Working with LLMs (Large Language Models) can feel a bit like working with a new colleague: highly motivated, but (understandably) not yet familiar with all the details. If not all relevant information is included in the prompt, the LLM will try to fill in the gaps – relying on what it considers the most likely scenario. Well-intentioned, but not always helpful.
New team members often act in the same way: “If I ask, I’ll come across as annoying,” or “They’ll think I don’t know what I’m doing.” But just like in human collaboration, this kind of hesitation can lead to poor results. A healthy team culture assumes that not everything can be known from the start – and that asking questions is part of the process.
It’s the same with LLMs: encourage questions, reduce ambiguity, and provide context. Because without orientation, even the most powerful model can only guess – and might miss the mark.
How to encourage questions — human and machine alike
With people:
• Set up regular feedback and check-in sessions
• Foster open communication
• Actively encourage questions – instead of asking “Any questions?”, try:
“Which questions are still open?”
With LLMs:
• Group prompts around the same context — this helps the model understand connections
• Be clear and concise — skip filler words and vague phrasing
• Actively invite clarification, e.g.:
“What information is missing to give a solid answer?”
Why this matters
LLMs generate answers based on probabilities. If there are three interpretations — one with 90%, one with 7%, and one with 3% – the model will go with the first (probably right one), without asking. But if the probabilities are 60%, 25%, and 15%, the situation becomes riskier. Without an explicit prompt to clarify, the LLM still chooses – even though it might be unsure.
The better approach: give the model room to ask, before it answers. This can significantly improve the outcome.
Pro tip:
Even in team settings, question prompts can be made more effective by rephrasing. Try asking:
“Which three questions are still open?”
It creates a brief moment of pause – and leads to much more thoughtful engagement with the topic. 😉
