AI Prompt Engineering for Non-Technical People

How to ask AI the right way so you get usable results the first time

Executive Summary

Most people who try an AI writing tool and get a useless result assume the tool is the problem. Usually it is not. The problem is how the question was asked.

AI tools respond to instructions the way a new employee responds to a vague assignment — they fill in the gaps with assumptions, and those assumptions are rarely what you needed. Give them a clear brief and the output improves dramatically.

This guide gives you a simple four-part structure for writing better instructions to any AI tool. You do not need a technical background. If you have ever briefed a contractor, written a job description, or explained a task to a new hire, you already have the instincts. This guide shows you how to apply them.

What you will get from this guide:

• A repeatable structure for writing AI instructions that produce usable results faster
• Real before-and-after examples built around everyday business tasks
• Copy-and-paste templates you can adapt today
• A short guide to diagnosing prompts that go wrong
• A one-page quick reference you can keep nearby

The Problem This Solves

Here is a scenario that plays out constantly in small businesses:

A manager or owner has heard that AI can help with writing, so they open a tool and type something like:

Write me an email to follow up with a customer.

What comes back is a generic, stiff, corporate-sounding email that does not match their tone, does not reference anything specific about the customer, and would need to be completely rewritten before anyone would send it.

They try a few more times with slightly different wording. Same problem. After twenty minutes, they close the tab and go back to writing it themselves, now convinced the tool is overhyped.

The tool is not the problem. The instruction was too thin.

AI tools are not search engines. Typing a short phrase and expecting a finished result is like handing a new hire a Post-it note and expecting a completed project. They need more to work with.

The good news: once you know what "more" looks like, writing a better instruction takes about sixty additional seconds, and the output is almost always usable on the first try.

The Core Idea: A Prompt Is a Brief, Not a Search

When you use a search engine, you type a few words and get links. You do the reading and thinking yourself. The search engine just finds things.

AI tools work differently. You are not searching — you are delegating. You are asking the tool to think, write, or organize something on your behalf. And like anyone you delegate to, it needs context to do a good job.

Think about the last time you brought in a contractor or a freelancer for a specific task. You probably told them:

• What role they were filling
• What the situation was
• What you needed them to produce
• What the deliverable should look like

That is exactly what a well-written AI instruction does. The technical term for this instruction is a "prompt" — you may have heard that word in the context of "prompt engineering." That phrase makes it sound complicated. It is not. It is just learning how to give a clear brief.

The Four-Part Structure That Works

Every strong AI instruction has four components. You do not always need all four — but the more of them you include, the better your result.

Let's look at each one in more detail.

Role

Telling the AI what role to take on is one of the fastest ways to improve output quality. Without a role, the AI picks one for itself — usually some generic version of a helpful assistant.

With a role, it calibrates its language, tone, and approach to match someone with relevant expertise.

Generic: Write a job posting.

With a role: Act as an experienced HR manager at a small electrical contracting firm.

The difference in output quality is significant. The second instruction produces writing that sounds like it came from someone who understands the trade, not a generic template.

You do not need to make the role elaborate. Just give it a job title and an industry context.

Context

Context is the situation the AI needs to understand before it can help. The more relevant detail you give here, the less the tool has to guess.

Context answers questions like:

• Who is the audience?
• What happened before this moment?
• What is the business or service?
• What is the relationship?

Thin context: A customer is upset.

Useful context: A customer called to say their water heater installation was scheduled for Tuesday but no one showed up and no one called. They have been a customer for three years. This was our mistake — a scheduling error on our end.

With the second version, the AI can write something that actually fits the situation. With the first, it guesses — and usually guesses wrong.

Task

The task is the specific thing you need produced. This sounds obvious, but it is where most instructions are too vague.

Too vague: Help me with a response.

Clear task: Write a short apology email that acknowledges the mistake, takes responsibility without making excuses, and offers to reschedule within 24 hours.

Notice that the clear version also includes what the response should accomplish. The more you describe the outcome you want, the closer the first draft will be.

