{
 "@context": "https://schema.org",
 "@type": "ScholarlyArticle",
 "@id": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#paper",
 "aboutThisFile": {
  "whatThisIs": "A research brief: a structured summary of one paper, prepared for seanrintel.com from the paper's full text.",
  "howItWasMade": "This brief was prepared with AI assistance and published by Sean Rintel. It is not the authors' published text.",
  "whatIsVerbatim": "Only 'quote' fields, figure captions and 'abstract' are the authors' own words. Research questions follow the paper's wording where it states them. Takeaway, summaries, answers, definitions and method are the site's plain-language paraphrase.",
  "howItIsChecked": "Before publication every quote is matched against the paper and every number in the brief is found in it. 'verified' is the date of the last check.",
  "howToUse": [
   "For a quick answer, use 'takeaway' and 'questions'. For evidence, use 'findings' (each has a verbatim quote, its location in the paper and a 'url' anchored to that finding on the paper's page).",
   "Locations such as '§5.2, p. 9' refer to the version at 'pageNumbersReferTo'.",
   "For anything not covered here, read the paper itself: 'fullTextSources' lists free versions first.",
   "If 'researchOverview' is present, the paper is part of one of Sean Rintel's research overviews: it gives the overview's thesis, the claims this paper is evidence for and its role in each, and the other papers in the overview.",
   "To find other papers by Sean Rintel, use the catalogue at https://seanrintel.com/publications/index.json."
  ],
  "authorship": "Sean Rintel's role on each paper: lead author when he is first author, sole author when he is the only author, and otherwise senior author. Senior author here means the senior researcher on the team, whatever his position in the author list.",
  "howToCite": "Cite the paper, not this file: use 'citation' or the ready-made 'bibtex' (DOI where given) and name every author. Attribute paraphrase to the paper's findings, and use quotation marks only for 'quote' text."
 },
 "name": "Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks",
 "author": [
  {
   "@type": "Person",
   "name": "Ian Drosos"
  },
  {
   "@type": "Person",
   "name": "Jack Williams"
  },
  {
   "@type": "Person",
   "name": "Advait Sarkar"
  },
  {
   "@type": "Person",
   "name": "Nicholas Wilson"
  },
  {
   "@type": "Person",
   "@id": "https://seanrintel.com/#sean-rintel",
   "name": "Sean Rintel",
   "url": "https://seanrintel.com/",
   "sameAs": [
    "https://orcid.org/0000-0003-0840-0546",
    "https://scholar.google.com/citations?user=xc25GRgAAAAJ",
    "https://www.wikidata.org/wiki/Q102369654",
    "https://www.microsoft.com/en-us/research/people/serintel/",
    "https://www.linkedin.com/in/seanrintel"
   ]
  },
  {
   "@type": "Person",
   "name": "Payod Panda"
  }
 ],
 "datePublished": "2025-06",
 "doi": "10.1145/3729176.3729203",
 "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html",
 "citation": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203",
 "bibtex": "@inproceedings{drosos2025dynamic,\nauthor = {Drosos, Ian and Williams, Jack and Sarkar, Advait and Wilson, Nicholas and Rintel, Sean and Panda, Payod},\ntitle = {Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks},\nbooktitle = {CHIWORK 2025},\nyear = {2025},\nmonth = {June},\npublisher = {ACM},\nurl = {https://doi.org/10.1145/3729176.3729203},\nnote = {Microsoft Foundry Labs},\ndoi = {10.1145/3729176.3729203},\n}",
 "abstract": "Effective prompting of generative AI is challenging for many users, particularly in expressing context for comprehension tasks such as explaining spreadsheet formulas, Python code, and text passages. Prompt middleware aims to address this barrier by assisting in prompt construction, but barriers remain for users in expressing adequate control so that they can receive AI-responses that match their preferences. We conduct a formative survey (n=38) investigating user needs for control over AI-generated explanations in comprehension tasks, which uncovers a trade-off between standardized but predictable support for prompting, and adaptive but unpredictable support tailored to the user and task. To explore this trade-off, we implement two prompt middleware approaches: Dynamic Prompt Refinement Control (Dynamic PRC) and Static Prompt Refinement Control (Static PRC). The Dynamic PRC approach generates context-specific UI elements that provide prompt refinements based on the user’s prompt and user needs from the AI, while the Static PRC approach offers a preset list of generally applicable refinements. We evaluate these two approaches with a controlled user study (n=16) to assess the impact of these approaches on user control of AI responses for crafting better explanations. Results show a preference for the Dynamic PRC approach as it afforded more control, lowered barriers to providing context, and encouraged exploration and reflection of the tasks, but that reasoning about the effects of different generated controls on the final output remains challenging. Drawing on participant feedback, we discuss design implications for future Dynamic PRC systems that enhance user control of AI responses. Our findings suggest that dynamic prompt middleware can improve the user experience of generative AI workflows by affording greater control and guide users to a better AI response.",
