Every AI interaction starts with a prompt. These can from a simple question to a series of instructions and requirements, even including new information or data that must be passed to the AI to give the desired output. Whatever the complexity, it is essential that you use a well-crafted and carefully designed prompt to generate the required output, particularly when AI is being used within teaching activities or assessments.
This page provides guidance and links to resources that will help you gain confidence in writing prompts for AI.
When starting with an AI assistant, it is useful to understand some of the mechanics of getting a relevant response. It consists mostly of trial and error, of reframing the question to elicit the desired outcome, but this can be made more efficient through a new expertise of “Prompt Engineering”.
Prompting is not a magic trick and you do not need to understand any kind of technology or coding language, indeed there are numerous cheat sheets available on the internet or social media, all designed to make you a prompt expert. All current AI is based on machine learning over vast sets of data, so getting the required response requires practice and a bit of trial and error.
So how does prompting help? The AI assistant tries to understand the request you enter in order to formulate an appropriate response; hence it is useful to give it some context, or parameters. You can ask it to produce an outcome in a specific style, language, or format.
Prompts are cumulative so that, following a response, you can provide the AI with a further prompt that contains new information (i.e. more detail, asking for an amendment, setting additional boundaries) that helps to refine its next response. You can build on these responses, providing as many additional prompts as needed to achieve your desired output.
Prompt Categories: RTRI Structure
There are numerous ways to frame your prompt, what they all have in common is the need to be well communicated, and descriptive. If you frame your prompt in the following way, it should go some way to providing a relevant response.
The following categories are recognised across the sector as a strategic approach to engineering prompts, known as the Role, Task, Requirements, Instructions (RTRI) structure. However, these are provided as guidelines and you are encouraged to trial different approaches to suit your needs.
Ask the tool to act in a particular role. For example:
Act as a:
- Project Manager
- Website developer
- Job interviewer
Give it details of the task, what exactly are you asking the assistant to do. For example:
- A headline
- An article
- A summary page
- A Blog
- A Cover letter
- A Lesson plan
- A Web page
- A Product description
- Some interview questions
- A short quiz
Setting clear parameters for the AI. This can include both essential elements that need to be incorporated, including style and tone, but also set boundaries for what the AI should not do. For example:
- Use poetic language.
- Use a formal tone.
- Use 400 words or less.
- Use approved research papers.
- Use approved research papers.
- Use scientific sources.
- Use a specific coding language (HTML / JSON / C#)
- Be creative / formal / persuasive.
You can then choose the output style, telling the AI how it should act on the prompt and what it needs to do to drive towards a very specific outcome. For example:
Show as/must contain:
- A table
- A list
- A summary
- Gannt chart
- Bullet points
- Word cloud
- 3 questions with correct answers
Remember, this structure is intended as a starting point, and you should use the cumulative nature of Ai models to build and refine your prompts until you are satisfied that it produces a suitable output.Once an initial response has been received, you can continue to refine the output, by generating follow on questions. You can ask it to elaborate or focus on a particular aspect or even change the tone and style completely. It may even be useful to give the assistant an example of the output you would like to achieve.
If this is being used with students, they will often need to be able to replicate the same setup you have used. In these situations, you should keep a record of how you have achieved your desired outcome, though you may wish to simplify or make it more concise if you have used a lot of prompts. Bing Chat does not save your conversations between sessions, so you must remember to download a copy of your conversation if you wish to return to the same task at a later date.
For creating images, following a similar set of instructions will enable the assistant to generate the output you intend. The DallE-3 image generator within BingChat for Enterprise is a leading model and fairly accurate when the prompt is defined. You can save the image if it meets your requirements.
As with the text-based query, you need to be descriptive about the output. For example:
Set the style
- Pencil drawing
- Specific artist style e.g. Van Gogh
- Digital Image
- Pixar animation style
Be as descriptive as you can:
“In the style of Van Gogh create me a painting of a cockerpoo wearing a bowler hat and smoking a pipe”, with the most important elements of the desired output upfront.
Ask the Assistant
As you can see, there are no wrong ways to ask for your assistant to help formulate a document or image, but here are ways to support a more desired outcome.
The key thing to remember is that you can always just ask the assistant!
BingChat for Enterprise response to “How should I design a prompt to get the best outcome?”
Designing a prompt that elicits the best outcome is an art in itself. Here are some tips that can help you create a well-formed prompt:
- Be specific: The more specific your prompt is, the better the outcome will be. For example, instead of asking “Can you draw me a picture of a dog?”, try asking “Can you draw me a picture of a golden retriever playing fetch in the park?”.
- Provide context: Providing context can help the model understand what you’re looking for. For example, if you’re asking for a recipe, provide information about dietary restrictions or ingredients you have on hand.
- Avoid ambiguity: Avoid using ambiguous language or phrases that could be interpreted in multiple ways. For example, instead of saying “Can you write me a story about a man and his dog?”, try saying “Can you write me a story about a man who rescues a dog from the pound and they become best friends?”.
- Use complete sentences: Use complete sentences when asking for something. This helps the model understand what you’re looking for and what information is important.
- Be polite: Being polite and using proper grammar can help the model generate more accurate responses.
I hope these tips help! Let me know if there’s anything else I can do for you.
BingChat for Enterprise (2023) Using AI to design effective AI prompts, 24 October.
Bingchat for Enterprise response for “give me an example of a prompt when someone is acting in a specific role, for example as a project manager write me a product description for a new training course”.
Sure! Here’s an example of a prompt for a project manager to write a product description for a new training course:
“Please write a product description for a new training course that will help project managers develop their skills in managing complex projects. The course should cover topics such as project planning, risk management, team leadership, and communication. The target audience is experienced project managers who are looking to take their skills to the next level.”
I hope this helps! Let me know if there’s anything else I can do for you.
Using that prompt:
Here’s a product description for a new training course that will help project managers develop their skills in managing complex projects:
Course Title: Advanced Project Management
This course is designed to help experienced project managers take their skills to the next level. The course covers a range of topics that are essential for managing complex projects, including project planning, risk management, team leadership, and communication. The course is taught by experienced project managers who have worked on a variety of complex projects.
- Develop advanced skills in project planning and execution
- Learn how to manage risks and uncertainties in complex projects
- Develop effective team leadership skills
- Learn how to communicate effectively with stakeholders
- Understand the importance of project governance and compliance
This course is designed for experienced project managers who are looking to take their skills to the next level. The course is ideal for those who have worked on complex projects in the past and are looking to develop advanced skills in project management.
To enroll in this course, you should have at least 5 years of experience in project management. You should also have experience managing complex projects and be familiar with project management methodologies.
The course is delivered online and consists of 10 modules. Each module includes video lectures, case studies, and interactive exercises. You will also have access to a range of resources, including reading materials and online forums.
The course is designed to be completed over 10 weeks. You should expect to spend approximately 5 hours per week on coursework.
I hope this helps! Let me know if you have any other questions.
BingChat for Enterprise (2023) Using Ai to provide an effective prompt for a project manager wanting to write a product description for a new training course, 24 October.
BingChat for Enterprise: Things to Try
Using the Edge browser and the Edge Sidebar, whilst logged into your University account:
- Open a document in the browser and ask the assistant to “summarise this page”.
- Upload some specific text into the query box and ask it to rewrite this text in a formal / amusing / informative style.
- Choose the compose tab and use the different options for composing various communications.
Below are some additional resources that you may want to use to help you with engineering AI prompts.