each one to the two or three specialists who fit. Each specialist only handles what they're good at, and the hospital as a whole covers everything.
In an MoE model, the router plays the nurse. For each word, it picks a handful of experts out of dozens or hundreds, and their outputs are blended into the next step. The idea was set out for modern language models in a 2017 paper.
Why it matters to a marketer
MoE is mostly about cost: it lets a company run a bigger model for the same money. For you, that means larger, more capable models can sit behind more of the answers people see.
It also adds to the variety in answers. Small differences in a question can send words to different experts, and the platforms also change their models often.
For example, "best coffee subscription UK" and "best UK coffee subscription" mean the same thing to a person. To a model, they're different text, and can produce answers that name slightly different brands. That's one reason a single answer tells you little, and a week of answers tells you more.
In AnswerOrbit
We ask each prompt once a day on each platform, and every number is based on seven days of answers, so one unusual answer doesn't swing your numbers. See How we count.