Prompt engineering for SMEs: better prompts, better output
From vague question to usable answer: four ingredients and five patterns to get the most out of ChatGPT, Microsoft Copilot and other AI tools.
- A good prompt contains four ingredients: a role ('You are a marketing copywriter'), context (situation and product), task (what you want to achieve) and format (list, table, email, maximum X words). Without a format instruction, AI delivers continuous prose you still have to reformat yourself.
- Five patterns work best for SMEs: writing email (role plus tone plus recipient), summarising (document type plus length plus focus), analysis (data source plus output type), answering customer queries (product context plus brand tone) and generating code (language plus specification).
- The three most common mistakes are: giving too little context (AI fills gaps arbitrarily), not specifying a format (output lacks structure for direct use) and forgetting to assign a role (making tone and knowledge level unpredictable).
Why does a better prompt make such a big difference?
A prompt with a clear role, context, task and format delivers 40 to 60 percent more usable output than a short, vague question. AI models such as ChatGPT (GPT-4o), Claude, Microsoft Copilot and Google Gemini are powerful, but they fill in missing information based on statistical patterns, not your specific situation. A well-structured prompt gives the model enough framing to produce a specific, consistent and immediately usable result. This article explains the four-ingredient model and gives five SME patterns with practical examples.
The four ingredients of an effective AI prompt
An effective AI prompt has four components: role (who is the AI in this task), context (what is the situation), task (what should the AI do) and format (what should the output look like). Each ingredient reduces the chance of vague or generic output.
Step 1: Give the AI a role
Start your prompt with 'You are a...' and describe the expertise you need. Example: 'You are an experienced copywriter who writes for Dutch small business owners.' The model calibrates its tone, vocabulary and knowledge level to this role. Without a role, ChatGPT writes from a neutral position; with a role it writes from a specific expertise. This is the most underestimated ingredient.
Step 2: Provide context
Describe the situation, the product or the target audience. Example: 'We sell accounting software for independent hairdressers in the Netherlands. Our USP is automatic VAT filing in three clicks.' The more specific the context, the less the AI needs to fill in itself. Product context, target audience, brand tone and relevant background belong here.
Step 3: Specify the task
State exactly what the AI should do. 'Write' is vague; 'Write an email of at most 120 words that acknowledges a dissatisfaction and offers a voucher of 10 euros' is concrete. The task determines what the model does; the context determines how it approaches that. Use action verbs: write, summarise, analyse, translate, rephrase or generate.
Step 4: Specify the format
Say how the output should look: 'as a numbered list', 'as a Markdown table with three columns', 'as an email with subject line', 'maximum 200 words', 'in a professional but accessible tone'. Without format instructions, AI delivers continuous prose. With format, it delivers output you can copy and paste directly.
Five prompt patterns for SMEs: by use case
The five most commonly used SME prompt patterns are: writing email, summarising, analysis, answering customer queries and generating code. Each pattern applies the four-ingredient model to a specific work context.
Email writing: role plus tone plus recipient
You are a customer service agent for [company name]. Our tone is friendly but professional. Write an email to a customer whose order was missed. Acknowledge the inconvenience, explain the solution and close with a direct next step. Maximum 150 words. This pattern works for complaint handling, quotes and follow-ups. In Microsoft Copilot (Outlook) you use the same instruction structure via the Draft prompt.
Summarising: document type plus length plus focus
Here is a three-page meeting report. Create a summary of at most 150 words. Focus on: decisions made, open action items and who is responsible. Use a numbered list for actions. This pattern works for meeting notes, reports, contracts and email threads. In ChatGPT paste the document after your instruction; in Copilot in Word use Summarize.
Analysis: data source plus output type
Here are our website visitor figures for the past four weeks (table attached). Identify the three pages with the highest bounce rates and suggest one concrete improvement per page. Present the result as a table with three columns: page, bounce rate, recommendation. This pattern works for customer data, sales figures, survey results and process data in Excel. In Copilot in Excel use the Analyze button with a follow-up instruction.
Answering customer queries: product context plus brand tone
You are an agent for [webshop name]. We sell organic coffee. Our tone is warm, direct and honest. Answer the following customer query: 'Can I still cancel my order?' Give a clear yes or no, explain the conditions and refer to customer service for exceptions. Maximum 80 words. This pattern is the basis of a chatbot script or standard-reply library.
Generating code: language plus specification
Write a Python script that reads a CSV file, filters all rows where the column 'revenue' is less than 1000, and saves the filtered rows as 'low-revenue.csv'. Use the pandas library. Add comments per section. This pattern works for simple automations in Python, JavaScript or Excel formulas. Naming the specific language and library is required; without it ChatGPT writes a generic implementation.
Three mistakes most SME owners make when prompting
The three most common AI prompting mistakes are: giving too little context, not specifying a format and forgetting to assign a role to the AI. All three can be fixed by adding one step to your existing prompt. Mistake 1: too little context. AI fills in missing information arbitrarily from its training data. If you do not say what your product is, who the customer is or what the tone should be, you get a generic answer that still needs editing. Fix: always add a context sentence describing product, target audience and tone. Mistake 2: not specifying a format. Without a format instruction, AI delivers continuous prose. If you want to use the output in an email, presentation or Notion page, say so explicitly. Fix: always end your prompt with 'Give the result as [format]'. Mistake 3: forgetting to assign a role. Without a role, ChatGPT writes from a neutral, academic position that is often too formal or too generic for business use. Fix: always start with 'You are a...' followed by the expertise you need.
Frequently asked questions about prompt engineering for SMEs
What is prompt engineering?
Does prompt engineering work the same in ChatGPT, Copilot and Gemini?
How long should a good prompt be?
Can I put customer data in an AI prompt?
What is the difference between a prompt and a system prompt?
Want to know how AI can truly speed up your workflows?
Book a free call. We look at your current workflows together and advise which AI tools and prompts will save the most time — free and without commitment.
Book a free call →Free and without obligation · Reply within one business day · Fixed price up front, no surprises