 "fullTextSources": [
  {
   "kind": "freeAccess",
   "url": "https://doi.org/10.1145/3729176.3729203",
   "label": "DOI (Free Access)"
  },
  {
   "kind": "freeAccess",
   "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.pdf",
   "label": "Author's Copy"
  }
 ],
 "fullTextAvailable": true,
 "takeaway": "Ephemeral UI (controls an AI generates fresh for each prompt) can help people steer AI explanations. Knowledge workers rated this dynamic prompt middleware more effective than fixed controls and said it eased context-giving and prompted reflection. But how the AI would apply an option often stayed hard to predict.",
 "researchOverview": {
  "name": "Better Thinking With AI",
  "page": "https://seanrintel.com/better-thinking-with-ai.html",
  "json": "https://seanrintel.com/overviews/better-thinking-with-ai.json",
  "markdown": "https://seanrintel.com/overviews/better-thinking-with-ai.md",
  "thesis": "Generative AI moves people's thinking into stating what they want and judging what they get. The easiest way to use it can help least, so good tools ask people to think first.",
  "claimsThisPaperFeeds": [
   {
    "claim": "What feels helpful is not always what helps",
    "role": "supporting evidence",
    "url": "https://seanrintel.com/better-thinking-with-ai.html#a4"
   },
   {
    "claim": "Asking for goals before AI drafts can improve the work",
    "role": "supporting evidence",
    "url": "https://seanrintel.com/better-thinking-with-ai.html#a6"
   }
  ],
  "otherPapers": [
   {
    "study": "Review: productivity loss with AI",
    "title": "Ironies of Generative AI: Understanding and Mitigating Productivity Loss in Human-AI Interaction",
    "page": "https://seanrintel.com/publications/ironies-of-generative-ai-understanding-and-mitigating-productivity-loss-in-human-ai-interaction.html",
    "researchBrief": "https://seanrintel.com/publications/ironies-of-generative-ai-understanding-and-mitigating-productivity-loss-in-human-ai-interaction.json"
   },
   {
    "study": "Survey: critical thinking at work",
    "title": "The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers",
    "page": "https://seanrintel.com/publications/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers.html",
    "researchBrief": "https://seanrintel.com/publications/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers.json"
   },
   {
    "study": "Position paper: what skill at work means",
    "title": "The Future of Skill: What Is It to Be Skilled at Work?",
    "page": "https://seanrintel.com/publications/the-future-of-skill-what-is-it-to-be-skilled-at-work.html",
    "researchBrief": "https://seanrintel.com/publications/the-future-of-skill-what-is-it-to-be-skilled-at-work.json"
   },
   {
    "study": "Conceptual analysis: metacognition and AI",
    "title": "The Metacognitive Demands and Opportunities of Generative AI",
    "page": "https://seanrintel.com/publications/the-metacognitive-demands-and-opportunities-of-generative-ai.html",
    "researchBrief": "https://seanrintel.com/publications/the-metacognitive-demands-and-opportunities-of-generative-ai.json"
   },
   {
    "study": "Prompting study: AI-assisted data analysis",
    "title": "“It’s like a rubber duck that talks back”: Understanding Generative AI-Assisted Data Analysis Workflows through a Participatory Prompting Study",
    "page": "https://seanrintel.com/publications/its-like-a-rubber-duck-that-talks-back-understanding-generative-ai-assisted-data-analysis-workflows-through-a-participatory-prompting-study.html",
    "researchBrief": "https://seanrintel.com/publications/its-like-a-rubber-duck-that-talks-back-understanding-generative-ai-assisted-data-analysis-workflows-through-a-participatory-prompting-study.json"
   },
   {
    "study": "Method paper: participatory prompting",
    "title": "Participatory prompting: a user-centric research method for eliciting AI assistance opportunities in knowledge workflows",
    "page": "https://seanrintel.com/publications/participatory-prompting-a-user-centric-research-method-for-eliciting-ai-assistance-opportunities-in-knowledge-workflows.html",
