Prompt Engineering for Business: How to Get Useful Results from ChatGPT and Claude

"I tried ChatGPT and the answer was generic." I hear this every week. Almost every time, the prompt was one sentence long and the model did exactly what it was told: produce something generic.

Prompt engineering sounds technical, but for business use it is mostly about giving the model the same context you would give a new assistant. Once you do that, the quality jumps dramatically. Here is the structure I use in daily client work and inside automated workflows.

The five-part prompt structure

Every reliable business prompt has five parts. You do not need all five every time, but the more you include, the more consistent the output.

1. Role

Tell the model who it is for this task.

"You are an experienced customer support manager for a small logistics company in Nairobi."

This sets the tone, vocabulary and level of detail.

2. Context

Give the background a human would need.

"We deliver parcels within Kenya. Customers usually contact us about delivery delays, wrong addresses and refunds. Our tone is warm, direct and never defensive."

3. Task

Say exactly what you want, with a verb.

"Draft a reply to the customer email below."

4. Constraints and format

Length, structure, language, things to avoid.

"Keep it under 120 words. Use plain English. Do not promise a specific delivery time. End with a clear next step."

5. Input

Paste the material the model should work with, clearly separated.

"Customer email: ..."

Put these together and you get output you can send with minor edits, instead of something you rewrite from scratch.

Templates you can copy

Email reply

You are [role] at [company]. Our tone is [three adjectives].
Draft a reply to the message below.
Constraints: under [n] words, plain English, [things to avoid], end with a clear next step.
Message:
[paste]

Research summary

You are a research assistant. Summarise the text below for a busy business owner.
Give: (1) a three-sentence summary, (2) five key facts with numbers where available, (3) two risks or open questions.
Only use information from the text. If something is unclear, say so.
Text:
[paste]

Content ideas

You are a content strategist for [business type] targeting [audience].
Suggest 10 blog post titles that answer real questions this audience searches for.
For each title, give the search intent in five words and a one-line angle.
Avoid generic titles; be specific to [industry] in [country/region].

Data extraction

Extract the following fields from the text below and return them as a table: name, company, email, phone, service requested, budget (if mentioned).
If a field is missing, write "not provided". Do not guess.
Text:
[paste]

Meeting notes

You are an executive assistant. Turn the transcript below into: a 4-line summary, a list of decisions, and a list of action items with owner and due date if mentioned.
Use the exact names from the transcript.
Transcript:
[paste]

Seven habits that make prompts reliable

1. Show an example. One sample of the output you want (a "one-shot" example) improves consistency more than any clever wording.
2. Ask for a format. Tables, numbered lists and JSON are easier to use downstream, especially inside automations.
3. Say what to do when information is missing. "Write 'unknown' rather than guessing" prevents confident nonsense.
4. Split big tasks. Ask for an outline first, then the sections. Quality drops when you ask for a 2,000-word article in one go.
5. Give the audience. "For a first-time buyer" and "for a procurement manager" produce very different, better-targeted results.
6. Iterate, don't restart. "Make it shorter and remove the second paragraph" works well. The model remembers the conversation.
7. Save what works. Keep a shared document of tested prompts for your team. This is the single biggest productivity gain most businesses miss.

Prompting inside automations

When a prompt runs inside a Make, Zapier or n8n workflow, nobody is there to fix a weird answer. That changes the rules slightly:

• Lock the format. Ask for JSON with named fields, then validate them in the next step.
• Lower the creativity. Set temperature low for classification and extraction tasks.
• Add a fallback. If the model returns something unexpected, route the item to a human instead of sending it on.
• Log inputs and outputs. When something goes wrong, you need to see what the model saw.

I use exactly these rules in the workflows described in AI automation for small businesses.

ChatGPT or Claude?

Both are excellent for business writing and analysis. In my experience:

• Claude is particularly strong at following long, detailed instructions, working with large documents and producing natural, less "AI-sounding" writing.
• ChatGPT has a very broad ecosystem, strong image features and is often the one clients already have.

Most businesses do fine with either. What matters is the prompt structure above and a saved library of prompts your team actually uses.

Common mistakes to avoid

• One-line prompts with no context.
• Pasting confidential data (passwords, card numbers, unredacted client records) into consumer accounts.
• Trusting facts without checking. Models can invent statistics and sources. Verify anything you will publish.
• Letting AI write your whole brand voice. Use it to draft and edit, then add your judgement. The guide to content creation with AI covers this balance.

Frequently asked questions

Is prompt engineering a real skill or just a buzzword? It is a real, learnable skill, but it is closer to "writing a good brief" than to programming. Anyone who can delegate clearly can prompt well.

How long should a prompt be? As long as it needs to be to include role, context, task, format and input. For business work that is often 100–250 words, and that is fine.

Can you set this up for my team? Yes. Prompt engineering and reusable templates are part of my AI automation service, and I also build them into virtual assistance work so routine writing is fast and consistent.