A boundary is a pair of unlikely markers wrapped around user text, with an instruction that says only text inside them is data. It is not perfect, and it does not need to be: it turns an invisible edge into a visible one.
About 1 min of reading
Wrap user text in a marker pair
Pick a marker that is unlikely in text
Before you run it, say out loud what it will print. Then open the answer.
Name the region in the instruction
Before you run it, say out loud what it will print. Then open the answer.
Check the user text cannot close the fence
Before you run it, say out loud what it will print. Then open the answer.
Quoting tells a human. The model reads the same tokens whether or not a quote was used, so quoting changes nothing.
The boundary is built by the code that assembles the request, never by a person writing the prompt.
A marker the user cannot type is a marker they cannot fake. Random per request markers are the strong version.
Naming the region gives the model something to point at. Without a name, the marker is decoration.
A user who guesses the marker still wins. The boundary is a barrier, not a wall.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.