    "researchBrief": "https://seanrintel.com/publications/participatory-prompting-a-user-centric-research-method-for-eliciting-ai-assistance-opportunities-in-knowledge-workflows.json"
   },
   {
    "study": "Commentary: role-based prompts",
    "title": "Productivity implications for generative AI role-based prompts as a networked hermeneutic (Commentary)",
    "page": "https://seanrintel.com/publications/productivity-implications-for-generative-ai-role-based-prompts-as-a-networked-hermeneutic.html",
    "researchBrief": "https://seanrintel.com/publications/productivity-implications-for-generative-ai-role-based-prompts-as-a-networked-hermeneutic.json"
   },
   {
    "study": "Randomised trial: how students learn with AI",
    "title": "Learning behavior accounts for background-related advantage in AI-assisted education",
    "page": "https://seanrintel.com/publications/learning-behavior-accounts-for-background-related-advantage-in-ai-assisted-education.html",
    "researchBrief": "https://seanrintel.com/publications/learning-behavior-accounts-for-background-related-advantage-in-ai-assisted-education.json"
   },
   {
    "study": "Randomised experiment: note-taking versus AI for reading",
    "title": "Effects of LLM Use and Note-Taking On Reading Comprehension and Memory: A Randomised Experiment in Secondary Schools",
    "page": "https://seanrintel.com/publications/effects-of-llm-use-and-note-taking-on-reading-comprehension-and-memory-a-randomised-experiment-in-secondary-schools.html",
    "researchBrief": "https://seanrintel.com/publications/effects-of-llm-use-and-note-taking-on-reading-comprehension-and-memory-a-randomised-experiment-in-secondary-schools.json"
   },
   {
    "study": "Interviews: AI in higher education",
    "title": "The New Calculator? Practices, Norms, and Implications of Generative AI in Higher Education",
    "page": "https://seanrintel.com/publications/the-new-calculator-practices-norms-and-implications-of-generative-ai-in-higher-education.html",
    "researchBrief": "https://seanrintel.com/publications/the-new-calculator-practices-norms-and-implications-of-generative-ai-in-higher-education.json"
   },
   {
    "study": "Randomised experiment: scaffolds for AI-assisted writing",
    "title": "Metacognitive Scaffolds Create Productive Friction for AI-Assisted Knowledge Work",
    "page": "https://seanrintel.com/publications/metacognitive-scaffolds-create-productive-friction-for-ai-assisted-knowledge-work.html",
    "researchBrief": "https://seanrintel.com/publications/metacognitive-scaffolds-create-productive-friction-for-ai-assisted-knowledge-work.json"
   },
   {
    "study": "Workshop proposal: AI as a tool for thought",
    "title": "Tools for Thought: Understanding, Protecting, and Augmenting Human Cognition with Generative AI — From Vision to Implementation",
    "page": "https://seanrintel.com/publications/tools-for-thought-understanding-protecting-and-augmenting-human-cognition-with-generative-ai-from-vision-to-implementation.html",
    "researchBrief": "https://seanrintel.com/publications/tools-for-thought-understanding-protecting-and-augmenting-human-cognition-with-generative-ai-from-vision-to-implementation.json"
   },
   {
    "study": "System design: AI agents for diverse ideas",
    "title": "YES AND: A Generative AI Multi-Agent Framework for Enhancing Diversity of Thought in Individual Ideation for Problem-Solving Through Confidence-Based Agent Turn-Taking",
    "page": "https://seanrintel.com/publications/yes-and-a-generative-ai-multi-agent-framework-for-enhancing-diversity-of-thought-in-individual-ideation-for-problem-solving-through-confidence-based-agent-turn-taking.html",
    "researchBrief": "https://seanrintel.com/publications/yes-and-a-generative-ai-multi-agent-framework-for-enhancing-diversity-of-thought-in-individual-ideation-for-problem-solving-through-confidence-based-agent-turn-taking.json"
   },
   {
    "study": "Interviews and experiment: how people and machines interpret",
    "title": "Meeting in the Middle: The Interpretation Gap Between People and Machines",
    "page": "https://seanrintel.com/publications/meeting-in-the-middle-the-interpretation-gap-between-people-and-machines.html",
    "researchBrief": "https://seanrintel.com/publications/meeting-in-the-middle-the-interpretation-gap-between-people-and-machines.json"
   }
  ]
 },
 "whyItMatters": {
  "stateOfTheArt": {
   "summary": "Prompting generative AI is hard, from expressing context in text to the metacognitive demands it makes. Graphical 'prompt middleware' helps users build prompts, but has not been generated dynamically for the user's task in comprehension and learning.",
