A word is essentially a computer instruction. The word size is the number of bits the CPU can process in one bus cycle.
We need a fixed unit to transfer data around a computer; each unit is called a word. Therefore, word size refers to the number of bits, a “package of bits” that a CPU can transfer in parallel in one bus cycle. The most common word size now is 64 bits. Older computers and computers embedded in appliances may have word sizes of 8, 16, or 32 bits.
Conventionally, word size dictates how many bits a CPU is as well; on a 32-bit CPU, the word size is typically 32-bit. This is also associated with a CPU’s internal architecture, including the size of its general-purpose registers and ALU. While CPUs may have certain specialised registers of a different size than the word length, general-purpose registers will almost always be the same size.
In general, it is faster to have a register size match or exceed the word size, though this depends on the size of the computed data as well. Assuming the same clock rate, adding a 16-bit number will not be faster with a 32-bit register than a 16-bit register. But a 24-bit number will be, as the 32-bit register can fit the entire number whereas the 16-bit register can’t.
However, word size doesn’t necessarily dictate the width of external buses. For example, a 32-bit CPU may have a 64-bit order wider external data bus to increase memory bandwidth, allowing the CPU to transfer more bits in less time.