Format

Format tells the AI what the finished product should look like. Without this, you get whatever default the tool defaults to — often too long, too formal, or structured in a way that does not match how you communicate.

Format instructions can cover:

• Length ("under 100 words," "three short paragraphs")
• Tone ("friendly but professional," "direct, not salesy")
• Structure ("use bullet points," "no headers," "subject line included")
• Voice ("write in first person," "avoid jargon")

Format is especially important if you are using the output in customer-facing communication. A 400-word formal email is not the same as a 75-word conversational one — but the AI does not know which you need unless you say so.

Before and After: Real Business Examples

The following examples show what a weak instruction produces versus what a structured one produces. These are the kinds of tasks business owners use AI for every week.

Example 1: Follow-Up Email After a Sales Call

Before (weak instruction):

Write a follow-up email after a sales call.

What you get: A stiff, template-sounding email that says things like "It was a pleasure speaking with you" and "Please do not hesitate to reach out." Nothing specific. Nothing that reflects the actual conversation.

After (structured instruction):

Act as a sales manager at a small commercial cleaning company.

Context: I just had a 20-minute call with the office manager at a mid-sized accounting firm. They are unhappy with their current cleaning company because of inconsistent service on Fridays. They mentioned they have 30 employees and their space is about 4,000 square feet. I did not give them pricing yet — we agreed to follow up with a proposal.

Task: Write a follow-up email that thanks them for the call, references the Friday reliability issue so they know I was listening, and sets up the next step of sending a proposal by Thursday.

Format: Keep it under 120 words. Friendly and professional. No bullet points. Include a subject line.

What you get: An email that sounds like it came from a real person who was in that conversation, references their specific pain point, and moves the deal forward.

Example 2: Job Posting for a Technician

Before (weak instruction):

Write a job posting for a plumber.

What you get: A generic job description that could apply to any company anywhere. It lists "competitive salary" and "team environment" — phrases that every candidate has seen a hundred times.

After (structured instruction):

Act as an operations manager at a family-owned residential plumbing company in a mid-sized city.

Context: We are hiring a licensed journeyman plumber. Our company has 8 employees. We do primarily service calls and bathroom remodels — no new construction. We want someone who can work independently, is good with customers, and is reliable. We pay above market rate and offer health benefits and a company truck.

Task: Write a job posting that will attract experienced candidates who want stable work at a quality shop, not just anyone looking for any plumbing job.

Format: Under 300 words. Use a few short sections with simple headers. Lead with what makes us a good employer before listing requirements. Avoid corporate language.

What you get: A posting that sounds like a real company, leads with the value to the applicant, and filters for the kind of candidate you actually want.

Example 3: Summarizing a Customer Complaint

Before (weak instruction):

Summarize this customer complaint.

What you get: A restatement of the complaint in different words, often longer than the original.

After (structured instruction):

Act as an operations manager reviewing customer feedback.

Context: The following is a written complaint from a customer. I need to share a summary with my service team at our weekly meeting.

Task: Summarize the complaint in three bullet points: (1) what the customer says happened, (2) how it affected them, and (3) what they are asking for.

Format: Three bullets, plain language, under 60 words total. Do not include the customer's name.

[Paste the complaint text here]

What you get: A clean, structured summary your team can act on in thirty seconds.

Example 4: Responding to a Negative Online Review

Before (weak instruction):

Write a response to a bad Google review.

What you get: An overly formal response that starts with "We are sorry to hear about your experience" and does not sound like a real person wrote it.

After (structured instruction):

Act as the owner of a small landscaping company responding to a customer review online.

Context: A customer left a 2-star review saying our crew left clippings on their driveway after a job and did not respond when they called to complain. We did drop the ball on the callback. This response will be public.

Task: Write a response that acknowledges the specific complaint, takes responsibility for the missed callback, and invites them to contact us directly to make it right. Do not get defensive.

Format: Under 80 words. Conversational and genuine. First person. Do not use phrases like "we apologize for any inconvenience."

