Applied AI — Articles
AI Strategy & Implementation Insights
All Articles
Productivity · July 2025 · 7 min read

How to Write AI Prompts That Actually Work

Most people treat AI like a search engine — type a few words, hope for the best. The results are mediocre and they blame the AI. The real problem is the prompt. Here's the practical framework we use every day to get consistently excellent results from Claude, ChatGPT, and any other AI model.

JM
John Martines
Applied AI — NEPA & Lehigh Valley

The Gap Between Average and Excellent

Most people get average results because prompts are vague. AI doesn't fill in context from your head — it fills in context from its training data. In other words, it averages. The difference between a 5-minute answer and a 30-second answer is usually the prompt, not the AI.

Here's a concrete example. If you ask an AI "Write a proposal for a client," you'll get a generic proposal template that could apply to anything. It's technically correct, but it's not what you needed. If you ask it the same thing with proper context and detail, you get exactly what you wanted in a fraction of the time.

❌ Bad

Write a proposal for a client

✓ Good

Write a 1-page project proposal for a manufacturing client in the Lehigh Valley. They need help automating their quality control documentation. Tone: professional but approachable. Include: project scope, 3 deliverables, a 6-week timeline, and a next-steps CTA. Format: headers and bullets.

The second one took maybe 30 seconds longer to write. The AI's response will be 10 times better.


The Five Elements of a Strong Prompt

Every excellent prompt has five ingredients. You don't always need all five, but when your results are mediocre, one of these is missing.

1

ROLE

Tell the AI who it's being. "You are a senior accountant reviewing..." gets different results than a bare question. Role-setting activates different knowledge patterns. It primes the AI to think like someone in that position would think.

2

CONTEXT

Give it the background it needs. Industry, company size, audience, constraints. What does it need to know to answer correctly? The more specific, the better. "Manufacturing company in NEPA" is better than "company." "Quality control team with 5 years average experience" is better than "our team."

3

TASK

Be specific about what you want. Not "help me with this email" but "rewrite this email to be more direct, reduce it by 30%, and move the ask to the first paragraph." Vague tasks get vague results.

4

FORMAT

Tell it how to structure the output. Bullet list, table, numbered steps, paragraph length, JSON, email format. If you don't specify, it guesses — and usually guesses wrong for what you need.

5

CONSTRAINTS

What should it avoid? "Don't use jargon," "keep it under 200 words," "don't mention our competitor by name," "assume the reader has no technical background." Constraints are how you steer the AI away from bad outputs.


Common Mistakes

These are the patterns we see that consistently kill prompt quality:

❌ One-liner questions

"Analyze our sales data"

✓ Full context

"You are a sales analyst. Here's our Q2 data [attached]. Identify the top 3 trends, flag anomalies, and suggest one action for each. Format: brief report with bullets."

❌ Asking for everything at once

"Write our entire marketing plan"

✓ Breaking into steps

"First, let's define our positioning statement. Here's our current elevator pitch and our target audience. What should change?"

❌ Accepting the first answer

AI gives you something, you use it as-is

✓ Iterating

"Make it shorter / more formal / add an example / cut the jargon." Each iteration refines the output.

❌ Treating AI as a search engine

Type and hope; move on

✓ Treating it as a collaborator

Use follow-ups and refinements; guide it toward what you actually need


Prompting for Business Tasks

Here are four common business tasks with weak and strong prompts side-by-side. Study the difference:

Task Weak Prompt Strong Prompt
Meeting Summary Summarize this meeting Summarize this 45-min sales call. Bullet key points, action items with owners, and any objections raised. Max 300 words.
Follow-up Email Write a follow-up email Write a follow-up email to a prospect who went quiet after our demo 2 weeks ago. Warm tone, no pressure. Include a specific next step and keep it under 150 words.
Data Analysis Analyze our sales data You are a sales analyst. Here is our Q2 data [data]. Identify the top 3 trends, flag any anomalies, and suggest one action for each. Format as a brief report.
SOP / Procedure Write a procedure Write a step-by-step SOP for our receiving department to process inbound freight. Audience: new warehouse staff with no prior experience. Include safety notes and a checkpoint list.

Notice the pattern? The weak prompts are one to four words. The strong prompts give role, context, task, format, and constraints. That's the framework.


Iterating: The Real Secret

Here's the truth most people don't realize: the first response is a draft, not a deliverable. Prompting is a conversation, not a command. You're collaborating with the AI to refine an output until it's exactly what you need.

Common iteration prompts that work:

  • "Make it more concise"
  • "Add a specific example for [X]"
  • "Rewrite the opening — it's too formal"
  • "Now put that in a table"
  • "Use simpler language"
  • "Add 2-3 more bullet points"

Chain prompting is also powerful: break big tasks into steps, each building on the last. "First, write the positioning statement. Then, write 3 value propositions based on that. Then, write a 100-word elevator pitch using those props." You're guiding the AI through a process.


Applied AI's Prompting Principles

How We Build AI Skills

We build these same five principles into every skill we deploy: role, context, task, format, and constraints — baked in so your team gets expert-level results without having to become prompt engineers themselves. That's what grounded AI is: we embed the expertise into the tool, not the user.


The Real Opportunity

Once you understand how to prompt well, AI stops feeling like a toy and starts feeling like a superpower. A 30-second investment in a better prompt saves you 15 minutes of editing. That compounds across your entire team.

This is exactly why we build custom AI tools for businesses. The same prompting discipline that makes you better at ChatGPT makes us better at building solutions tailored to your specific workflows.

Want AI That Works Without The Learning Curve?

That's what Applied AI does. We bake expert-level prompting and domain knowledge into custom tools so your team gets excellent results without becoming prompt engineers. Reach out for a demo of what grounded AI can do for your business.