   "quote": "close attention has not yet been paid to generating middleware dynamically, adapted to the user's specific task at hand, in the context of comprehension and learning tasks",
   "location": "§2, p. 3"
  },
  "contribution": {
   "summary": "Controls designed in advance cannot anticipate the breadth of knowledge work. The authors build a system that generates controls for each request instead, compare it in a controlled study with fixed generic controls, and derive design implications.",
   "quote": "In this paper, we propose instead to generate prompt middleware customized to the user and task.",
   "location": "§1, p. 2"
  },
  "whoCares": {
   "summary": "Designers and developers of generative AI tools, especially for comprehension and learning in knowledge work, and researchers of AI as a tool for thought.",
   "quote": "Our findings suggest that dynamic prompt middleware can improve the user experience of generative AI workflows.",
   "location": "Abstract, p. 1"
  }
 },
 "researchQuestions": [
  "What are user preferences for control of AI responses in comprehension tasks? [RQ1]",
  "Which form of control is perceived as most effective for controlling AI responses? [RQ2]",
  "What are the trade-offs of Dynamic PRC and Static PRC prompt middleware for controlling AI responses? [RQ3]"
 ],
 "concepts": [
  {
   "term": "Prompt middleware",
   "definition": "Graphical interface elements that help users construct and manipulate prompts instead of typing everything.",
   "quote": "a general category of techniques that use graphical interface elements for manipulating prompts, sometimes referred to as\" prompt middleware\" [44], to ease the cognitive burdens of text entry and manipulation",
   "location": "§2, p. 3"
  },
  {
   "term": "Dynamic Prompt Refinement Control (Dynamic PRC)",
   "definition": "The authors' system: an Option Module analyses the user's prompt and generates controls (radio buttons, checkboxes, free text) whose selections refine the prompt sent to a Chat Module.",
   "quote": "Dynamic Prompt Refinement Control (Dynamic PRC), which generates UI elements for prompt refinement based on the user's specific prompt",
   "location": "Abstract, p. 1"
  },
  {
   "term": "Static Prompt Refinement Control (Static PRC)",
   "definition": "The comparison condition: a fixed list of generic controls derived from the formative survey, offered for any prompt.",
   "quote": "Static Prompt Refinement Control (Static PRC), which offers generic controls",
   "location": "Abstract, p. 1"
  },
  {
   "term": "Session options",
   "definition": "Requested in natural language, edited as JSON or pinned from inline options. The paper's definition:",
   "quote": "Session options are options that apply to every prompt and AI response in a given session with the Dynamic PRC system.",
   "location": "§4.2, p. 6"
  }
 ],
 "method": {
  "design": "A formative survey with a design probe, then a controlled within-subjects lab study comparing Dynamic PRC and Static PRC, counterbalanced for condition and task-set order.",
  "prototype": "Dynamic Prompt Refinement Control (Dynamic PRC), a web-based chat assistant built with React and TypeScript, with responses from gpt4-turbo. The user types a request, for example to explain a spreadsheet formula. One language model (the Option Module) reads the request and generates controls tailored to it, such as radio buttons for explanation detail level or learning objective, with starting values already chosen. A second model (the Chat Module) answers using the request plus those selections, and regenerates the answer whenever the user changes a control. Users can also ask in plain language for controls that apply to the whole session (for example, response format), edit them directly as JSON, or pin a generated control so it applies to every later request. It was compared with Static PRC (the same chat with a fixed list of generic explanation controls derived from the formative survey).",
  "prototypeFigure": 1,
  "setting": "Formative survey of knowledge workers at a large software company. Lab study with knowledge workers from diverse occupations, located on various continents.",
  "sample": "Formative survey n = 38. Lab study n = 16 (8 men, 8 women), mostly regular generative AI users.",
  "n": 16,
  "duration": "Each lab session took approximately 90 minutes: 6 tasks (code explanation, complex topic understanding, skill learning), 7 minutes per task.",
  "analysis": "Post-task, post-condition and post-study questionnaires (including an adapted NASA Task Load Index) compared with Mann-Whitney U tests. Think-aloud and semi-structured interview data analysed with iterative open coding and thematic analysis, refined by negotiated agreement."
