Morphic animated logo
AI + Creative Intelligence

Why Does AI Follow My Instructions but Still Produce Bad Design?

Short answer: AI can follow a creative instruction correctly without knowing whether that instruction improves the final work. Compliance and judgment are different capabilities: a model may satisfy every individual request while the overall design becomes less clear, less coherent, more generic, or less effective.
By Weston Baker · Founder, Morphic

One of the biggest misconceptions about AI design is that better instruction-following automatically produces better design.

It doesn't.

You can give an AI exactly what you asked for and still end up with something you don't want.

Following instructions and making good creative decisions are different objectives.

Your instruction may not contain the solution

Imagine telling a designer:

“Put our six benefits into cards.”

A designer might ask:

Why
Why cards?
Emphasis
Do all six benefits deserve equal emphasis?
Grouping
Could three be grouped together?
Format
Would a diagram communicate this better?
System
Does the page already have too many cards?

The instruction contains a requested implementation.

It doesn't necessarily contain the best solution.

General-purpose AI is often optimized to be helpful and responsive, so it tends to accept the premise.

You asked for cards. It gives you cards. Success.

Except perhaps the website is now worse.

Creative work contains hidden decisions

Even a simple instruction like:

“Make this feel more premium.”

contains dozens of unanswered questions.

Tone
Does premium mean more restrained, more editorial, more luxurious, more established, or more technically sophisticated?
Typography
Should typography change?
Imagery
Should imagery change?
Density
Should density decrease?
Motion
Should motion become subtler?

The correct answer depends on the company, audience, category, positioning, and existing visual system.

Without that context, AI has to infer.

And inference often pushes toward familiar patterns.

Prompt quality isn't the only bottleneck

There's an enormous amount of advice about writing better prompts.

Better prompts absolutely help.

But prompting can become a way of transferring the entire burden of expertise to the user.

If you need to know exactly how a great creative director would solve the problem in order to instruct the AI to solve it, the system hasn't eliminated the need for creative expertise.

It has simply changed the interface.

How much should the system already understand?

A good creative system should occasionally disagree

This may sound counterintuitive.

But a system designed for quality shouldn't blindly follow every creative instruction.

Blind compliance

Do exactly what was requested

“Make every section more visually interesting.” → add more visual complexity everywhere.

Creative judgment

Protect the whole

“The page already has several high-emphasis sections. I'd recommend increasing emphasis here while simplifying the section above it.”

That's not worse instruction-following.

It's better judgment.

Context turns instructions into decisions

The same instruction can have completely different meanings depending on context.

“Make this bolder” for a conservative investment firm is different from “make this bolder” for an entertainment brand.

A creative system should interpret the instruction through what it already knows:

Company
Who is this company?
Audience
Who is the audience?
Brand
What is the brand?
Message
What are we trying to communicate?
History
What decisions have already been made?
Page
What does the rest of the page look like?

Without that context, prompting becomes a sequence of isolated commands.

With it, prompting becomes direction within a larger system.

That's a much more powerful way to work.

Common questions

Frequently asked questions

Short answers about why perfect instruction-following can still produce weak creative work—and why context and judgment matter beyond the prompt.

01

Why can AI follow my instructions and still produce bad design?

Because the instruction may specify an implementation without containing the best solution. AI can satisfy a request exactly while missing page hierarchy, audience needs, brand fit, or the effect that decision has on the rest of the experience.

02

Is instruction-following different from creative judgment?

Yes. Instruction-following asks whether the system did what you requested. Creative judgment asks whether that request was the right move in context and whether the result improves the work as a whole.

03

Should a good AI creative tool ever disagree with me?

Sometimes. If a requested change would weaken hierarchy, break the design system, repeat an overused pattern, or conflict with the stated objective, a quality-oriented system should be able to flag the issue and recommend a better approach.

04

How much context does AI need to make better design decisions?

At minimum it should understand the company, audience, goal, positioning, brand principles, existing page, surrounding sections, and important prior decisions. The relevant context will vary by task.

05

Do better prompts fix instruction-following problems?

They help by clarifying intent and constraints, but they do not replace persistent context or an evaluation layer. For complex creative work, the system needs to know more than whatever happens to be written in the latest prompt.

Morphic

Turn better thinking into better creative work.

Morphic combines company context, creative intelligence, design systems, and AI to help teams create better websites and brand materials—and improve them over time.