
Prompt Engineering: key considerations for effective use of ChatGPT
When using ChatGPT, it's crucial to approach prompt engineering thoughtfully, meaning how you formulate your request. ChatGPT can retrieve almost any information, and if you're not satisfied with the result, the issue often lies in how you've phrased your query. We’ll adhere to the principle that the question determines the answer.
What is a prompt? A prompt is the input or task provided to an AI model to generate text or perform a specific task. It serves as the starting point from which the model begins to generate text or solve a problem.
In the context of language models like GPT, a prompt can be a statement or question that guides the model on a specific topic or task. For example, a prompt might be "Write an article on the application of neural networks in medicine" or the question "What are the advantages and disadvantages of using robots in manufacturing?"
It’s important to understand that the quality and content of the prompt play a key role in the outcome received from the model. The clearer and more precisely the prompt is formulated, the more relevant and informative the model's output will be.
Why should non-technical specialists study prompt engineering?
The short answer: to effectively communicate tasks to the model and receive the most accurate desired responses.
Long answer:
For marketers:
- Understanding consumers: Effective prompt engineering allows for deeper analysis of customer behavior and preferences, facilitating targeted advertising campaigns.
- Content optimization: The ability to craft precise queries that extract useful data aids in creating personalized and engaging content for the audience.
For product managers:
- Market research: Prompt engineering helps conduct effective market analysis, identifying trends and preferences that can influence new product development.
- Prototyping: Rapid iteration of product ideas using language models allows for concept testing before launching the process.
For customer service specialists:
- Automating responses: Well-formulated prompts can be used to create automated yet personalized responses to frequently asked questions, reducing operator workload.
- Analyzing inquiries: Deep analysis of customer inquiries can identify common issues and wishes, improving service quality.
For communications specialists:
- SEO and content strategy: Using prompt engineering to optimize content for SEO helps enhance product visibility in search engines. It's essential to describe precisely what you want.
- Social media management: Effective engagement on social media through automated but deeply personalized messages strengthens connections with the audience.
For sales specialists:
- Personalizing communication: Prompt engineering enables the creation of highly personalized communications with potential clients, increasing the likelihood of successful deals.
- Automating routine inquiries: Using language models to automate responses to standard customer questions frees up time for more critical tasks, improving productivity.
For business development specialists:
- Assessing market opportunities: Prompt engineering can analyze large volumes of market data, helping to identify and evaluate market opportunities.
- Strategic planning: Language models can simulate various business scenarios, aiding in making informed strategic decisions.
- Accelerating innovation: Using prompt engineering for quick validation and adaptation of new ideas and approaches fosters innovation and the proactive development of the company.
To formulate prompts effectively at an advanced level, here are some recommendations (from general to more specific and in-depth):
- Be Specific: Understand the outcome you want. Clarifying your request will yield a more accurate and informative response. Instead of saying, "Tell me about Spain," ask, "Provide information on Spain's cultural history and its influence on contemporary European art."
- Describe Context and Circumstances: Explain why and for what purpose you need the model to solve this task. Specify, for instance, the type of company involved and any prerequisites, like if you’re testing hypotheses and need quick results.
- Ask Sequential Questions: Break the task into smaller subtasks for diverse and unique ideas. After a basic response, refine your request with follow-ups like, “Tell me more about…” or “Change the structure of the answer; I need it in a table format.”
- Use Personas and Instructions: For example, “Respond as a senior data analyst. Formulate a task for your colleague to forecast leads for the next six months based on marketing department data from the last six months.” Or “Imagine you are a neuromarketing expert…”
- Chain of Thoughts Method: Ask ChatGPT to explain its reasoning. “Step-by-step, describe your thought process.” This allows you to better understand the topic or verify the model’s response direction.
- Request Sources of Information (for GPT-4): Depending on the task, it may be necessary to find the latest information and verify sources. Mention this in the prompt, e.g., “...Answer based on data from 2023-2024. Use at least six different sources and provide links…”
- Include Emotions: Surprisingly, AI language models can be quite empathetic. Research shows that incorporating emotional elements in prompts can improve response quality and accuracy by up to 10%. Use phrases that emphasize task importance, like “I need this data for an important meeting in a few hours” or “This is VERY important!”
