A test double is a simple stand-in for a real part of a program during a test. It is like using a cardboard model of a door to check where it fits before installing the real one.
A test might use a pretend payment service that returns a predictable answer instead of charging a real card. This makes the test safer and easier to repeat.
The stand-in should behave closely enough for the situation being tested. It cannot prove that the real service works, so teams still need checks that use the actual connection where appropriate.