In plain English
Like showing your work in math class, the AI works through a problem one step at a time before answering. That usually leads to better answers on hard questions, and it lets people see how it got there.
Example
Asked how many weekdays are left in the year, a model first notes today's date, counts the weeks left and subtracts holidays, then answers. Without those steps, it's more likely to guess wrong.
In short
Showing its work. The step-by-step thinking a model does before it answers.
Why it matters
Chain of thought is a big reason reasoning models are better at math, coding and planning. Reading the steps also helps people catch mistakes. Researchers caution that the written steps don't always match how the model actually reached its answer.
Common confusion
Chain of thought vs. reasoning model. Chain of thought is the technique: thinking in steps. A reasoning model is trained to do it automatically, often at length and sometimes out of view.