My approach is to set up the user to tell me when they are done, with the caveat that I'll gently stop them in the interest of keeping us moving. I've rarely had to intervene as long as I let the user know they can't fail, it's them testing our solution vs us testing them.
In setting up the test I'll tell the user to indicate when they are done with a task, or when they believe they cannot complete it. I'll usually say "thank you" or "ok" and move on to reading the next task description. Primary intent is to acknowledge the transition, so the response itself is more important than what I say. I also work to keep consistency and neutrality so I'm minimizing any chance I influence the test.
The first few clients where I did this I felt really weird, but once I got my own rhythm going it was a bit more comfortable. Hope this helps!