Key Aspects of the Research:

The researchers built HopeBot, a GPT-4o-based chatbot that administers a standardized depression screening questionnaire (PHQ-9). Instead of just reading the questions aloud, the bot held a conversation, could ask follow-up questions when it got ambiguous answers, and kept an empathetic tone, drawing on information from a knowledge base (RAG). The experiment showed that users trusted the results from the conversation with the bot significantly more than the results they got from filling out the same questionnaire on their own.

Key Result: An interactive, empathetic conversation structure that can clarify information makes an LLM agent a more reliable and preferred tool for users compared with filling out a static form.

The Core of the Method, Explained:

The method described in the study is a shift from the "request-response" model to a "guided conversation" model. Instead of trying to write one perfect, exhaustive prompt, the user sets up a multi-step process where the LLM actively helps gather and clarify information.

For the user, the practical technique boils down to three core principles:

1. A phased prompt structure: A complex task is broken into logical stages right inside the LLM's instructions. The study uses this model:

   • Phase 1: Rapport Building. Open with a friendly introduction and ask general warm-up questions.
   • Phase 2: Core Task. Work through the key steps of the task in order.
   • Phase 3: Feedback. A structured wrap-up, summary, and recommendations.

2. A built-in Clarification Loop: This is the heart of the method. The prompt includes a direct instruction for the LLM: if the user's answer is unclear, ambiguous, or incomplete, the model must ask a follow-up question rather than guess or interpret on its own. This sharply reduces the risk of "hallucinations" and wrong conclusions.

3. Contextual enrichment (User-level RAG): HopeBot used RAG to answer based on vetted sources. An everyday user can simulate this by providing relevant context in the prompt (for example, an article's text, a product description, a previous email thread) and instructing the LLM to base its answers only on that information.

Practical Applicability Analysis:

• Direct applicability: Users can start structuring their prompts around the phased model right away. The simplest, most powerful thing you can do is add this instruction to any complex prompt: "If any part of my request is unclear to you, ask me clarifying questions before generating an answer". That directly implements the study's main finding.

• Conceptual value: The study helps you see that to get a high-quality result, you need to "help" the model rather than expect a miracle. It shows that an LLM does better as an assistant-interviewer than as an all-knowing oracle. The key concept: the quality of the final output depends directly on the quality of the data gathered during the conversation.

• Adaptation potential: The method is universal. Instead of depression screening, you can adapt it to any task that requires gathering information from the user:

  • Trip planning: Phase 1 (Greeting) -> Phase 2 (Pinning down budget, dates, and vacation preferences through follow-up questions) -> Phase 3 (A finished itinerary).
  • Writing a resume: Phase 1 (Explaining the goal) -> Phase 2 (Gathering information on work experience, skills, and education through a series of questions) -> Phase 3 (A finished resume in several versions).
  • Product-buying advice: Phase 1 (Greeting) -> Phase 2 (Figuring out needs, budget, and use cases) -> Phase 3 (A shortlist of the top 3 models with explanations).

Practical Example:

You are an experienced marketing strategist who specializes in helping small businesses. Your job is to help me develop a content plan for my project. Keep the conversation friendly and supportive.

Follow this three-phase protocol strictly:

**Phase 1: Getting to know the project.**
Start with a greeting. Then ask me a few key questions to understand the essence of my business. For example: "Tell me about your product or service," "Who is your target audience?", "What are your main goals (increase sales, build awareness, etc.)?" Wait for my answers.

**Phase 2: Gathering and clarifying information (the most important stage).**
Once I've answered the Phase 1 questions, we'll move on to the details. Your task is to suggest content ideas.

**[KEY CLARIFICATION RULE]:** If during our discussion my answers are general or vague (for example, if I say "I want social media posts" or "we need something about our advantages"), you **MUST** ask clarifying questions to get specifics.
Examples of clarifying questions:
- "Great, social media posts. Which platform should we focus on first: Instagram, TikTok, or LinkedIn?"
- "When you talk about advantages, what's the one most important advantage we should get across to customers first?"
- "Which content format do you think will appeal most to your audience: short videos, articles, infographics?"

Keep asking questions until you have enough information to build a detailed plan.

**Phase 3: Building the content plan.**
Once you've gathered all the information you need, pull it together and present it as a structured one-week content plan in table format. The table should have the columns: "Day of the week," "Post topic," "Format," "Key idea/talking points."

Let's begin. I'm ready to answer your questions.

Why It Works:

This prompt works because it faithfully reproduces the successful model from the HopeBot study, transplanting it from healthcare into marketing:

1. A structured conversation: The prompt doesn't require the user to hand over all the information at once. It sets a clear three-phase structure that guides the LLM and makes the interaction predictable and logical.
2. Active clarification: The [KEY CLARIFICATION RULE] construction is a direct implementation of the study's main finding. It forces the LLM to ask again instead of filling in the blanks. That prevents useless, generic content (like "write interesting posts") and leads to a personalized, relevant plan.
3. More trust and engagement: Just as in the study, the user feels "heard." Clarifying questions create the impression of working with a thoughtful specialist rather than a machine. That increases trust in the final result and engagement in the process.

Another Practical Example

You are a personal event-planning assistant. Your job is to help me organize a birthday celebration. Your communication style is caring, detail-oriented, and highly organized.

We'll work step by step, in three phases.

**Phase 1: Setting the overall frame.**
Start with a friendly greeting. Then ask me 3-4 main questions to understand the big picture of the party. For example: "Who are we throwing the birthday for, and how old are they turning?", "What's the approximate date and budget?", "How many guests are you planning to invite?"

**Phase 2: Detailing and clarifying.**
After my answers, we'll dig into the details. Your goal is to gather all the information needed to put together a plan.

**[INTERACTIVITY RULE]:** If my answers are incomplete or leave you too much room for imagination (for example, if I say "the food should be tasty" or "entertainment for everyone"), you **MUST** ask clarifying questions to narrow down the options.
Examples of clarifying questions:
- "What cuisine does the birthday person prefer: Italian, Asian, maybe home-style comfort food?"
- "When you say 'entertainment for everyone,' will there be both kids and adults? Should we plan a separate program for the kids?"
- "Which vibe feels more like you: a loud party with dancing or a cozy evening with close friends and family?"

Your job at this stage is to be my brainstorming partner, suggest options, and pin down my preferences.

**Phase 3: Putting together the final plan.**
Once you've gathered all the details, give me a final birthday prep checklist. It should be broken into categories: "Venue," "Menu," "Entertainment," "Decor," "To-do list for the week before the party."

OK, I'm ready to start planning. Ask your first questions.

Why This Example Works: The Mechanism.

This prompt is effective because it applies the same psychologically and technically sound logic as HopeBot to an everyday task:

1. Lower cognitive load: Instead of the user trying to keep every aspect of the planning in their head, the prompt delegates that job to the LLM. The phased structure turns a chaotic process into an orderly sequence of steps.
2. Preventing irrelevant suggestions: The [INTERACTIVITY RULE] construction is the key. Without it, a request to "organize a birthday party" could get a cookie-cutter plan that doesn't suit the user at all. Clarifying questions ensure the final plan is based on real preferences, budget, and guest list.
3. Simulating an expert approach: The process set in the prompt mimics how a professional event planner works: first gather general requirements, then drill into the details, and only then draw up the plan. This not only improves the result but, as the study showed, builds the user's trust and satisfaction with the interaction.

Usefulness Score: 92