Skip to Main Content
Research Guides@Tufts

AI for Admins: Prompt Design

Using AI to Streamline Administrative Tasks

OpenAI o1

With its new reasoning-focused model, OpenAI is asking us to turn that conventional prompt design on its head. The reason: o1 does most of the heavy lifting on its own, so cluttering it with complicated prompts can lead to lower-quality outputs.

  1. Keep prompts concise; include only essential information.
  2. Add relevant details to queries for more meaningful responses.
  3. Assign a role to the chatbot (e.g., "You are an expert in...").
  4. Use triple quotes to delineate specific parts of the query.
  5. Clearly outline the steps needed to complete a task.
  6. Provide examples to clarify expectations.
  7. Specify the desired output length.
  8. Break complex tasks into simpler subtasks.
  9. Allow time for the model to process and respond.

ChatGPT3 & GPT4

The typical blueprint for prompt engineering: Make your request as detailed as possible. Ask the model to break things down step-by-step, so it doesn’t get confused. And instill confidence by emphasizing its expertise in the subject matter.

Six strategies for getting better results

  1. Write Clear Instructions
    Provide specific details to minimize confusion. If the model outputs too much or too little information, adjust your query to ask for precise responses.

    • Example: Instead of "Schedule a meeting," specify, "Schedule a 30-minute budget meeting next Tuesday at 10 AM."
  2. Provide Reference Text
    If you need factual answers, especially on complex topics, supply the model with reference materials to avoid inaccuracies.

    • Example: Attach the relevant budget report when asking for a summary of the fiscal year.
  3. Split Complex Tasks into Simpler Steps
    Break down large or multi-step tasks into smaller ones to reduce errors.

    • Example: "First, gather last month’s financial reports. Then, prepare a comparison with the current month."
  4. Give the Model Time to “Think”
    For tasks that require reasoning or multi-step processes, request a step-by-step approach.

    • Example: "Create a plan for this year’s orientation, starting with venue booking, followed by catering, then invitations."
  5. Use External Tools
    Enhance model performance by integrating other tools for tasks like calculations or pulling from databases.

    • Example: "Use the system to retrieve the student enrollment data from the central database."
  6. Test Changes Systematically
    When implementing new processes or adjustments, make sure to evaluate the results carefully.

    • Example: After adjusting workflows for document approvals, measure time savings over the next month.