 },
 "findings": [
  {
   "headline": "Preferred, though not more successful.",
   "summary": "Knowledge workers in a lab study used both tools on comprehension tasks. Compared directly, most preferred the controls generated for each request over the fixed ones, and felt slightly more successful but also slightly more stretched with them. Their ratings of task success after each tool did not differ significantly.",
   "quote": "all bumper rails at the bowling alley for getting to where you need to go",
   "location": "§6.3, p. 13",
   "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-p-13",
   "cite": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203 (§6.3, p. 13)"
  },
  {
   "headline": "Rated more effective for steering.",
   "summary": "Participants rated the generated controls significantly more effective than the fixed ones at controlling the AI's explanations, and said they needed less extra control. The generated controls were not rated more helpful for understanding the task's concepts, and for explaining Python code the fixed controls helped understanding more.",
   "quote": "found the controls generated in Dynamic PRC to be significantly more effective at helping control AI output and explanations over Static PRC",
   "location": "§6.2, p. 12",
   "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-2-p-12",
   "cite": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203 (§6.2, p. 12)"
  },
  {
   "headline": "Easier to give the AI context.",
   "summary": "Several participants said the generated controls made it easier to tell the AI about their situation and needs. For some, the options shown with each answer replaced much of the typing of follow-up prompts.",
   "quote": "I felt like dynamic would really help me shortcut what I'm doing manually today.",
   "location": "§6.3, p. 12",
   "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-p-12",
   "cite": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203 (§6.3, p. 12)"
  },
  {
   "headline": "A scaffold for thinking.",
   "summary": "Several participants said the generated options worked as guidance, showing them what they could ask the AI for and prompting them to reflect on their task.",
   "quote": "were things that I would maybe assume the system couldn't handle or that it wouldn't be able to do.",
   "location": "§6.3, p. 13",
   "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-p-13-2",
   "cite": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203 (§6.3, p. 13)"
  },
  {
   "headline": "Changing options took effort.",
   "summary": "Some participants found the generated controls more mentally demanding, because a new set of options appeared with each request, though self-reported mental load did not differ significantly between the two tools.",
   "quote": "required more activation [mental] energy",
   "location": "§6.3, pp. 13–14",
   "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-pp-13-14",
   "cite": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203 (§6.3, pp. 13–14)"
  },
  {
   "headline": "Hard to predict what an option does.",
   "summary": "With both tools, participants found it hard to tell how the AI would apply a selected option. Responses sometimes changed in unexpected places, changed too little, or ignored one option because of another.",
   "quote": "some options nullify other options that you put in",
   "location": "§6.3, pp. 14–15",
   "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-pp-14-15",
   "cite": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203 (§6.3, pp. 14–15)"
  }
 ],
 "questions": [
  {
   "question": "Should the controls in an AI interface adapt to each request, or stay fixed?",
   "answer": "Each has strengths. Controls generated for each request make it easier to give the AI context, and people preferred them and rated them more effective for steering explanations. Fixed controls were not rated significantly less useful for understanding concepts, and helped more when explaining code.",
   "source": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203",
   "basedOn": [
    {
     "location": "§6.3, p. 13",
     "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-p-13"
    },
    {
     "location": "§6.2, p. 12",
     "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-2-p-12"
    },
    {
     "location": "§6.3, p. 12",
     "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-p-12"
    }
   ]
  },
  {
   "question": "Are controls that change with every request harder to use?",
   "answer": "Somewhat. Because new options appear with each request, some people felt they took more mental energy, and overall they were judged slightly more demanding. Yet self-reported mental load did not differ significantly from fixed controls.",
   "source": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203",
   "basedOn": [
    {
