Ask Before or After? Clarification Timing and User Experience in Generative AI
AI-initiated clarification questions have shown promise in resolving ambiguity in generative AI systems. However, little is known about how a clarification question reshapes different dimensions of user experience, what trade-offs it introduces, and whether those effects vary by clarification timing and task modality. This study investigates how clarification questions, asked before or after generation,
affect user experience across text generation and image generation tasks. We conducted a controlled task-based study comparing three interaction strategies: (1) No Questions, where the AI generates immediately without asking anything; (2) Ask First, where the AI asks clarification questions before generating; and (3) Ask After, where the AI generates an initial output first, then asks questions to revise it. Thirty participants completed paired text and image generation tasks and provided immediate post-task Likert-scale survey responses across six tasks, followed by semi-structured interviews. We evaluated three UX dimensions, namely perceived Cognitive Overhead, Verification Burden, and Trust and Reliability, as comprehensive indicators of interaction quality, and examined whether clarification timing effects differ by task modality, as text and image generation impose distinct demands on users. Our results show that clarification conditions consistently increased Trust and Reliability relative to no clarification, while also increasing Cognitive Overhead. Timing alone (Ask First vs. Ask After) produced no significant UX differences overall, with one exception: Ask After yielded significantly higher trust in text generation tasks. Qualitative findings reveal that clarification functions as a thinking scaffold rather than purely an output-quality tool, and that its benefit scales with task uncertainty and output irreversibility. This study contributes to generative AI UX research by showing that clarification questions are not merely tools for improving output accuracy, but interaction mechanisms that shape users’ cognitive effort, verification behavior, and trust. Our findings suggest that the value of clarification depends less on timing alone and more on how uncertainty and irreversibility are experienced within a task. These results provide practical design implications for future AI creative tools by identifying when clarification can meaningfully support user thinking and alignment.
Additional Key Words and Phrases: Generative AI, Human-AI Interaction, Clarification Questions, Clarification Timing, User Experience, Task Modality, Text Generation, Image Generation
Research Questions
This study examines three questions about AI-initiated clarification questions: whether the presence of clarification questions addresses the three UX challenges users face, whether their benefits vary by task modality, and whether the timing of clarification questions matters.
We hypothesize that when users can clarify their intent, the AI could better align outputs with their needs. The improved alignment should reduce Cognitive Overhead in crafting prompts, lessen Verification Burden in checking outputs, and increase Trust and Reliability in the system.
- H1a: Clarification questions will decrease Cognitive Overhead relative to No Questions.
- H1b: Clarification questions will decrease Verification Burden relative to No Questions.
- H1c: Clarification questions will increase Trust and Reliability relative to No Questions.
We further expect clarification questions to be moderated by task modality. Image generation imposes greater
communication challenges, as users often struggle to verbalize visual intent in text, making clarification potentially
more impactful.
- H2: Clarification benefits will be larger for image generation than for text generation.
Finally, we examine whether clarification timing—Ask First or Ask After—differentially addresses the three UX
challenges. Prior work offers no basis for predicting which timing is superior, so we treat this as exploratory.
- H3 (exploratory): Ask First and Ask After will differ in their effects on the three UX challenges.