What you get: A response that sounds like a real business owner, acknowledges what actually happened, and gives readers confidence that the company handles problems well.

Copy-and-Paste Prompt Templates

Use these as your starting point. Fill in the brackets with your details.

Template 1: Any Customer-Facing Email

Act as [your role, e.g., "the owner of a small HVAC company"].

Context: [Describe the situation — who is the customer, what happened, what is the relationship?]

Task: Write an email that [describe what the email should accomplish — thank them, apologize, follow up, confirm, etc.].

Format: [Length, tone, structure, e.g., "Under 100 words, friendly and direct, include a subject line, no bullet points."]

Template 2: Job Posting

Act as the hiring manager at [describe your company briefly — size, type, location context].

Context: We are hiring a [job title]. Key details: [list 3–5 things that matter — experience required, type of work, company culture, pay or benefits if you want to mention them].

Task: Write a job posting that appeals to [describe your ideal candidate] and accurately represents the role.

Format: [Length, sections you want, tone, e.g., "Under 300 words, lead with why we're a good employer, then requirements, avoid buzzwords."]

Template 3: Internal Summary or Briefing

Act as [an operations manager / office manager / team lead] summarizing information for [your team / your manager / a meeting].

Context: The following is [a customer complaint / meeting notes / a support ticket / a report]. The audience is [who will read this and why].

Task: Summarize the key points in [format you want — bullets, short paragraph, numbered list]. Specifically pull out: [what matters most, e.g., what the issue is, what action is needed, who is responsible].

Format: [Under X words, plain language, no jargon.]

[Paste your source text here]

Template 4: Online Review Response

Act as the owner of [type of business] responding publicly to a customer review.

Context: The review says [summarize what the reviewer said — the complaint or the compliment]. [Add any relevant context about what actually happened, if you know.]

Task: Write a response that [acknowledges the feedback / thanks them / addresses the issue / invites them to contact us].

Format: Under [X] words. Genuine and conversational. First person. Do not use corporate phrases like "we apologize for any inconvenience" or "we take this very seriously."

Template 5: Social Media or Marketing Post

Act as a marketing writer for [type of business].

Context: [Describe the business, the audience, and any relevant detail — a promotion, a season, a service feature, a recent project].

Task: Write a [platform, e.g., Facebook / Instagram / LinkedIn] post that [goal — promotes a service, shares a tip, announces something, etc.].

Format: [Length appropriate for the platform, tone — conversational / professional / local and friendly, include or exclude hashtags, include or exclude a call to action].

When a Prompt Fails: How to Diagnose and Fix It

Even with a good structure, you will occasionally get output that misses the mark. Instead of starting over, learn to read the failure.

Here is a quick diagnostic guide:

The fastest fix is usually one sentence.

If the response was too formal, add: "Use a conversational, friendly tone — not corporate language."

If it missed the context, add one more specific sentence about the situation.

If the length was wrong, add a word count.

You do not need to rewrite the whole instruction. Find the gap and patch it.

Common Mistakes

1. Treating it like a search engine Typing a short phrase and expecting a finished result. AI tools need instructions, not queries.

2. Being vague about the task "Help me with customer communication" is not a task. "Write a two-paragraph email that apologizes for a late delivery and offers a discount on the next order" is.

3. Skipping the format If you do not say how long or what tone, you will get the default — which is usually too long and too formal.

4. Not giving it a role The AI will pick a role. It might pick one that does not fit your industry or audience. Give it one that does.

5. Starting over when you should be iterating If the output is 70% right, do not scrap it. Add one clarifying sentence and ask the tool to revise. This is faster than rewriting from scratch.

6. Pasting in sensitive information More on this below — but as a habit, do not paste in customer data, financial details, or confidential business information unless you have confirmed how the tool handles that data.

7. Asking for too many things at once If you need an email, a social post, and three subject line options, do those as separate instructions. Stacking too many tasks in one instruction usually produces mediocre results across all of them.

