Prompt Engineering for Beginners: A Practical Guide

Infinex·December 8, 2025·6 min

TL;DR: 80% of what you get from an AI tool depends on how you communicate with it. Prompt engineering isn't a technical skill reserved for developers — it's a communication method that any business owner or team member can master in a few hours of practice.

Why Your AI Disappoints You (And It's Not the AI's Fault)

"I tried ChatGPT but the results weren't great." That's the sentence we hear most often in early conversations.

Invariably, when we ask to see the prompts that were used, we find vague instructions, absent context, and unclear objectives. The AI returned something generic because it had no other choice.

An LLM (GPT, Claude, Gemini) is an ultra-sophisticated completion tool. It doesn't read your mind. It doesn't know your business, your industry, your tone, or your constraints — unless you give them.

The Structure of an Effective Prompt

A good prompt relies on four elements, remembered with the acronym RICE:

R — Role

Tell the AI who it's supposed to be in this exchange.

"You are a B2B communication expert specializing in industrial SMBs."

Assigning a role calibrates the language level, assumed expertise, and tone of responses. Without a role, the AI adopts a generalist register that fits everyone and impresses no one.

I — Instruction

The precise action you're requesting. Use clear action verbs: write, summarize, list, compare, translate, rewrite, analyze.

"Write a follow-up email for a prospect who hasn't responded to our quote in 10 days."

Avoid open-ended prompts like "tell me about" or "explain." They produce essays. Prefer closed formulations that imply a specific deliverable.

C — Context

The information the AI needs to personalize its response.

"Our company makes material handling equipment. The prospect is Operations Director at an 80-person SMB. Our quote was for 3 forklifts at $50,000."

The richer the context, the less generic the response. Don't be afraid to provide a lot of context — modern LLMs handle long prompts very well.

E — End format (Output)

The expected format for the response.

"Format: professional email of 5 to 8 lines, warm but direct tone, no excessive pleasantries."

Specify the length, format (list, paragraphs, table, JSON), tone, and any specific constraints (don't mention pricing, use informal register, avoid jargon).

Practical Examples by Use Case

Writing Sales Emails

Role: You are an experienced B2B salesperson in the business services sector.

Instruction: Write a first-contact email for a prospect met at a trade show.

Context: The person is an HR Director at a 120-person SMB (logistics sector). We talked for 5 minutes about their difficulties with continuous training for their field teams. Our offer: a custom AI training program.

Output: 6-8 line email, subject line included, professional but not corporate tone, one value proposition, one closing question to open dialogue.

Meeting Summary

Role: You are a precise and concise executive assistant.

Instruction: Summarize this meeting transcript into clear action points.

Context: [paste raw meeting notes here]

Output: List of decisions made, list of actions with owner and deadline, list of open items. Maximum 200 words total.

Product FAQ Generation

Role: You are a product expert for a scheduling software for tradespeople.

Instruction: Generate 10 frequently asked questions that new clients have during onboarding, with the associated answers.

Context: The software allows managing appointments, sending SMS reminders, and invoicing directly from the calendar. Target users: plumbers, electricians, carpenters with 2 to 8 technicians.

Output: Q&A format, answers 2-3 lines maximum, simple and direct language.

Iteration: The Most Underrated Skill

The best prompt is never the first one. Iteration is the real driver of quality.

When the result doesn't work for you, don't start from scratch. Change one element at a time:

Result too long? Add "Maximum 100 words."
Tone too formal? Add "Conversational tone, like you're writing to a colleague."
Content too generic? Enrich the context with specific details.
Wrong format? Specify exactly what you want to see.

Save your best prompts in a file or Notion page. A reusable good prompt is worth far more than a single perfect result.

Advanced Techniques to Go Further

Chain-of-Thought (Step-by-Step Reasoning)

For complex tasks that require analysis, ask the AI to reason step by step before concluding.

"Before answering, list the 3 key factors to consider, then analyze each one, then formulate your recommendation."

This technique significantly reduces hallucinations on topics that require reasoning.

Few-Shot Prompting (Examples as Input)

Provide one or two examples of the expected response type before your actual request.

"Here's an example of an email I like: [example]. Using the same style and structure, now write an email for [situation]."

Negative Constraints

Specify what you don't want, not just what you want.

"Avoid filler phrases like 'in today's fast-changing world' or 'don't hesitate to reach out.' Avoid buzzwords. No bullet points."

Reusable Templates for Your Team

Build a library of validated prompts for your team's recurring tasks. A few examples to adapt:

| Use case | Key RICE elements | |----------|------------------| | Sales follow-up email | Sales role, specific prospect context, output length + tone | | Meeting summary | Assistant role, action list instruction, 200-word max output | | Product sheet | Product expert role, technical specs context, structured format output | | Response to negative review | Customer service role, verbatim review context, empathetic tone + solution output |

Standardizing these prompts takes a day. The impact on your team's consistency and production speed is immediate.

Where to Start

Pick a repetitive task you do at least twice a week. Write a RICE prompt for that task. Test it three times, iterate each time. Share the validated prompt with your team.

That's it. No two-day training session, no additional tool. The skill builds through use.

To go further, see our guides on structuring an AI training program for your business, choosing between ChatGPT and Claude for your company, and free AI tools for SMBs.