Two people ask an AI chatbot for the same thing: “Write an email about a network outage.” The first gets a generic, flat text that has to be rewritten from scratch; the second gets a text that is almost ready to send. The difference is usually not the tool but the instruction they wrote. That instruction is called a “prompt,” and writing one is a skill any employee can learn in a short time.
Why the prompt matters so much
Chatbots such as ChatGPT, Claude or Gemini know nothing about your company, your audience or your goal unless you tell them. Wherever information is missing, the model fills the gap with a generic guess. The clearer the context, the less guessing and the more accurate the result.
The parts of a good prompt
- Context: Who you are and what the situation is. For example, “I am the IT manager of a distribution company with 40 employees.”
- A specific task: Exactly what you want: write, summarize, compare, translate or list.
- Audience: Who the text is for: the CEO, non-technical staff or the technical team.
- Output format: Length, structure and tone. For example, “At most 100 words, formal, in three short paragraphs.”
- Constraints: What should not be there. For example, “Don’t use technical jargon” or “If you’re not sure, say you don’t know.”
- Example: If you want a particular style, give one of your own previous texts as a sample.
An example: from a weak prompt to a good one
Weak prompt:
“Write an email about a network outage.”
Better prompt:
“I am the IT manager of a 40-person company. Write a short, formal email to all staff announcing that on Thursday from 14:00 to 16:00, internet and email will be unavailable because of a firewall update. The readers are not technical. Keep it under 120 words, tell them to save their files before the outage, and to call extension 205 for urgent issues.”
The second version tells the model everything it needs to know: who, to whom, what, when, how long and in what tone.
Have a conversation, not a one-shot command
Don’t treat the first answer as final. Refine it with short messages: “Make it shorter,” “Make the tone friendlier” or “Remove the second paragraph.” For larger tasks, work in stages: ask for an outline first, approve it, then ask for the full text. If a prompt works well for a recurring task, save it in a shared file so colleagues can use it too.
Ask the model to raise its questions before answering: “If you need more information for this task, ask me first.” This is the simplest way to find the gaps in your prompt.
Don’t forget the limitations
A good prompt reduces the chance of errors but does not eliminate it. Language models can confidently invent a number, a source, a command name or a fact, so check every output, especially technical commands before running them on a server. Don’t put confidential information such as passwords, internal network addresses, customer data or contracts in a prompt, or replace them with made-up placeholders before sending. Also, the features and terms of these services change, and access to some of them from Iran may be limited.
Checklist
- You have stated your context and role.
- The task and audience are clear.
- You have set the length, format and tone of the output.
- You have written down constraints and things to avoid.
- There is no confidential information in the prompt.
- You have checked the output before using it.
Writing good prompts is a personal skill, but using these tools safely in a company also depends on infrastructure. ITPlus724 engineers can help secure your company’s network, manage user accounts and protect your data.