Prompts that deliver results

Martin Hrnko

30.11.2024

Effective prompts are key to getting better and more relevant answers from large language models, such as OpenAI ChatGPT or Copilot in Microsoft 365.

The following general principles will help you achieve better results.

Clarity and Comprehensibility → The prompt must be clear and unambiguous so that the model understands as accurately as possible what you expect from it.
Context → Providing enough context helps the model better understand your needs and provide more relevant answers.
Specificity → Be specific and detailed in what you ask the model to minimize the risk of incorrect or general answers.
Iteration → Don’t be afraid to modify the prompts and test different variations to find the best way to communicate with the model.

In addition to the above general principles for creating a prompt, there are also a number of different approaches, or rather strategies, on how to formulate a query against a language model. In the following text, we will describe and discuss those that I have encountered, but I do not have the ambition to describe all of them, I probably would not be able to do that. However, I believe that it will inspire you to establish your own prompting strategy adapted to the specific tasks you are solving.

General prompting strategies

In the following text, I present well-known prompting strategies with examples of their use. In the introduction, I always explain the abbreviation and follow with an example from work and personal use. If you have the time and desire, try the examples and of course do not be afraid to modify them to better suit your specific needs. Enter the prompt examples in separate communication threads so that the language model is not influenced by the response to the previous prompt. The strategies are listed in alphabetical order, there is no other logic in their order. Slovak translations of the prompt components are adapted to the purpose of the prompting strategy, do not consider them as literal translations, the goal is to enable a better understanding of the individual approaches.

APE (Action, Purpose, Expectation)

The meaning of prompt items (APE → task, purpose, expectation).

Action→ Define the task or activity that you expect from the language model.
Purpose→ Describe the goal or purpose of the task you are entering.
Expectation→ State the expected result, what you want to achieve by entering.

Example of prompt (APE) in business.

Prepare an hour-long training for the Microsoft 365 package.
The goal is to inform team members about the latest features in Microsoft 365 so that they can use the most modern and effective tools at work.
The desired result is that my team members have a clear idea of the possibilities that Microsoft 365 offers them and that they are able to apply them to common tasks as soon as possible.

Example of prompt (APE) in personal life.

Plan my daily 2-hour training to prepare for a marathon.
The goal is to improve my endurance and prepare for the upcoming marathon.
The expected result is to improve my physical condition and prepare consistently for the marathon.

BAB (Before, After, Bridge)

The meaning of the prompt items (BAB → before, after, bridging).

Before→ Describe the problem or situation before the action for which you expect a suggestion from the language model.
After→ State the result or state after the activity that you want the language model to achieve.
Bridge→ Describe what you think should lead to the change in the state before and after the activity that you expect the language model to achieve.

Example of a prompt (BAB) in business.

Many employees in our company struggle with organizing their emails, which leads to missing important tasks and chaos in the workplace.
I would like all employees to be able to process incoming mail and not be late with responses, nor to be late for meetings.
Design a system to solve the situation.

Example of a prompt (BAB) in personal life.

I struggle with organizing household chores and completing them on time, I don't enjoy them and I don't have time for them. This mainly concerns cleaning and ironing.
I would like to have more time for my children.
Suggest what I should do about it.

CARE (Context, Action, Result, Example)

The meaning of the prompt items (CARE → context, task, result, example).

Context→ Explain the situation.
Action→ Assign a task to the language model.
Result→ Describe what you expect the result to be.
Example→ Give an example of the expected response.

Example of a prompt (CARE) in business.

Many employees waste time searching for documents on SharePoint, which, however, leads to their reduced productivity.
Design a centralized document management system in SharePoint that would include organizing files into logical folders and setting access permissions based on job position.
After implementing the new system, all employees will be able to quickly and easily find the necessary documents.
For example, while working on the reconstruction of a local cultural center, the electrician did not get lost in the complex project documentation, but immediately found a map of electrical wiring in his folder and was able to immediately search for a specific document by name and time it was created.

Example of a prompt (CARE) in personal life.

I have problems managing my home budget and tracking expenses.
Create an Excel spreadsheet for me to track income and expenses with categories for different types of expenses.
After implementing this spreadsheet, I will be able to better control my finances and track expenses.
For example, a monthly expense report will help me identify areas where I can save money.

COAST (Context, Objective, Action, Scenario, Timing)

The meaning of prompt items (COAST → context, goal, task, scenario, time).

Context→ Describe the situation for which you are looking for a solution.
Objective→ Formulate the goal you want to achieve.
Action→ Describe the task you want the model to perform.
Scenario→ Describe the situation you need to solve.
Timing→ State the timeline, or time perspective, that the model should consider when designing an answer.

Example of prompt (COAST) in business.

