Prompting Is a Design Skill

Most disappointing AI output starts with a one-line prompt like "write error messages for a signup form" and a lot of optimism.

And the model does exactly what it was asked. You get "Invalid email address" and "This field is required", the same two messages every form on the internet already has.

Now try it with what you actually know. The users are technical, so skip the friendly filler. Keep each message under eight words and tell people what to do instead of what they did wrong. This time you get "Enter a work email address." It’s the same model. The only difference is that the thinking happened before anyone hit enter.

The prompt says who it’s for and what good looks like

Prompt

TaskInline error messages for email, company name and password in a B2B SaaS signup form.

AudienceTechnical users. Not first-timers.

ToneDirect. Say how to fix it. No "Oops!"

ScopeUnder 8 words each. One message per field.

Output, in the signup form

Work email⚠ Enter a work email, like name@company.comCompany name⚠ Enter your company namePassword⚠ Password must be 8 characters or more

Each message says what to do, in the field’s own words. Try fixing the fields.

Illustrative example

Don’t

The prompt only names the task

Prompt

Write some error messages for a signup form

Output, in the signup form

Work email⚠ Please enter a valid email address so we can reach you.Company name⚠ This field is required. Please fill it in to continue.Password⚠ Your password is too short. Please choose a longer, more secure password.

It filled the gaps with defaults: an "Oops!" banner and long, generic sentences.

Illustrative example

That’s why I think prompting is a design skill, and a pretty old one. It’s the same work as writing a good brief or explaining to an engineer exactly what should happen when a filter returns nothing. You take something that’s clear in your head and make it clear on the page, for someone who wasn’t in any of your meetings.

The model knows nothing about your product. It doesn’t know your users are compliance analysts on a deadline, or that you already tried the cheerful tone and nobody liked it. Whatever you leave out, it fills in with the most average answer it can find. Most bad output I see comes from missing context that felt too obvious to mention.

It also changes when the thinking happens. When I design by hand, I work out what I want while I’m making it, and the file and my opinion grow up together. With a model I have to know what I want before I start typing, or I get something confidently generic and spend the afternoon arguing with it.

With AI, the thinking has to happen before the making

How the work goes without AI and with it

Without AI

Thinking happens during makingYou discover what you mean by trying things

Artifact and intent co-evolveThe work teaches you what the work should be

Intuition fills the gapsTacit knowledge guides decisions mid-process

Iteration is explorationEach version is also a question

With AI

Thinking happens before makingIntent must be expressed before the model can act

Intent must be explicit firstThe model has no intuition to fall back on

Gaps produce generic outputWhat you leave unspecified gets filled with defaults

Iteration is refinementEach prompt is more informed than the last

The thinking moves earlier. The better prompt comes from knowing what good looks like before opening the tool.

The people who are good at this don’t treat it as one shot. They read the output, notice what they don’t want, change one thing and run it again. It’s basically sketching, with more typing.

So if you want to get better at prompting, I wouldn’t buy anyone’s list of fifty magic prompts. I’d start by writing sharper briefs and deciding what you care about before you open any tool.

Tamara Milakovic is a senior product designer specialising in B2B SaaS, design systems, and the decisions upstream of the brief.

When to Standardise and When to Leave It Alone

Clarity Was Never the Finish Line

The Brief Is Rarely the Whole Problem