     "location": "§6.3, p. 13",
     "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-p-13"
    },
    {
     "location": "§6.3, pp. 13–14",
     "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-pp-13-14"
    }
   ]
  },
  {
   "question": "Can prompt controls help people think about their task, not just the AI's output?",
   "answer": "They can. Options generated for a request show what one could ask for. Several people said this helped them work out what they did not know and reflect on their task, not only shape the answer.",
   "source": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203",
   "basedOn": [
    {
     "location": "§6.3, p. 13",
     "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-p-13-2"
    }
   ]
  },
  {
   "question": "Why is it hard to predict what an AI control will do?",
   "answer": "Users could not always tell what an option would change or whether the AI had applied it. Options altered unexpected parts of a response, changed too little, or cancelled each other out. Trying each option and watching the answer regenerate helped.",
   "source": "Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203",
   "basedOn": [
    {
     "location": "§6.3, pp. 14–15",
     "url": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.html#at-s6-3-pp-14-15"
    }
   ]
  }
 ],
 "cites": [
  {
   "url": "https://seanrintel.com/publications/its-like-a-rubber-duck-that-talks-back-understanding-generative-ai-assisted-data-analysis-workflows-through-a-participatory-prompting-study.html",
   "title": "“It’s like a rubber duck that talks back”: Understanding Generative AI-Assisted Data Analysis Workflows through a Participatory Prompting Study",
   "location": "p. 18"
  },
  {
   "url": "https://seanrintel.com/publications/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers.html",
   "title": "The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers",
   "location": "p. 19"
  },
  {
   "url": "https://seanrintel.com/publications/the-metacognitive-demands-and-opportunities-of-generative-ai.html",
   "title": "The Metacognitive Demands and Opportunities of Generative AI",
   "location": "p. 19"
  }
 ],
 "quantitative": [
  {
   "measure": "Formative survey: importance of having control over the AI response (5-point scale)",
   "value": "Median=4.0 [Agree], Mean=4.3",
   "location": "§3.3, p. 4"
  },
  {
   "measure": "Overall tool preference (1 = Dynamic PRC, 7 = Static PRC, 4 = equal)",
   "value": "median=2.0, mean=2.81",
   "location": "§6.1, p. 10"
  },
  {
   "measure": "Which tool was more demanding to communicate with (1 = Dynamic PRC, 7 = Static PRC)",
   "value": "median=3.0, mean=3.31",
   "location": "§6.1, p. 10"
  },
  {
   "measure": "Helpfulness for controlling AI responses (7-point scale), Dynamic vs Static PRC",
   "value": "Dynamic PRC median=6.5, mean=6.38 vs Static PRC median=6.0, mean=5.69",
   "location": "§6.1, p. 11"
  },
  {
   "measure": "Controls effective at helping control AI output, Dynamic vs Static PRC",
   "value": "mean=6.44 vs mean=5.81; U: 73.5, P-value: .0245, r: .363 (medium effect), power: .275",
   "location": "§6.2, p. 12"
  },
  {
   "measure": "Needed greater control over the explanations, Dynamic vs Static PRC",
   "value": "mean=2.75 vs mean=4.44; U: 190.5, P-value: .0174, r: .42 (medium effect), power: .349",
   "location": "§6.2, p. 12"
  },
  {
   "measure": "Controls useful for understanding task concepts, Dynamic vs Static PRC (not significant)",
   "value": "mean=6.44 vs mean=5.88; U: 94, P-value: .164",
   "location": "§6.2, p. 12"
  },
  {
   "measure": "Per-task rating of how much options helped understanding, Task 4 (explaining Python code): Dynamic PRC worse than Static PRC (the only significant per-task difference)",
   "value": "U: 12.0, P-value: 0.0355",
   "location": "§C.1, p. 22"
  }
 ],
 "implications": [
  {
   "summary": "Dynamic controls should give appropriate mental load, for example through progressive disclosure, rather than overwhelming users.",
   "quote": "Dynamic PRC should provide appropriate mental load to users.",
   "location": "§7.1, p. 15"
  },
  {
   "summary": "Give users control over and transparency about how and where the AI applies options, for example with response diffs.",
   "quote": "Dynamic PRC should provide control of how and where the AI applies options.",
   "location": "§7.1, p. 15"
  },
  {
   "summary": "Use relevant user context, such as past AI use or open files, to generate and pre-select controls.",
   "quote": "Dynamic PRC should gather and leverage relevant context for the user.",
   "location": "§7.1, p. 15"
  },
  {
   "summary": "Dynamic options could act as metacognitive interventions, making such systems tools for thought.",
   "quote": "This suggests that Dynamic PRC can potentially act as a tool for thought by assisting user thinking around their tasks.",