Our team is facing problems with late deliveries of a software project.
We need to improve efficiency and meet deadlines.
Create a new project schedule with clearly defined tasks and deadlines.
The new schedule should allow for better tracking of progress and meeting deadlines.
We need to complete the activities within the project by the end of next week.

Example of prompt (COAST) in personal life.

I am having trouble organizing the responsibilities of organizing a birthday party.
But I would like to organize a beautiful and enjoyable celebration for my son's friends and classmates.
Suggest a place, program, invitation for the celebration.
I need a comprehensive plan from which I can easily and quickly select and carry out specific tasks.
We need to hold the celebration next Saturday at 2:00 PM in Bratislava.

ERA (Expectation, Role, Action)

The meaning of the prompt items (ERA → expectation, role, task).

Expectation→ Describe the expectation you want to achieve.
Role→ Describe the role in which the language model should act when proposing an answer.
Action→ Describe what you need to do.

Example of a prompt (ERA) in business.

I need to prepare a detailed plan to increase the sales of our fruit products.
Act as an experienced international business consultant.
Design an expansion strategy, target market analysis, recommended marketing tactics, and logistical solutions for effective distribution.

Example of a prompt (ERA) in personal life.

I want to improve my physical health and fitness.
Act as a personal trainer and nutrition consultant.
Help me create an exercise plan and diet tailored to my needs.

RACE (Role, Action, Context, Expectation)

The meaning of the prompt items (RACE → role, task, context, expectation).

Role→ State the role in which the language model is acting.
Action→ Describe the action you want to take.
Context→ Describe the situation you are in.
Expectation→ Describe the expected result you want the language model to achieve.

Example of a prompt (RACE) in business.

As an experienced HR manager from a multinational corporation,
Design a strategy for recruiting new talent,
For a fast-growing IT company facing a shortage of qualified workers.
Provide me with a detailed plan with recommended recruitment channels, selection procedures and retention strategies for each type of job.

Example of a prompt (RACE) in my personal life.

As a career advisor,
Help me plan my career path.
I am a recent college graduate in the field of food engineering,
Provide me with a step-by-step plan on how to proceed in managing my future career.

RISE (Role, Input, Steps, Expectation)

The meaning of the prompt items (RISE → role, input, steps, expectation).

Role→ Explain what role the language model is supposed to play.
Input→ State the information or sources of information that the language model is supposed to provide in its answers.
Steps→ What steps the language model should take to provide the answer you need.
Expectation→ State what form you expect the result from the language model to take.

Example of a prompt (RISE) in business.

Working as a consultant for sustainability and environmental policy in the corporate sector.
My company is a medium-sized manufacturer of packaging materials that wants to reduce its environmental impact and implement sustainable practices in production.
Using supplementary questions, analyze current production processes and identify areas with the greatest environmental impact, propose possible solutions and technologies to reduce waste and emissions, and assess the economic efficiency of the proposed measures and their impact on the business.
The goal is to provide a detailed sustainability plan containing specific steps, implementation timeline, and expected results, including potential savings and improved company image.

Example of prompt (RISE) in personal life.

Working as a personal trainer and nutritionist.
I am an individual, a 40-year-old manager who feels overwhelmed by work and has difficulty maintaining a work-life balance. I am starting to feel burned out.
Using additional questions, identify the main stressors in my work and personal life, suggest techniques and strategies for managing stress and improving time management, and provide recommendations to support mental and physical well-being.
Create an individual stress management plan that includes practical steps, relaxation techniques.

ROSES (Role, Objective, Scenario, Expected Solution, Steps)

Meaning of prompt items (ROSES → role, objective, scenario, expected solution, steps).

Role→ State the role that the language model should play.
Objective→ State exactly what you expect from the language model.
Scenario→ Describe the situation you are in and within which the language model should search for an answer.
Expected Solution→ Describe the ideal result you want the language model to achieve.
Steps→ State the steps that the language model should take or consider when preparing an answer.

Example of a prompt (ROSES) in business.

Work as a strategic consultant for sustainable development.
Help our company create a strategy for reducing its carbon footprint and increasing energy efficiency.
Our manufacturing company is facing increasing pressure to reduce its environmental impact and is looking for effective ways to do this.
I need you to propose a detailed plan of action for implementing sustainable practices, including recommendations for technologies and processes that reduce emissions.
Ask follow-up questions to take the following steps: analyze the company's current energy use and emissions, identify areas with the greatest potential for improvement, and propose specific measures.

Example of a prompt (ROSES) in your personal life.

You are a personal language tutor.
Help me create a strategy to improve my Chinese skills.
I want to be able to speak Chinese fluently so that I can communicate better in my personal life.
I need a specific study plan including recommendations for resources, learning methods, and practical exercises to improve my speaking, listening, reading, and writing.
Ask me follow-up questions that you need to know.

