What is prompt engineering? A plain guide with real examples

If you have asked ChatGPT for something and got back a bland, generic answer, you have met the problem that prompt engineering solves. So what is prompt engineering? It is the craft of briefing an AI well: telling it who it is helping, what good looks like and how to check its own work. You do not need to code. This guide explains what it means in 2026, the techniques that work, with short examples, and where it pays off in a South African business.

What is prompt engineering, in plain words

A prompt is the text you give an AI model: a question, an instruction, pasted material, examples. Prompt engineering is writing and refining that text so the model consistently produces what you need. OpenAI describes it in almost exactly those terms in its developer guide. Google calls the same work prompt design. Anthropic treats it as something you test against clear success criteria, rather than a bag of tricks.

The models behind ChatGPT, Claude and Gemini are large language models, which predict text based on patterns learned from huge amounts of writing. If you want the basics of how they work, Zaiq explains what an LLM is. The short version for prompting: the model only knows what it learned in training and what you put in front of it. Everything else it guesses.

Why prompt engineering matters less, and more, in 2026

It matters less because models understand plain requests far better than early chatbots did. Magic phrases and elaborate tricks count for little now. OpenAI’s own guide notes that its reasoning models do best with a high-level goal, while its standard GPT models benefit from precise, explicit instructions.

It matters more because AI no longer lives only in a chat window. It sits inside automations, chatbots and agents that run the same prompt thousands of times, and a vague instruction repeats its mistake every time. Anthropic’s guidance starts with defining what success looks like and how you will test it, and points out that some problems, such as speed or cost, are better solved by choosing a different model. Its engineers have also started talking about context engineering: managing everything the model sees, not only the prompt.

The core prompt engineering techniques

Give it a role and context

The model cannot see your business. Tell it who it is, who it is talking to and the facts it needs.

Weak:   Write a reply to this customer.

Better: You are the front desk of a guesthouse in Durban. A guest asks on
WhatsApp whether we have parking and whether they can check in early on
Friday. We have free off-street parking. Early check-in from 11:00 costs
R250 if the room is ready. Reply in a short, friendly WhatsApp message.

Show an example

One or two examples of what you want beat a paragraph of description. Google’s prompting guide warns that prompts without examples are likely to be less effective. Paste a past quote you were happy with, or a reply in your house style, and ask the model to match it.

Say exactly what format you want

Name the structure: a table, a numbered list, three bullet points, a 60-word WhatsApp message. For long prompts, OpenAI suggests Markdown headings or simple tags to separate the instructions from the material the model should work on.

Break big jobs into steps

Ask for one stage at a time, or spell the stages out in order. Google recommends splitting complex tasks into simpler steps, sometimes as separate prompts where each answer feeds the next.

Check the output

Ask the model to quote the part of the document it relied on, then check that part yourself. Verify every number, name, date and legal point. For a prompt you will reuse, keep a handful of test cases and run them again whenever you change the wording.

South African business examples

Drafting a quote

You are helping a Pretoria plumbing business write quotes. Using the job
notes below, draft a quote with a table: item, quantity, unit price (R) and
line total (R). Add 15% VAT and a grand total at the bottom. Keep our
standard terms: 50% deposit, quote valid for 14 days. Do not invent prices;
if a price is missing, write "price to confirm".

Summarising a tender

Read the attached tender document. First, list every compulsory requirement
and returnable document. Second, list the closing date, any compulsory
briefing session and where bids must be submitted. Third, flag anything a
small construction company might not meet. Quote the page number for each
point.

Replying to WhatsApp enquiries

Here is our price list and our opening hours. Draft replies to the three
enquiries below in a warm, brief tone. If a customer writes in Afrikaans,
reply in Afrikaans. If you are unsure of an answer, say a colleague will
confirm rather than guessing.

Before you paste real enquiries into a public AI tool, remove names, phone numbers and anything else that identifies the customer. POPIA still applies when AI does the work, and our guide on putting customer data into ChatGPT covers the safe way to do it. For a wider view of using these tools at work, see Zaiq’s guide on how to use ChatGPT for business.

Common prompt mistakes

• Vague asks. “Make this better” gives the model nothing to aim at. Say better for whom, and how.
• No audience. A reply for a customer, a board and a supplier should read differently.
• Too many jobs at once. Five tasks in one prompt usually means three done badly.
• Trusting facts unchecked. Models state wrong figures, dates and laws with complete confidence.
• Pasting personal information into public chatbots, which puts your business on the wrong side of POPIA.
• Random rewrites. Changing three things at once means you never learn which change helped.

Is prompt engineer a real job?

In 2023, prompt engineer was one of the most talked-about new AI jobs. By April 2025, The Wall Street Journal was reporting that the stand-alone role was already becoming obsolete. That fits what the vendors’ own guides show: as models got better at plain requests, the tricks mattered less and the basics of a good brief became something everyone is expected to know.

The skill has not gone away. It has moved into other jobs: AI engineers write and test prompts inside software, automation specialists build them into workflows, and content and support teams use them daily. There is also paid remote work writing and grading prompts on AI training platforms such as Outlier. Our guide to AI jobs in South Africa covers those roles, and AI courses in South Africa lists where to learn, from free options to university short courses.