Can users articulate their intent before exploring what an AI system finds?
This reads the question as: does a user arrive with a fully-formed intent to state, or does intent only take shape through interaction with what the system surfaces? The corpus leans hard toward the second.
This explores whether intent comes before exploration or grows out of it — and the collection's answer is that the premise of "articulate first, explore second" is mostly backwards. Intent isn't a fixed thing waiting to be typed in; it's something that matures. One line of work frames intent formation as continuous maturation through progressive constraint resolution, with stability that fluctuates rather than a switch that flips from absent to present How do users actually form intent when prompting AI systems?. So asking users to articulate before they've seen anything is asking them to do the part of the work that the exploration itself is supposed to enable.
This is named directly as a "gulf of envisioning": users can't say what they want, and AI, because it responds rather than probes, doesn't help them get there Why can't users articulate what they want from AI?. The interesting move in that work is shifting the cognitive load — instead of open-ended "tell me what you want," the system presents generated options so the user evaluates rather than invents. Articulation becomes recognition. You often don't know your intent until you see a few candidate shapes of it.
The cost of ignoring this is measurable. When users reveal goals incrementally across a conversation, even strong models reach full intent alignment only about 20% of the time, and uncover under 30% of preferences through active querying — they make premature assumptions instead Why do AI agents miss most of what users actually want?. Part of why is structural: conversational models are built to react, not to initiate, plan, or lead, so they don't naturally do the probing that would draw intent out Why can't conversational AI agents take the initiative?.
The corpus also offers a vocabulary for the missing behavior. Conversation analysis supplies "insert-expansions" — a formal account of when an agent should pause to clarify or scope before acting, heading off misunderstanding rather than recovering from it When should AI agents ask users instead of just searching?. And the passivity isn't a hard limit: clarification-seeking and proactivity are trainable, jumping from near-zero to ~74% with reinforcement learning, the real challenge being how to probe without becoming intrusive Why do AI agents fail to take initiative?. Done well, proactivity can cut conversation length by up to 60% Could proactive dialogue make conversations dramatically more efficient?.
So the honest answer to the question is: usually no, and that's fine — intent and exploration are supposed to co-produce each other. The design failure isn't the user's inability to articulate up front; it's a system that demands a finished answer instead of helping author one. The thing worth knowing you wanted to know: the fix isn't a smarter model that reads your mind, it's an interaction that turns "describe what you want" into "react to what I found."
Sources 7 notes
Human intent matures through progressive constraint resolution with fluctuating stability, not as a simple present-or-absent condition. The STORM framework and Clarify metric reveal that AI systems fail partly because they cannot access users' internal cognitive states during this evolution.
Intent develops through interaction, not in isolation. Since AI models respond rather than probe, they miss opportunities to help users discover unarticulated requirements. Structured dialogue that presents model-generated options shifts the cognitive burden from open-ended envisioning to constrained evaluation.
UserBench measured multi-turn interactions where users reveal goals incrementally and found models achieve full intent alignment just 20% of the time. Even top models uncover fewer than 30% of user preferences through active querying, suggesting passivity and premature assumption-making are systematic failures.
Research shows LLMs including ChatGPT cannot initiate topics, plan strategically, or lead conversations because their training optimizes for responding to queries, not creating dialogue from agent goals. This passivity is reinforced by alignment objectives and masked by fluent-sounding outputs.
Tool-enabled LLMs drift from user intent through silent tool chaining. Conversation analysis reveals insert-expansions—clarifying intent, scoping responses, enhancing appeal—as a formal framework for proactive user consultation that prevents misunderstanding instead of recovering from it.
Research shows next-turn reward optimization structurally removes initiative from models, but proactive behaviors like critical thinking and clarification-seeking are trainable (0.15% to 73.98% with RL). The core challenge is balancing proactivity with civility to avoid intrusion.
Simulations show proactivity—providing relevant information without being asked—cuts dialogue turns by 60% in medium-complexity domains. This behavior mirrors human conversation and Grice's maxims but is almost entirely absent from AI datasets and research benchmarks.