Writing AI prompts that work: structure, examples and templates for business
A good prompt saves more time than any new AI model. Learn the five building blocks of a useful prompt, see how a weak prompt becomes a strong one, and copy templates that work in everyday office tasks.
The five building blocks of a good prompt
Not every prompt needs all five. The more the result matters, the more of them you should include.
- 1. Role. From whose perspective should the AI answer? For example a sales manager, a tax assistant or an editor.
- 2. Task. What exactly should come out? Start with a verb: write, summarise, compare, check.
- 3. Context. What does the AI need to know? Audience, occasion, background, your notes or the document in question.
- 4. Format. What should the result look like? Length, tone, bullet points or prose, a table with fixed columns.
- 5. Limits. What should not happen? No invented figures, no jargon, no more than 120 words.
Before and after: turning a weak prompt into a good one
Weak
“Write an email to a customer about the delivery delay.”
The AI has to guess: who is the customer, how long is the delay, how formal should it be? The result is generic text.
Good
“You are an account manager at a machine-building company. Write an email to a long-standing customer: the spare parts they ordered will arrive two weeks late because a supplier dropped out. Apologise briefly, give the new date and offer a partial delivery. Factual, friendly, no more than 120 words.”
Why it works
Role, task, context, format and limits are all there. The AI does not have to invent anything, and the text needs hardly any editing.
Eight prompt templates for everyday work
Replace the details in square brackets with your own.
- Reply to an email. “Reply to the following email on behalf of [name, role]. Answer every question, suggest [date or next step], no more than 150 words. Email: [text]”
- Summarise a meeting. “Summarise these minutes in five bullet points. Then list all tasks as a table with the columns task, owner, due date. Minutes: [text]”
- Draft a proposal. “Write proposal text for [service] aimed at [audience]. Name three benefits, avoid superlatives, half a page. Our notes: [bullet points]”
- Job advert. “Write a job advert for [position] in [location]. Tasks, requirements and what we offer as short lists each, gender-neutral, no clichés. Key facts: [bullet points]”
- Shorten a text. “Cut the following text to half its length. Keep the content and technical terms, remove filler words. Text: [text]”
- Structure data. “Turn this list into a table with the columns [column 1], [column 2], [column 3]. Mark missing information with a question mark instead of filling it in. List: [text]”
- Explain a term. “Explain [term] so that a customer without technical knowledge understands it. Three sentences and one example from [industry].”
- Review a draft. “Check this draft for clarity, contradictions and missing information. Give me a list of concrete improvements, do not rewrite the text. Draft: [text]”
Three mistakes companies often make
Confidential data in the wrong tool
Customer data, contracts or HR files only belong in an AI with a data processing agreement and processing in the EU. More on GDPR-compliant AI.
Using results without checking
AI models sound convincing even when a detail is wrong. Figures, names and deadlines are always checked by a person. This rule belongs in every AI policy.
Everyone writes their own prompts
Good prompts get lost when they only live in one user’s history. A shared collection works better, see a prompt library as a team standard.
Frequently asked questions about AI prompts
What is a prompt?
A prompt is the instruction you give an AI, the question or task in the input field. The more clearly role, task, context, format and limits are described, the more useful the answer.
Do the same prompts work in every AI model?
Broadly, yes. The five building blocks help with all major language models. Differences show in the length and style of the answers. For an overview of the models see AI models.
What is prompt engineering?
Prompt engineering means deliberately developing and testing prompts until a result is reliably good. For everyday office work, the five building blocks and a few proven templates are usually enough.
May employees use company data in prompts?
Only in tools the company has approved, with a data processing agreement and a clear rule on data processing. Which data is allowed is best set out in an AI policy.
Try the prompts, with company data hosted in Germany
Test the templates with novendix.AI: several AI models on one platform, hosting in Germany, data processing agreement included.
Your secure AI platform for the Mittelstand. Secure. Intelligent. Integrated. Custom database integration, personally supported.
novendix GmbH · Industriestraße 6 · 91126 Schwabach
Locations: Schwabach · Weißenburg · Nuremberg
A company of the L&S Lange & Schermer Group