RTF (Role, Task, Format)

The meaning of the prompt items (RTF → role, task, format):

Role→ State the role in which the language model is to perform.
Task→ State what the language model is to do for you.
Format→ State the form in which you expect the result to be.

Example of a prompt (RTF) in business.

Act as an experienced senior HR professional.
Prepare an advertisement for a job position as a junior programmer with a focus on JavaScript for our startup company where we create games.
Provide me with the complete text of the advertisement, including requirements, job description and benefits.

Example of a prompt (RTF) in personal life.

Act as a chef specializing in Slovak cuisine.
Suggest me a recipe for a romantic dinner for two using seasonal ingredients.
Provide me with a shopping list and step-by-step instructions for preparing dinner.

TAG (Task, Action, Goal)

The meaning of prompt items (TAG → goal, task, result).

Task→ Describe the task, what exactly you expect from the language model.
Action→ State what the language model should do for you.
Goal→ Define the result that you want to achieve using the result from the language model.

Example of a prompt (TAG) in business.

I need to create an attractive description for our new toilet paper on our website.
Describe an engaging and informative text that highlights the key benefits and features of the product.
The goal is to increase potential customers' interest in our product.

Example of a prompt (TAG) in personal life.

I need to write a poem on the topic of love and loss for a literary competition.
Compose an emotional poem that expresses the complex emotions associated with love and loss.
I would like to impress the jury with original content and style and increase my chances of winning the competition.

TRACE (Task, Request, Action, Context, Example)

The meaning of prompt items (TRACE → problem, request, steps, context, example).

Task→ The problem you want to solve using the language model.
Request→ The task you want the language model to do.
Action→ What steps should the language model do.
Context→ What is the additional information about the problem you are solving.
Example→ An example of the expected answer.

An example of a prompt (TRACE) in business.

Our company needs to raise awareness of a new eco-friendly product on the market.
Create a creative slogan and design an advertising campaign for this product.
Analyze the product features, identify the target audience, and create a slogan along with a short description of the campaign.
The product is a biodegradable cleaning product designed for households that care about sustainability and an eco-friendly lifestyle.
I expect a slogan along the lines of "Cleanliness without compromise for our planet" and a campaign design focused on social media and eco-blogs.

Example of a prompt (TRACE) in personal life.

I want to plan a vacation abroad, but I can't decide where to go.
Recommend three vacation destinations according to my preferences.
Consider my interests and requirements, compare the suggested destinations, and briefly describe why they would be suitable for me.
I don't like history, but I like good music and good cuisine. I'm planning a vacation for two weeks in the summer months and prefer places where you can relax pleasantly.
I don't expect recommendations like Greece for its ancient monuments and beautiful beaches, Italy for its rich history and excellent cuisine, or Spain for its combination of culture, beaches, and gastronomy.

Final Summary and Recommendations

Finally, let’s summarize some generally valid rules for creating prompts.

Be specific→ Clearly define what you expect from the model and provide enough details.
Provide context→ Describe the situation or problem so that the model better understands your needs.
Iterate and test→ Don’t be afraid to modify the prompts and try different variations to find the best results.
Use clear and understandable language→ Make sure that your prompt is understandable and unambiguous.

For those of you who have read this Newsletter to the end, the most important advice for free at the end.

Review→ Approach all outputs of the language model critically. Verify information and advice, never forward them without evaluation and do not present them as the results of your work without verifying the results. This will avoid surprises.
Encourage→ encourage the language model to ask additional questions, at the end of each prompt you can write the sentence "Ask additional questions so that you can prepare the best possible answer in this area".
Experiment→ experiment with different approaches to creating prompts and assigning tasks, this will achieve more creative results. Tell the model not only what your output should look like, but also what it should not look like. Do not tell the model what it should do and what the result should look like, but also what it should not look like. Do not tell it where you want to go on vacation, but where you do not want to go and what you do not want to do there.
Check→ if you think that language models cannot yet help you with the design of procedures, answers, emails, try them at least to check typos or the correctness of the formulations of your own texts. You will be surprised in what areas language models can already advise you today.

We hope these tips and strategies will help you create effective prompts and improve your work with large language models. Consider the above procedures as inspiration for your own creation, don't be afraid to experiment 💪.

If you have any questions or need further assistance, do not hesitate to contact us at MH THREE TECHNOLOGIES.

We are here to help you turn complex challenges into clear solutions and your visions into reality!

If you are interested in the topic of Enterprise Architecture in your company, arrange a consultation with us via email martin.hrnko@mh3.sk, or directly here: https://calendly.com/martinhrnko/45-minute-meeting.