Despite the advent of messengers, video calling capabilities, and voice messages, emails remain a part of our lives. A cover letter accompanying a resume, explanatory notes, invitations to relatives for a wedding, a business proposal for manufacturing or marketing—in fact, we write a lot. We spend time and effort on this. But why should we when there’s a chat bot in the GPT model that can generate any text or article itself. And if you know how to formulate commands for ChatGPT for emails, a boring routine or a complex task turns into a 5-minute job.
Of course, you could simply send a request to the bot in the style of “Write an invitation to the restaurant for my friend”. But the result will not be as you expect because the “electronic brain” is not like a search engine. It requires details, hints. The more questions the neural network receives, the more accurate the answer.
Such a hint is a prompt—an instruction or code for writing an article, letter, or another text. Creating prompts is not complicated (at a basic level), although there are many tricks that turn GPT into a multi-functional tool of the “how did I ever manage without it” variety.
Do not use requests as if you were talking to a search engine. You need to provide the bot with as much data as possible, but in a concise form. To simplify the task, use this structure:
Now let’s break down how to write a prompt. Example—a cover letter to a dentist’s resume.
Prompt: You are playing the role of a dentist. I will give you my resume. Study it. Write a cover letter to the resume, no more than 300 words. List such achievements: achievement 1, achievement 2, achievement 3. Use a formal, but friendly style.
For GPT to write what you need, address it as you would a person. Yes, the request should be specific, but with details. The more information the bot receives, the more accurate and complete its response will be.
Give the bot time to think about the task. If the response isn’t what you wanted, send the request again, with the same input. Or adjust the task for a better quality response. When communicating, the model remembers all discussions, but only within the current dialogue. If you start a new one, the answers will be different.