Verification: “How would you rate your answer on a scale from 1 to 10?” or “Is this your final answer? Maybe it’s worth double-checking?”
Motivation: “I know this isn’t easy, but challenges are opportunities for growth. Every obstacle overcome brings you closer to success.”
- Change the Interaction Model (Cognitive Verifier Template): If you have a high-level request or are unsure how to approach a task, ask the model to pose questions to you instead of answering your queries. This method breaks complex questions into smaller sub-questions. ChatGPT will start asking clarifying questions and update its understanding with each response.
- How to Start This Dialogue: For example, “I can’t calculate the economics of a new product. Imagine you’re a unit economics expert. Let’s simulate a dialogue where you ask me clarifying questions, and I answer. This will help me structure what I already know and identify what I need to learn.”
Another example: “When I ask my question, come up with ten additional questions to help you give a more precise answer. Ask these questions sequentially. If needed, increase the number of questions. Once I answer all, combine my responses to provide a final answer to my original question.”
When using ChatGPT for content generation, you can train the model to respond in your writing style. Upload several texts you’ve written and use this prompt: “Please study my writing style based on the following materials to use it for generating your responses.”
Once you train the model with a few examples, you can start the task and describe it in the prompt.
An almost-perfect formula for a prompt

Task: It is recommended to always start a prompt with an action verb like “write/analyze/provide/create/generate/come up with,” etc.
Context: The more context you provide, the better. However, as a person, you can discern which context is most important for the task. Key elements to include in the context are task prerequisites, goals, environment, and industry.
Examples: If possible, provide examples of the responses you want to receive.
Persona: In what field do you want the GPT model to act as an expert? How specifically should it assist you? From the perspective of which profession? (See above for more details.)
Format: What structure do you expect from the AI’s response? Bullet points/comparison table/text in five paragraphs, each no more than 200 characters/poem/script, etc.
Tone: Formal/friendly/neutral/slang/“as if writing a social media post,” etc.
Examples of Use Cases for ChatGPT-4 (without API)
- Jobs-to-be-Done Research: You can use AI to identify the “jobs” or pain points of your target audience. Ask GPT to articulate these jobs and formulate interview questions. For example: “Act as a product manager for a grocery delivery app serving the Toronto region. I want to conduct consumer research to understand what jobs they need to accomplish. Which personas should I focus on, and what questions should I ask to uncover the tasks they address using the Jobs-to-be-Done framework? Base your questions on the book 'Mom Test.'”
- Customer Development (Interviews): With effective prompt engineering and a clear understanding of your target audience, GPT-4 can understand their needs with about 80% accuracy. First, describe your audience segments—age, gender, life situation, country, income, etc.—using the persona method outlined earlier. Instruct the model to behave like this target audience and provide detailed responses to your questions. You can then ask the same questions as in real interviews.
- Code Generation from Images: This can be useful for creating MVPs. You can find many business applications for this scenario. Here’s how it works: upload a hand-drawn page or screenshot to GPT-4 and ask it to generate a script based on that image, including all details. The model will provide you with HTML, JavaScript, and CSS scripts (all three if you upload a clear image with variations in font size, image placement, etc.). You can then copy these scripts into a text file or Visual Studio Code, save the files in the correct formats in one folder, and link the CSS and JS files in the HTML. You can ask GPT-4 how to do this. Finally, save the file and open it in a browser! This entire process takes, without exaggeration, about five minutes.
- Writing UX Interview Scripts from Screenshots: You can upload a screenshot of the product you want to conduct a UX interview on and add a prompt for GPT to generate relevant interview questions.
- Understanding Developer Output for Non-Technical Specialists: For project/product managers and other leaders, you can upload a piece of code written by a developer and ask the model to explain what the developer accomplished—essentially, what tasks were completed.
- Task Definition and Time Estimation: This can save time when planning your workweek. You can explain to ChatGPT what tasks you need to accomplish, mention when you feel most comfortable doing these tasks (if you have such preferences), and ask for an estimate of the time required. Then, request a weekly schedule that considers these tasks, your available time, calls, and other relevant factors.