How to write AI prompts that give useful business results
Vague requests get generic answers. Learn the six parts of a useful business prompt, see before-and-after examples, copy ready-made templates for customer replies, meeting notes and job posts, and build a habit of checking the output before it goes anywhere.

Most disappointing AI results come from a vague request, not a weak tool. If you ask for "a marketing email," you get something generic. If you explain who it is for, what you want them to do and what your business sounds like, you get something you can actually use.
This guide shows small business owners how to write prompts that produce useful drafts, with templates you can copy. It is general information, not professional advice, and anything an AI writes should be read by a person before it is used.
The idea behind a good prompt
Anthropic's documentation describes the AI as something like a capable new employee who lacks context on your norms and workflows. That is a helpful picture. A new hire given a one-line instruction will guess. A new hire given a clear brief will do a better job. The Anthropic guidance on clear and direct prompts also offers a simple test: show your prompt to a colleague with little context on the task, and if they would be confused, the AI probably will be too.
The same guidance says to ask explicitly for what you want rather than hoping the AI infers it. "Can you suggest some changes?" may get suggestions, while "change this paragraph to be shorter" gets the change made.
The six parts of a useful business prompt
You do not need every part every time, but the more of these you include, the less you will have to rework.
1. Role and purpose
One sentence on who the AI should act as and why. "You are a customer service writer for a small landscaping company" focuses tone and vocabulary. OpenAI's prompt engineering guide similarly recommends describing the assistant's purpose, communication style and goals up front.
2. Context
Say who the reader is, what happened and what you know. Explaining why something matters helps too. Anthropic gives the example of telling the AI a response will be read aloud by a text-to-speech engine, so it avoids ellipses, rather than just banning ellipses.
3. The task
State the action in one clear sentence: summarize, rewrite, compare, draft, list, classify.
4. Format and length
Say what you want back: three bullet points, a 120-word email, a table with these columns. Both vendors' guides stress specifying output format and constraints.
5. Examples
A few samples of what good looks like are among the most reliable ways to control tone and structure. Anthropic suggests including a handful, making them relevant and varied, and keeping them clearly separate from your instructions.
6. Source material
Paste the text the AI should work from, and tell it to use only that. OpenAI's guide describes adding relevant reference text and marking where it begins and ends with simple delimiters so the AI does not confuse it with instructions.
Before and after
| Weak prompt | Stronger prompt | |---|---| | Write a follow-up email | Write a friendly 100-word follow-up email to a customer who requested a quote for a kitchen remodel last week and has not replied. Offer a 15-minute call this week. Tone: warm, no pressure. End with one question. | | Summarize this | Summarize the email thread below in five bullets: decisions made, open questions, deadlines, who owns each action, and anything that needs my reply today. | | Make a social post | Write three short social posts announcing our new Saturday hours. Audience: local families. Voice: cheerful and plain. Include the hours: Saturdays 9 to 3. No hashtags. |
Templates you can copy
Fill in the brackets and keep the structure.
Customer reply: "You are a support writer for [business]. Reply to the message below. Be [tone]. Answer only from the facts I give you. If something is missing, say what we need to find out rather than guessing. Facts: [paste]. Message: [paste]."
Meeting notes: "Turn the notes below into: a three-sentence summary, a list of decisions, and a table of action items with owner and due date. If an owner or date is missing, write 'unassigned'. Notes: [paste]."
Job post: "Draft a job post for a [role] at [business] in [location]. Include responsibilities, must-have and nice-to-have skills, and the pay range [range]. Keep it under 300 words and use plain language. Avoid requirements not tied to the work."
Work in steps, not one giant request
Complex jobs go better when broken up. Ask for an outline first, correct it, then ask for the full draft. Anthropic suggests numbered steps when order or completeness matters, and OpenAI's guide recommends having the model break complicated work into sub-tasks.
Iterate like an editor
- Run the prompt.
- Note what is wrong: too long, wrong tone, missing facts.
- Tell the AI exactly that, in plain words, and ask for a revision.
- When you get a result you like, save the final prompt as a template.
Checking the output
A good prompt improves the odds but does not guarantee accuracy. Build a short review habit.
- Check every fact, number, name and date against your own records.
- Ask for gaps. Add "list anything you were unsure about" to the prompt.
- Keep sensitive data out. Do not paste customer personal details or payment information into tools you have not vetted. See AI and customer data privacy.
- Test on a few real cases. OpenAI's guide recommends building tests so you can see how a prompt behaves before relying on it. For a small business, ten past emails run through your template is a fair test.
- Keep a human sign-off for anything customers or regulators will see.
Build a small prompt library
Prompts you reuse are worth saving. Keep a shared document with the name, purpose, the prompt text, an example of a good result and the date you last checked it. Review it quarterly. Our guide on practical and safe AI uses suggests which tasks to start with, and measuring time and money saved shows how to tell whether a template pays off.
Common mistakes
- One-line requests. Short prompts force the AI to guess.
- Hiding the real goal. Say what the output is for.
- No format instruction. You will get whatever shape the tool prefers.
- Pasting everything. More text is not always better. Include what is relevant and mark it clearly.
- Accepting the first answer. One round of feedback usually makes a large difference.
- Treating the result as final. The AI drafts; you decide.
Where WorkBOS fits
Good prompts are most useful when the AI can see the right information. In WorkBOS, your contacts, deals, projects, invoices and documents sit in one system, so an AI agent can work from your own records instead of text you paste in by hand. Agents propose actions and a person approves them. Browse the features, check pricing, read the docs or take a look at the Academy to learn more.
Frequently asked questions
Q: How long should a prompt be? A: As long as it needs to be to give context, the task and the format. A few sentences is common for everyday work. Longer prompts are fine when each part is useful.
Q: Do I need special wording or tricks? A: No. Plain, specific instructions work best. Explain who the reader is, what you want and what a good answer looks like.
Q: Should I give examples? A: Yes when tone or format matters. A few relevant, varied examples help the AI match your style, and you should keep them separate from the instructions.
Q: Why does the AI sometimes ignore part of my prompt? A: Long, mixed-up prompts can blur priorities. Split the job into steps, use bullets for requirements, and ask for the change you want explicitly.
Q: Can I trust the facts it gives me? A: Not without checking. Give it your source text, tell it to use only that, and verify numbers and names yourself.
Sources are linked in the text above: Anthropic documentation, OpenAI documentation.
Run it all from one place
Projects, clients, finance, people and AI employees in one system. Start free, no card needed.
Start free →