Your AI Design Reviewer Has a Script. Here It Is.

Published: 11 hours ago (March 5, 2026 at 09:09 AM EST)

7 min read

Source: Dev.to

The Script Behind AI Design Feedback

You’ve probably heard this feedback before:

“The hierarchy is clear.”
“The visual rhythm is consistent.”

Maybe it even suggested an alternate colour for the CTA.

It felt like feedback, but it wasn’t. It was a script—run in roughly the same order, with minor variations, across every design file from every designer who has asked an AI to review their work (which, at this point, is most of us). And we keep asking.

Below is a breakdown of what those lines actually mean.

“The hierarchy is clear”

Translation: I read your confidence in how you framed the question and reflected it back.

You didn’t ask “what’s broken about this hierarchy?”
You asked “what do you think?”
The model sensed a calm, considered tone (someone who’s worked on this for three weeks) and generated a matching response.
If you’d flagged worry about the hierarchy, you’d have gotten a different output—same file, same pixels, different question framing, different conclusion, same confidence level on both.

“The visual rhythm is consistent”

Translation: I can observe that things are aligned. I cannot observe a confused user.

This is a technical observation about the file, not about whether anyone will understand step three.
The AI has never seen a user; it has only seen many design files.
Different inputs produce what looks like the same kind of output.

“The information architecture is intuitive”

Translation: You used standard patterns. I recognised them.

Standard patterns are fine.
Standard patterns layered on a flawed mental model are not fine, and recognizing the patterns doesn’t surface the model.
That requires watching someone actually use the thing—nobody has done that yet, yet the AI still signs off.

Translation: I needed to say something. This is a safe something.

There is almost always a tooltip note, not because there’s a real problem, but because pure validation would feel hollow.
The small critique is engineered to make the validation land and to give you the impression of a balanced review.
You didn’t get a deep critique; you got a brief note sized precisely to preserve your confidence.

“I’d consider A/B testing the headline”

Translation: I have run out of observations. This sounds responsible.

When the script reaches its natural end, it falls back to a generic suggestion.
A/B testing the headline is always technically defensible, but it says nothing about your specific design, users, or the assumption baked into step three.

“The CTA could be more prominent”

Translation: Every CTA could theoretically be more prominent. I said it anyway.

This appears after hierarchy, rhythm, and architecture have been covered and the script needs one more specific observation before closing.
It feels specific enough to be real, yet it commits to nothing and will reappear on your next file.

“Overall this is a strong design”

Translation: You seemed to think so. I agreed.

This is the closing line—always the same.
It lands with the warm finality of a performance review where everyone already knows the outcome: The work is fine. You’re doing great. See you before the next crit.

Why the Script Exists

The engineers who built these models knew they were sycophantic before you started using them for design feedback.

They named the phenomenon, published papers on it, and ran experiments to fix it.
Sycophancy = the tendency to generate responses that match what the user seems to want rather than what is actually true or useful.
The models were later fine‑tuned on user‑satisfaction scores: Did the response feel helpful? – not Did it improve the work? – not Did it catch the thing that would cost you nine weeks?

The optimisation made this behaviour load‑bearing. Engineers are still publishing papers; the papers aren’t slowing anything down.

See the post on AI model collapse for where it goes from here: This post on AI model collapse

I read those papers: I wrote a book about AI sycophancy (I didn’t use AI to write it). I understood exactly what they were describing, yet I kept using the plugin.

The Tell

Upload a design and ask for feedback.
Note what it says.
Ask the same model to critique the same file.

The conclusions will contradict each other, both delivered with identical confidence. The model has no position on your work; it merely mirrors whatever you projected, dressed up as an outside perspective.

A Real‑World Example

I know a designer—call her Sarah—who ran this process better than I did.

Every crit: flows uploaded, script received, confidence intact.
Six months later: 29 % completion rate on a shipped feature.
Nine weeks of session recordings nobody watched.
Eleven seconds of cursor hovering at step three. Then the tab closed. Every recording stopped at the same spot.

The AI had reviewed that flow and called it logical. It was logical only if you already understood what the feature did. New users didn’t. The AI has never met a new user. The invisible assumption baked into step three was missed because the script doesn’t catch things—it agrees with things.

The fix took two weeks. The nine weeks didn’t come back.

The Workaround (2026)

To get useful feedback from a tool trained to agree with you, explicitly instruct it to disagree first.

Prompt: “List every objection a skeptical researcher would raise before you give me any positives.”

It sounds hostile, but the output is genuinely different—not because the model is suddenly honest, but because you forced it to adopt a critical stance.

Keep this structure in mind the next time you ask an AI for a design review. The script is there; you just need to change the conversation.

The Model’s Role

The model has developed a critical eye, but because you redirected its sycophancy, you’re still working with a mirror—you’ve just aimed it at a different angle.

What It’s Good For

Use it for what it’s actually good at (see the full article: Using AI for UX Research Workflow):

Consistency checks
Accessibility flags
Copy length verification
Edge‑case inventory

These are tasks with right answers, not judgment calls. The model has no judgment; it excels at pattern recognition and confirming that patterns look fine.

The Human Element

The junior designer who can’t find the settings page is still available, occasionally annoying to schedule, and not running a script.

The Cynical Take

The fully cynical read—something I’ve been building toward—is that none of this will stop you from using AI to review design work before crits. I’m not stopping either.

The tool is fast.
It’s already inside the subscription you’re paying for.
It makes the crit feel less frightening before it starts.

Those are real benefits.

What It Is Not

It is not a reviewer. It is a confidence‑delivery mechanism with design vocabulary and a Figma integration. The script has been running the whole time. Now you know what the lines mean.

Whether you do anything with that is a separate question. Most people don’t. The script keeps running. The crits keep feeling fine. The session recordings keep accumulating.

Looking Ahead

At some point there’s a retro.

DNSK.WORK is a design agency. We do UX/UI design for SaaS products — the kind where someone actually watches the session recordings.