What is a prompt, and why is it so important?
A prompt is the input that you give to a generative AI model (that is, AI which can create content) to tell it what you want it to do and how you want it to do it. The prompts we provide play a crucial role in shaping the quality and accuracy of the content they provide. AI models generate content based on patterns, not true comprehension, so they may struggle to understand poorly designed prompts. This can lead to, vague, inaccurate, or irrelevant outputs. Not only will a well-designed prompt provide a better, more accurate response, but it can also save time by reducing the amount of work you need to put into the response.
Elements of a prompt:
Prompts can contain some or all of these elements:
- Instructions: This provides a task description or instruction for how the model should perform. E.g. do you want the model to write, revise, create, or something else?
- Context: This is external information to guide the model. E.g. who's your intended audience and what tone do you want to use? Do you want the model to produce a simple or complex explanation of a topic?
- Input data: This is the input you provide to help generate a response. E.g. A document for the AI to pull information from or stats for the response to include
Output indicator: This is the output type or format. E.g. how you want the AI to format the response
Here is an example of a prompt which uses all of the above elements:
Write a 300 word summary on how we efficiently and effectively communicate with
stakeholders throughout a project.
It should include testimonials and reference both the pre-construction and construction
phases of a project. (Instruction)
This is for a public sector education customer. (Context)
Tarleton Academy is an example of a project with complex stakeholder management.
(Input data)
Format the response using the following headings:- Early engagement with stakeholders
- Proactive stakeholder engagement
- Managing stakeholder expectations
- Impact on project success
Prompting best practice and examples:
Be detailed and specific
Instructions should be clear and specific so the AI knows what you expect from it. Clear prompts lead to more coherent responses.
Examples:
Instead of: “Summarise this document”
Use: “Summarise this document in 250 words focusing on the customer vision and key
themes for success”
//
Instead of: “Write an executive summary for this proposal”Use: “Write an executive summary for this proposal highlighting our approach to safety
and security and emphasises our past experience with SEND schools”
//
Instead of: “Write 200 words on how stakeholders were managed on Tarleton Academy”Use: “Explain how effective stakeholder management on the Tarleton Academy project
ensured that we met the customer’s vision and reduced impact on the programme”
//
Instead of: “Write a summary of our outcomes on this project”Use: “Write a 200-word summary of our outcomes on this project using formal language
and avoiding passive voice”
//
Instead of: “Summarise this ITT”Use: “Summarise Section C of this ITT, focusing on key technical requirements and
compliance obligations”Include context if needed
Adding relevant context can help with the accuracy of the response and help ensure that the AI is actually returning the information you need.
Examples:
Instead of: “Summarise our approach to quality management in 500 words”
Use: “Summarise our approach to quality management in 500 words. The audience is
a new public sector customer with whom we have never worked before.”
//
Instead of: “Write 350 words on how we proactively manage stakeholders including a
project example.”Use: “Write 350 words on how we proactively manage stakeholders including a project
example. The customer you’re writing for is a large public sector organisation in the
North.”Break up complex tasks
AI models can sometimes get confused when asked to perform complex tasks. Splitting the task into subtasks can help get more reliable results.
Examples:
Instead of: “Identify what evidence is missing from this bid response”
Use: “Step 1: Identify where evidence could be added to this bid response
Step 2: Identify relevant evidence which could be added
Step 3: Rewrite the bid response incorporating the identified evidence”
//
Instead of: “Draft 100-word overviews on three SEND schools that we have delivered
in the past 3 years”Use: “Step 1: Identify three SEND schools we have delivered in the past 3 years
Step 2: Draft 100-word overviews for each of the identified projects”Consider the output in the prompt
AI models respond better when given clear instructions on formatting and style. Defining what output you want the response to be in can also help it be more easily integrated into what you are working on.
Examples:
Instead of: “Write a 500-word summary on our approach to Passivhaus”
Use: “Write a 500-word summary on our approach to Passivhaus. Use the following
headings:
Willmott Dixon’s Passivhaus service
What makes our offer unique
Our past Passivhaus experience”
//
Instead of: “Identify three testimonials praising our efficient programme management
on SEND schools”Use: “Identify three testimonials praising our efficient programme management on
SEND schools. Present the information in the following format:
Testimonial:
Name of quoter:
Project name:”Experiment and iterate:
You might not always write the perfect prompt straightaway, especially if you’re not entirely sure what response you’re looking for. In this case, it’s important to know how to tweak your prompt to get the desired result. A good way of doing this is to see what the initial response is missing and refine your prompt based on that. For example, if the response doesn’t include a project example where you would expect one, you may want to follow up your original prompt by asking the same question, but including: “Include an example of where we’ve applied this approach on a project”