We read a lot of interview transcripts, which is an unusual vantage point on questioning. A few patterns show up so consistently that they are worth writing down, and none of them are complicated.
Short beats thorough
The most common failure is a question that contains its own preamble. "So looking at your methodology section, where you discuss the sampling approach, I am interested in the trade-off you mentioned between representativeness and feasibility, could you expand on how you weighed those?"
By the end, the student is answering the last clause and has forgotten the first. Spoken questions have to survive one hearing. Twenty five words is a good ceiling. "You chose convenience sampling. What did that cost you?" does the same work and leaves room for the student to think.
Ground it in something they wrote
General questions get general answers, and general answers are exactly what a well-prepared student can produce without understanding much. "What is validity?" is answerable from a textbook. "Your sample was thirty people from one class. What does that do to your claim?" is not answerable without having thought about their own study.
The question you want is one that a stranger to this specific piece of work could not answer, no matter how well they know the subject.
One question, one idea
Stacked questions are the second most common problem. "Why did you choose that framework, and how does it compare to the alternatives you rejected?" The student picks the half they are more confident about and the other half quietly disappears. Ask one, then ask the other if the first answer earns it.
Follow down, not sideways
The weakest follow-up is "can you give me another example", which tests breadth of recall. The strongest is some version of "why": why that choice, why not the obvious alternative, what would have changed your mind. Depth finds the edge of understanding. Breadth just moves along the surface.
Let the plain words do the work
Register matters more than people expect. "What are the epistemological limitations of your approach?" and "what can this study not tell you?" are the same question. The first tests whether the student knows the word. The second tests whether they know the answer.
That distinction is not a small stylistic preference. For a student working in a second language, the first question is partly a vocabulary exam wearing the clothes of an assessment about method, and it will produce a worse estimate of what they understand.