   "location": "§7.2, p. 16"
  }
 ],
 "limitations": [
  {
   "summary": "The authors did not compare either system with a baseline chat interface.",
   "quote": "we did not compare Dynamic PRC and Static PRC to a baseline chat",
   "location": "§7.4, p. 16"
  },
  {
   "summary": "Formative survey participants all worked at one tech-forward organisation.",
   "quote": "the participants all worked at the same organization",
   "location": "§7.4, p. 16"
  },
  {
   "summary": "Only comprehension tasks were evaluated, and not over longer periods of use.",
   "quote": "future work should study how the effects of Dynamic PRC over longer periods of usage",
   "location": "§7.4, p. 17"
  }
 ],
 "figures": [
  {
   "src": "https://seanrintel.com/publications/figures/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks/fig-1.webp",
   "caption": "Figure 1: Schematic overview of Dynamic Prompt Refinement Control interface for generating inline prompt refinement options to increase user control of AI-generated explanations (derived from real system use). (1) User prompts the system. (2) The Option Module takes the user's prompt as context to generate prompt options which provide prompt refinements for the user to select. (3) The Chat Module uses the user's prompt and pre-selected options to generate an initial response. (4) The user can initiate refinements by selecting their preferred options in the UI. (5) On each change, the Chat Module regenerates the response based on the new selections.",
   "alt": "Two chat panels for the prompt 'How do I use PyTorch?': generated options for experience level and learning objective appear above the answer, and switching experience from Beginner to Advanced regenerates a more advanced answer.",
   "location": "Fig. 1, p. 1",
   "credit": "Figure 1 from Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203"
  },
  {
   "src": "https://seanrintel.com/publications/figures/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks/fig-12.webp",
   "caption": "Figure 12: Participant rated effectiveness of each condition in the control of AI responses for the comprehension tasks they saw.",
   "alt": "Box plots on a 1-7 agreement scale comparing Dynamic and Static PRC: Dynamic controls rated more effective for controlling AI output, similar for understanding, and with less need for greater control.",
   "location": "Fig. 12, p. 12",
   "credit": "Figure 12 from Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203"
  },
  {
   "src": "https://seanrintel.com/publications/figures/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks/fig-10.webp",
   "caption": "Figure 10: Overall direct comparison of conditions for participant tool preference, mental demand, feelings of being rushed, successful, level of effort, and feelings of annoyance.",
   "alt": "Box plots on a scale from 1 (Dynamic PRC) to 7 (Static PRC), 4 equal: preference leans to Dynamic (median 2.0). Dynamic rated slightly more mentally demanding and effortful. Rushed and annoyed sit near equal.",
   "location": "Fig. 10, p. 10",
   "credit": "Figure 10 from Ian Drosos, Jack Williams, Advait Sarkar, Nicholas Wilson, Sean Rintel, and Payod Panda. 2025. Dynamic Prompt Middleware: Contextual Prompt Refinement Controls for Comprehension Tasks. CHIWORK 2025. https://doi.org/10.1145/3729176.3729203"
  }
 ],
 "pageNumbersReferTo": "https://seanrintel.com/publications/dynamic-prompt-middleware-contextual-prompt-refinement-controls-for-comprehension-tasks.pdf",
 "verified": "2026-10-25",
 "studyFigure": {
  "about": "The opening figure on this paper's page, made for the site with AI assistance from this verified brief. It is not the paper's own figure. Evidence entries index this file's quantitative, findings and limitations arrays.",
  "study": {
   "kind": "comparison",
   "summary": "Controlled lab study with 16 knowledge workers. Each used two AI chat tools on comprehension tasks in counterbalanced order, in a session of about 90 minutes, after a formative survey of 38 knowledge workers.",
   "columns": [
    "Comprehension tasks",
    "After each tool"
   ],
   "groups": [
    {
     "name": "Generated controls",
     "n": 16,
     "cells": [
      "Controls generated by the AI for each request",
      "Questionnaire on the controls"
     ],
     "highlight": [
      0
     ],
     "nLabel": "all 16 people"
    },
    {
     "name": "Fixed controls",
     "n": 16,
     "cells": [
      "A fixed list of generic controls",
      "Questionnaire on the controls"
     ],
     "nLabel": "all 16 people"
    }
   ],
   "after": {
    "label": "End of session: direct comparison",
    "note": "The same 16 people used both tools, then were interviewed"
   }
  },
  "tool": {
   "steps": [
    {
     "title": "Type a request",
     "text": "The person asks the AI chat to explain something, for example a spreadsheet formula."
