AI

Prompt Engineering

A practical guide to understanding what prompt engineering is and why it matters for AI product quality.

What prompt engineering involves, how it shapes AI output quality, and what product and design teams need to know to do it well.

22 May 20265 min read

What it is

is the practice of designing and refining the inputs given to an to reliably produce high-quality, useful outputs.

A is any instruction or input given to a language . is the deliberate craft of making those instructions as effective as possible.

The same can produce dramatically different results depending on how it is prompted. A vague instruction produces a generic . A well-structured, specific instruction produces a precise and useful one.

covers the design of user-facing inputs, , instruction sets for automated , and any other text that guides an AI model's behaviour.

It is an iterative — write a , evaluate the outputs, identify where it falls short, and refine.

When to use it

Understand when matters most.

It is most relevant when:

You are building or refining AI-powered features
Outputs are inconsistent or not meeting quality requirements
You want to improve AI reliability before investing in fine-tuning
You are designing system prompts for a product or workflow
Specific output formats, tones, or constraints are required

It is less relevant when:

The task is so simple that any reasonable prompt will work
Fine-tuning has been completed and further prompt refinement offers minimal gain

Key takeaway

Before considering fine-tuning or more complex solutions, prompt engineering is almost always the first and most cost-effective lever to pull.

How it works

Understand the basic mechanism. Effective share several characteristics. They are specific about the task and the expected output. They provide relevant . They set where needed. They use examples when a particular format or style is required.

Techniques such as — asking the to reason step by step before giving an answer — can significantly improve the quality of complex outputs.

Few-shot prompting — providing examples of good input-output pairs within the — helps the understand precisely what is expected.

What this means for designers and product teams. is not just for engineers. Anyone who understands what good output looks like, what the user needs, and what the requires can contribute meaningfully.

should be -controlled and tested systematically — not changed ad hoc. A change that improves one type of output can unexpectedly degrade another.

What to look for

Focus on:

Specificity — whether the prompt clearly communicates what is expected
Context — whether the model has enough background to respond well
Constraints — whether limits on format, length, or content are clearly stated
Consistency — whether the prompt produces reliable results across varied inputs
Examples — whether including sample outputs would help the model understand the target

Where it goes wrong

Most issues come from: Vague produce vague outputs — precision is the job.

Instructions that are too broad or open to interpretation
No examples when the expected output format is non-obvious
Testing on too few inputs and missing important failure cases
Changing prompts without a systematic way to evaluate the impact
Treating prompt writing as a one-time task rather than an iterative process

What you get from it

Understanding gives you:

A practical skill for improving AI output quality without technical overhead
A foundation for designing system prompts and instruction sets
More consistent, reliable AI feature behaviour
A lower-cost alternative to fine-tuning for many quality issues

Key takeaway

Prompt engineering is the most accessible and iterative way to improve AI quality. It belongs in the product and design toolkit, not just the engineering one.

FAQ

Common questions

A few practical answers to the questions that usually come up around this method.

What is prompt engineering?

is the practice of designing and refining the inputs given to an to consistently produce high-quality outputs. It involves understanding how a model responds to different instructions and crafting that guide it toward useful, accurate, and appropriate responses.

Is prompt engineering a technical skill?

Partly. The underlying concepts — how instructions, what techniques improve outputs — are learnable by anyone. What matters most is understanding what good output looks like for your use case and being systematic about testing and . You do not need to be an engineer to be good at prompt engineering.

What makes a good prompt?

, specificity, relevant , and clear . Telling the model exactly what you want, why, and in what format removes ambiguity and produces more reliable outputs. Including examples of good responses is often the single most effective improvement you can make.

How is prompt engineering different from fine-tuning?

guides an existing through better instructions. changes the model itself through additional training. Prompt engineering is faster, cheaper, and more reversible. Fine-tuning is more powerful for sustained performance improvements on a specific task but requires significant data and resource investment.

Does prompt engineering still matter if I am using a very capable model?

Yes. Even the most capable produce better results with well-designed . The gap between a vague and a precise prompt narrows as models improve, but it never disappears entirely. Good is always worth the effort.

Quick take

Writing better prompts is one of the highest-leverage skills for anyone building or designing AI products — and it is a design skill, not just a technical one.

Related Services

LET'S WORK TOGETHER

Ready to improve your product?

UX, research and product leadership for teams tackling complex digital services. The work usually starts where things have become harder than they need to be: unclear journeys, inconsistent products, competing priorities, or teams trying to move forward without a clear direction. I help simplify the problem, shape the right next step, and turn complexity into something people can actually use.

Previous feedback

Will Parkhouse

Senior Content Designer

01/20