Prompt Engineering: How to Get the Most Out of ChatGPT and Claude

As of April 2026. Models and their capabilities keep shifting, but the logic of good prompts has remained stable for years. This article focuses on the durable principles.

The quality of your AI outputs depends directly on the quality of your prompts. This article delivers the five core principles, three advanced techniques, and examples from everyday Swiss SME work. If you are looking for ready-to-use templates, you will find them in the prompt templates - eight copy-paste frameworks for emails, analyses, code, and more.

What is prompt engineering?

Prompt engineering is structured communication with a language model. It is not programming, but it follows recognisable patterns that consistently lead to better results.

The five core principles

1. Be specific

Bad: “Write a text about AI.”

Good: “Write a 300-word blog article about the benefits of AI for Swiss SMEs. Target audience: executives without a technical background. Tone: professional but accessible. Use formal English appropriate for a Swiss B2B context.”

2. Provide context

The more relevant context you provide, the better the result. Three questions you should always answer:

• Who is the target audience?
• What is the purpose (inform, persuade, decide)?
• Which tone is desired (formal, collegial, direct)?

3. Define the format

Tell the AI exactly what the result should look like: bullet points or flowing prose, word count, with or without headings. Clear specifications save the expensive iteration of “can you also give me that as a list?“.

4. Use examples

One-shot or few-shot prompting dramatically reduces stylistic drift:

Here is an example of the desired style:
[example]

Now write a similar text about [topic].

5. Iterate

The first output is rarely perfect. Refine rather than restart:

• “Make it shorter.”
• “More formal, please.”
• “Add a concrete SME example.”

Three concrete examples from everyday Swiss SME work

HR: structuring a job posting

You are an experienced HR manager in a Swiss industrial SME
(50 employees). Write a job posting for an order-processing
administrator (80-100 %).

Context:
- We are looking for a new hire to strengthen our five-person
  internal sales team.
- Requirements: commercial apprenticeship or equivalent,
  ERP experience (Abacus or SAP), native-level German,
  French is an advantage.
- We offer: 5 weeks of vacation, 2 days of home office per week,
  public-transport access.

Format: approx. 300 words, clear sections
(Responsibilities, Profile, What we offer), no empty phrases,
no over-familiar "we are looking for YOU" tone.

Finance: drafting a budget commentary

You are a controller in a Swiss services SME. Write a budget
commentary for the executive board covering Q1 2026.

Key figures:
- Revenue Q1: CHF 2.4 million (+8 % YoY, budget +12 %)
- Gross margin: 38 % (prior year 41 %)
- Personnel costs: CHF 920'000 (+5 % YoY)
- Liquidity: CHF 1.1 million (comfortable)

Format: one A4 page maximum, sections "Highlights / Variances /
Recommendations", no consultant-speak, clear cause-and-effect
statements. Language: English, formal Swiss business style.

Sales: drafting a quotation email

You are a sales representative in a Swiss mechanical-engineering SME.
Write a quotation email.

Recipient: Ms. Steiner, head of purchasing at an existing customer
(industrial, approx. 200 employees).
Occasion: she requested a quotation for a maintenance extension.
Content: two maintenance options (Standard CHF 4'500/year, Premium
CHF 7'200/year), available from May onwards, recommend a personal
conversation to clarify details.

Requirements: max. 180 words, professional and friendly tone,
formal address, clear call to action (propose a meeting).

Three advanced techniques

Chain of thought

Ask the model to think step by step:

Analyse this problem step by step:
1. Understand the situation.
2. Identify the core challenge.
3. Develop three solution options.
4. Evaluate the pros and cons.
5. Provide a recommendation with reasoning.

Role-based prompts

You are an experienced trustee (Treuhaender) with 20 years of
experience advising Swiss SMEs. Respond from that perspective.

Output structuring

Respond in the following format:

## Summary
[2-3 sentences]

## Key points
[bullet points]

## Recommendation
[1 paragraph]

Avoid common mistakes

1. Being too vague - more detail almost always produces better results.
2. Providing no examples - one example is more powerful than three descriptions.
3. Not iterating - the best outputs emerge through dialogue.
4. Feeding personal data into free tiers - relevant under the Swiss Data Protection Act (DSG), see DSG guide.

Conclusion

Prompt engineering is a learnable skill. With a little practice, your results become measurably more consistent, and rework goes down. The next level is the eight prompt templates for everyday SME work, which your team can deploy right away.

In our workshops for Swiss SME teams we walk through these principles with concrete examples from your industry.

---

Related reading on ai-edu.ch:

• Prompt templates: 8 frameworks for daily work
• Which AI tool for which task
• DSG and AI in Swiss SMEs

Sources:

• OpenAI: Prompt Engineering Guide
• Anthropic: Prompt Engineering Documentation
• Swiss Federal Act on Data Protection (DSG)