    },
    {
     "title": "Generate controls",
     "text": "One AI model reads the request and generates controls tailored to it, with starting values already chosen.",
     "fields": [
      "detail level",
      "learning objective"
     ],
     "part": true
    },
    {
     "title": "Answer",
     "text": "A second AI model answers using the request plus the selected options."
    },
    {
     "title": "Adjust and regenerate",
     "text": "Whenever the person changes a control, the answer is regenerated.",
     "part": true
    },
    {
     "title": "Keep options for the session",
     "text": "The person can ask in plain language for controls that apply to every request, or pin a generated control.",
     "part": true
    }
   ],
   "note": "Steps 2, 4 and 5 are the generated controls. The fixed controls tool had the same chat with a fixed list of generic explanation controls from the formative survey."
  },
  "findings": {
   "legend": "comparison",
   "stages": [
    {
     "title": "Rated after each tool",
     "sub": "the same people rated generated and fixed controls",
     "items": [
      {
       "dir": "up",
       "text": "Generated controls: rated more effective at steering the AI's explanations.",
       "detail": "Self-reported, mean 6.44 against 5.81 on a 1 to 7 scale. A medium effect, r = .363, though 16 people gave the test low power.",
       "evidence": [
        "quantitative[4]",
        "findings[1]"
       ]
      },
      {
       "dir": "up",
       "text": "Generated controls: people said they needed less extra control over explanations.",
       "detail": "Self-reported, mean 2.75 against 4.44 on a 1 to 7 scale. A medium effect, r = .42.",
       "evidence": [
        "quantitative[5]",
        "findings[1]"
       ]
      },
      {
       "dir": "flat",
       "text": "Generated controls: rated about as useful as fixed controls for understanding concepts.",
       "detail": "Self-reported, mean 6.44 against 5.88. Not a reliable difference, p = .164.",
       "evidence": [
        "quantitative[6]",
        "findings[1]"
       ]
      },
      {
       "dir": "down",
       "text": "Generated controls: rated less helpful for understanding when explaining Python code.",
       "detail": "Self-reported for that task, p = 0.0355. It was the only task where ratings of understanding differed reliably.",
       "evidence": [
        "quantitative[7]",
        "findings[1]"
       ]
      },
      {
       "dir": "flat",
       "text": "Generated controls: task success was rated about the same as with fixed controls.",
       "detail": "Self-reported after each tool. Ratings of task success did not differ reliably.",
       "evidence": [
        "findings[0]"
       ]
      }
     ]
    },
    {
     "title": "Compared directly, then discussed",
     "sub": "end of session preference and interviews",
     "items": [
      {
       "dir": "none",
       "text": "Generated controls: most people preferred them to fixed controls.",
       "detail": "Self-reported on a scale from 1, generated, to 7, fixed, median 2.0. People also judged generated controls slightly more demanding.",
       "evidence": [
        "quantitative[1]",
        "quantitative[2]",
        "findings[0]"
       ]
      },
      {
       "dir": "none",
       "text": "Generated controls: several people said they made it easier to give the AI context.",
       "detail": "For some, options shown with each answer replaced much typing of follow-up prompts. Several said the options also prompted reflection on the task.",
       "evidence": [
        "findings[2]",
        "findings[3]"
       ]
      },
      {
       "dir": "none",
       "text": "Both groups: with either tool, people found an option's effect hard to predict.",
       "detail": "Responses sometimes changed in unexpected places, changed too little, or ignored one option because of another.",
       "evidence": [
        "findings[5]"
       ]
      }
     ]
    }
   ],
   "legendText": "↑ better than with fixed controls   ↓ worse   – no reliable difference"
  },
  "keyTerm": "dynamic prompt middleware",
  "takeaway": "Dynamic prompt middleware, which generates controls for each request, can help people steer AI explanations and reflect on their task. People may also need to see how options change the answer."
 }
}