Do clarifying questions work better just because of how they're phrased, or is something deeper going on?
Can question framing alone explain improvements in clarification effectiveness?
This explores whether clarifying questions get better results mainly because of how they're worded, or whether timing, what they ask for, and how models are trained matter more.
This explores whether clarifying questions get better results mainly because of how they're worded, or whether something deeper is at work: when they're asked, what they ask for, and why models ask at all. The corpus says framing matters, but on its own it doesn't explain much. The phrasing that works best is usually a sign that the question is aimed at a real gap. In user studies, questions that target a specific missing detail ("What type of monitor?") beat open requests to restate the need ("What are you trying to do?") Which clarifying questions actually improve user satisfaction?. The difference isn't polish. Users engage when they can see how their answer will improve the result, and that depends on the question having found the right missing piece.
Timing matters as much as wording. A well-phrased question asked at the wrong moment is just friction. One useful approach estimates how many different things the user could plausibly mean. When a request's likely interpretations split into several distinct groups, a question is worth asking. When they mostly agree, asking adds nothing Can we predict when users need clarification questions?. Question quality also isn't one dial. Training models on separate qualities like clarity, relevance and specificity works better than training on a single overall score, especially in clinical settings Can models learn to ask genuinely useful clarifying questions?.
The biggest factor sits upstream of wording: whether the model is inclined to ask at all. Standard training rewards whichever reply looks most helpful right now, so models learn to guess instead of ask. Rewarding the value of the whole conversation brings back real intent discovery Why do language models respond passively instead of asking clarifying questions?. Likewise, reinforcement learning raised models' ability to spot missing information in deliberately flawed math problems from almost zero to about 74%. Giving untrained models more thinking time actually made this worse Can models learn to ask clarifying questions instead of guessing?. No rewording fixes a model that has learned not to ask.
Framing does have a real effect, though it shows up somewhere unexpected. LLMs answer more accurately when given an actual clarifying question-and-answer exchange than when given a minimally edited, unambiguous rewrite of the same input. The likely reason is that dialogue looks more like what the model saw in pretraining Do clarifying dialogues help LLMs answer better than rewritten inputs?. So framing affects the model as well as the user. That fits a broader pattern: small wording changes like prepending "wait a minute" Can simple discourse markers make LLMs question harmful assumptions? or adding emotional phrases Can emotional phrases in prompts improve language model performance? shift model behavior without adding any new information.
There's also a caution about measurement. If you judge a clarification by how satisfied users feel, wording can inflate the score. Fluent explanations raise people's trust in LLM answers whether the answers are right or wrong Do LLM explanations actually help people judge answer quality?. A smoothly worded question could feel effective without improving the outcome. The honest answer is that framing is one ingredient among several. Gains that hold up come from asking at the right moment, aiming at the right gap, and training models to want to ask, with good wording doing the last stretch of the work.
Sources 9 notes
Clarifying questions that target concrete information gaps ("What type of monitor?") consistently beat those that ask users to rephrase their needs ("What are you trying to do?"). Users engage most when they can foresee how answering improves results.
INTENT-SIM, which clusters LLM outputs by inferred user intent and measures entropy across clusters, consistently outperforms output-entropy methods at flagging inputs worth clarifying across QA, machine translation, and natural language inference tasks.
The ALFA framework breaks down question quality into theory-grounded attributes (clarity, relevance, specificity) and trains models on 80K attribute-specific preference pairs. Attribute-specific optimization outperforms single-score training, especially in clinical reasoning where asking the right clarifying question directly impacts decision quality.
CollabLLM demonstrates that standard RLHF training optimizes for immediate helpfulness, discouraging models from asking clarifying questions or offering multi-turn insights. Multi-turn-aware rewards that estimate long-term interaction value enable active intent discovery and genuine collaboration.
Reinforcement learning training increased proactive critical thinking accuracy from 0.15% to 73.98% on deliberately flawed math problems. Notably, inference-time scaling degraded this ability in untrained models but improved it after RL training, suggesting the capability is learnable but fragile without explicit training.
Show all 9 sources
LLMs answer more accurately when given clarifying questions and answers than when given minimally-edited disambiguated versions of the same ambiguous input. The gap exists because clarifying exchanges match the model's pretraining distribution better than artificially constrained rewrites.
Two minimal prompting interventions achieve 70% improvement on Cancer-Myth and 40% on SAGE-Eval across six LLMs. The discourse marker 'wait a minute' appears to work by triggering epistemic vigilance rather than adding new information.
Testing EmotionPrompt across ChatGPT, Bard, and Llama 2 showed consistent performance gains from appending psychological phrases like "This is very important to my career." The effect works through motivational framing rather than new information, with positive emotional words driving over 50% of improvements.
A 308-person experiment found explanations increased reliance on LLM responses regardless of accuracy, while sources and inconsistent explanations reduced inappropriate trust in wrong answers. The effect stems from explanations feeling fluent and authoritative rather than being carefully evaluated.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Knowing but Not Showing: LLMs Recognize Ambiguity but Rarely Ask Clarifying Questions
- HumanAgencyBench: Scalable Evaluation of Human Agency Support in AI Assistants
- Clarify When Necessary: Resolving Ambiguity Through Interaction with LMs
- QuestBench: Can LLMs ask the right question to acquire information in reasoning tasks?
- Referential ambiguity and clarification requests: comparing human and LLM behaviour
- Aligning LLMs to Ask Good Questions A Case Study in Clinical Reasoning
- Can Large Language Models Reason and Optimize Under Constraints?
- Accommodation and Epistemic Vigilance: A Pragmatic Account of Why LLMs Fail to Challenge Harmful Beliefs