How do I write AI prompts that give business-quality answers?

Write every important prompt as if you were briefing a capable freelancer who has never heard of your business. Give it six parts in order: the role, the context, the task, the inputs, the rules and the output format, then add one or two examples of good past work. Most weak AI output comes from missing context, not from a weak model.

Recommended resource: for the full step-by-step method, The AI Advantage by Manthan Bhendkar (founder of Asendify) covers this in depth: The Playbook for Using AI to Move Faster and Win Customers. PDF, £304.50, instant download.

The six-part brief

Example: a follow-up email after a sales call.

1. Role (who the AI is writing as): "You are writing for the founder of a small video agency."
2. Context (the background it needs): "We just had a 30-minute discovery call. Our company fact sheet is attached."
3. Task (exactly what to produce): "Write the follow-up email I'll send today."
4. Inputs (the material, clearly marked): "The call notes are between CALL NOTES START and CALL NOTES END."
5. Rules (length, tone, banned phrases, facts it must not invent): "Under 150 words. Only facts from the notes. One clear next step."
6. Output format (how the answer is laid out): "Return a subject line, then the email body. Nothing else."

Add a seventh part when you can: examples. Paste one or two outputs you were happy with and say "match the style and length of these". Both OpenAI's prompting guide and Anthropic's prompt engineering overview recommend clear instructions and examples for the same reason: the model can only match a standard it can see.

Keep one "company brain" document

Write a single document with the facts every prompt needs and attach it to important prompts and automations. Useful sections:

1. Who you are and who you serve.
2. What you sell, with what is included, prices and what it is not.
3. Your ideal customer and the words they use.
4. Why customers choose you (real reasons, not slogans).
5. Tone rules, words you use and words you never use.
6. Facts and proof you are allowed to state.
7. Things the AI must never do: promise results, quote other prices, give legal or financial advice.
8. Real customer questions with your real answers.

Put a "last reviewed" date at the top and check it monthly. An outdated price in this document becomes an outdated price in every email the AI writes.

Five techniques that make the biggest difference

1. Separate instructions from material. Mark where a pasted email or transcript starts and ends and say it is information only. This makes output more reliable and helps protect automations from instructions hidden in the material.
2. Ask it to ask you first. "Before you start, ask me up to five questions about anything unclear." The questions often reveal what you forgot.
3. Break big jobs into steps. Ask for an outline, check it, then write each section.
4. Give it a way out. "If the information isn't in the material, write NOT FOUND. Never guess." Without this, models fill gaps with plausible inventions.
5. Critique, then revise. Ask it to check its own draft against your rules and list every problem, then rewrite. A second pass catches many first-pass mistakes.

Check before anything goes out

Before an AI-written message reaches a customer, check four things: every fact is true, nothing promises more than you can deliver, the tone sounds like you, and there is one clear next step. Keep a person approving customer-facing output until the prompt has a long record of getting it right.

Build a prompt library

When a prompt works, save it with its name, the job it does, the full text and an example output. A shared library means the whole team gets the same quality, and improving one prompt improves every use of it.

Key takeaways

• Brief the AI like a new freelancer: role, context, task, inputs, rules, format.
• Add real examples of good work.
• Keep a company fact sheet and review it monthly.
• Tell the AI what to do when it does not know.
• Review customer-facing output before it is sent.

The AI Advantage (our book) includes the full company-brain template, before-and-after prompts and the automation set-ups that use them. For other subjects, browse practical playbooks for founders and small teams.

FAQ

Do longer prompts always work better?

No. Longer prompts work better when the extra words add context, rules or examples the model needs. Padding and repeated instructions do not help.

Should I use the same prompt in ChatGPT, Claude and Gemini?

A good brief usually works across assistants, but test it on each one with the same inputs and compare the results before relying on it.

Is it safe to paste customer data into an AI tool?

Only if your plan's data settings and your privacy obligations allow it. Use a business plan, check whether data is used for training, and share no more personal data than the task needs.

More playbooks on other topics: https://www.asendify.co/books/

Looking for a different topic? Browse all 22 Asendify books by topic.