When NOT to Use This

Prompt structure improves AI output — but there are situations where the tool should not be doing the work at all, no matter how good the instruction is.

Do not use AI to:

• Draft legal agreements or contracts. AI can produce something that looks like a legal document but is not reviewed for your jurisdiction or situation. Use a real attorney.

• Give customers specific advice on regulated topics. If you are in financial services, insurance, law, healthcare, or any other regulated field, do not use AI to generate advice for individual clients without compliance review.

• Make final decisions on hiring, termination, or HR matters. AI can help you draft a job posting or organize your notes — it should not be making or strongly influencing people decisions.

• Replace a human voice in high-stakes personal situations. If a long-term customer has had a serious problem, a form-letter AI response is worse than a phone call. Use AI to prepare what you want to say — not to replace saying it.

• Generate content you cannot verify. If the AI includes facts, figures, or claims you are going to publish or share with customers, check them. AI tools can state things confidently that are wrong.

Privacy Guidance

This is worth its own section because it is easy to overlook in the moment.

When you are writing a prompt and want to include context, it is tempting to paste in real customer names, email threads, complaint details, or account information. Before you do, consider a few things:

What you should avoid pasting into AI tools:

• Full names combined with contact details, account numbers, or complaint specifics
• Financial details, invoices, or payment information tied to a customer
• Private health-related information (especially relevant for medical or wellness businesses)
• Confidential employee information
• Proprietary business information you would not want stored outside your systems

What you can do instead:

• Anonymize before you paste. Replace the customer name with "the customer" or a generic placeholder. Replace specific dollar amounts with approximate figures if you do not need precision.
• Describe the situation in your own words rather than pasting the original source.
• Check your tool's privacy policy before using it for sensitive work. Different tools have different data retention and usage policies.

A practical rule: If you would not paste it into a public form on a website, do not paste it into an AI tool without checking how your data is handled first.

None of this means AI tools are unsafe — it means they are tools, and like any tool, they should be used thoughtfully.

Action Plan

Here is how to put this into practice starting today.

Step 1: Pick one task you did this week that involved writing. A customer email, a team update, a social post, a job listing — anything you wrote by hand that took longer than you wanted.

Step 2: Write a four-part instruction for that task. Use the structure: Role, Context, Task, Format. Spend no more than two minutes on it.

Step 3: Run it through any AI tool you have access to. Do not grade it harshly after one try. Read the output and identify what is close and what is off.

Step 4: Make one adjustment and try again. Use the diagnostic table in this guide to find the gap and add one sentence to fix it.

Step 5: Save the prompt that worked. Keep a simple document — a notes file, a Google Doc, anything — with the prompts that produced good results. Label them by task. Over time, this becomes a personal library of instructions you can reuse and adjust.

Once you have a working prompt for one task, you have the pattern. Apply it to the next task, and the one after that. The structure is the same every time.

Quick Reference Card

Print this or keep it in a tab. Use it every time you sit down to write an AI instruction.

The Four-Part Prompt Structure

Role → Who should the AI act as? "Act as an experienced [job title] at a [type of business]."

Context → What is the situation? "The customer / candidate / situation is... [relevant details]."

Task → What exactly do you need? "Write / summarize / draft / create [specific output] that [accomplishes X]."

Format → What should it look like? "[Length] words. [Tone]. [Structure — bullets, paragraphs, headers]. [Anything to include or avoid]."

Diagnostic: When the Output is Wrong

Privacy Checklist Before You Paste

• No full customer names paired with complaint or account details
• No financial information tied to a specific person
• No health-related or sensitive personal details
• No confidential employee information
• You have checked (or are comfortable with) how this tool stores data

One-Minute Prompt Check

Before you hit submit, ask yourself:

1. Did I tell it who to act as?
2. Did I explain the situation?
3. Did I describe exactly what I need?
4. Did I say what the output should look like?

If you can answer yes to all four, you are set.

This guide is part of the Local AI Foundations series. Local AI tests AI tools, workflows, and ideas so business owners do not have to.