All posts

April 15, 2026

Why AI-Generated UI Should Use Real Copy, Not Lorem Ipsum

Lorem ipsum is a design lie. It hides the most common source of layout failure, produces designs that generate useless feedback, and obscures the relationship between content and structure that determines whether a product works. Here's what changes when you design with real words.

Lorem ipsum dolor sit amet. Consectetur adipiscing elit. Sed do eiusmod tempor incididunt.

Nobody reads this. Nobody is meant to. That's the point — it's neutral filler, shaped like words, designed to let you evaluate a layout without the content getting in the way. Designers have used it for centuries, literally: the scrambled Latin text traces to a 16th-century typographer.

It has always been a workaround. In the era of AI-generated UI, it becomes a structural problem.

What lorem ipsum hides

The conventional defence of placeholder text is that it keeps evaluators focused on the design, not the words. This is true for a narrow set of design questions — grid structure, spatial rhythm, type scale relationships. For almost every other design question that matters, lorem ipsum is actively counterproductive, because it removes the information the design needs to do its job.

It hides layout failures caused by real content. A headline that fits in the allocated space when it's eight letters of gibberish might not fit when it's an actual product feature name. Navigation items that look well-spaced in placeholder mode might collide when labelled with real terminology. A card that looks clean with a single line of lorem ipsum might break with the two-sentence description that the actual product uses. None of these failures are visible until the real content arrives — and by then, the layout has been approved, the design has moved forward, and the fix is rework.

It makes stakeholder feedback useless. When a stakeholder looks at a screen full of placeholder text, they can react to the visual structure but not to the product logic. They can't tell whether the information hierarchy makes sense for the content, because there is no content. They can't notice that the empty state message sounds wrong, because there's no message. They can't spot that the call-to-action button label is ambiguous, because it isn't labelled. The feedback they give is abstract, and abstract feedback produces abstract revisions that don't address the real problems.

It disconnects layout from content strategy. Content-first design research is unambiguous on this: placeholder text leads to layouts that don't fit real copy, marketing teams that cram long headlines into tight spaces designed for filler, and product copy that has to be cut or rewritten to fit a design it was never consulted on. The layout and the content should be designed together, because each constrains the other. Lorem ipsum is an explicit decision to design the layout without the content — and pay the integration cost later.

What real copy reveals

When you design with real words, a different class of problems becomes visible — earlier, when they're cheaper to fix.

A button that says "Submit" is vague; a button that says "Confirm and charge card" is specific. The specific label reveals a design question that "Submit" concealed: is there enough visual emphasis on the financial consequence of this action? Should there be a summary of what's being confirmed? The label's specificity creates feedback that "Submit" never would.

An empty state that says "No items" is a missed opportunity; an empty state that says "You haven't added any team members yet. Invite your first teammate to get started." is a content design decision. Whether that message is the right one, whether the CTA placement is right, whether the tone matches the rest of the product — these are questions you can only ask when the content exists.

Nielsen Norman Group's research on usability testing with real vs. placeholder content confirms the mechanism: when users can read the content, they give feedback about the copy and how they feel about it, which pushes teams to reflect on the type of messages communicated and the tone of voice that makes people trust a platform. That feedback is impossible to collect from lorem ipsum.

The practical consequence is that real copy makes the review cycle useful. Stakeholders who see real labels, real empty states, real error messages, real CTAs are evaluating a product, not a visual structure. Their responses are specific and actionable. The designs that come out of those conversations are better, because the feedback was real.

The spec-to-copy pipeline

The conventional reason design uses placeholder text is practical: the copy isn't written yet. Design and content are separate workstreams, and design usually runs ahead.

Spec-driven generation breaks this dependency. When the product spec includes real information — actual feature names, real user flow descriptions, actual product terminology — the generation system can derive real copy from it. Not polished, launch-ready marketing copy. But real content structure: real labels, real messages, real calls to action, real empty-state descriptions.

This is what Mowgli does by default. The questionnaire extracts the product's actual vocabulary — what the core features are called, what the user types are called, what the key actions are. The generation uses this vocabulary throughout. Navigation items use the product's actual terms. Empty states describe the actual missing content. Buttons have real labels.

The result is a design where the copy and the layout were designed simultaneously, from the same source of truth. Layout decisions account for actual content lengths. Content decisions account for actual space constraints. The two are coherent with each other, because they were never separated.

Why this matters for AI design specifically

The lorem ipsum problem is amplified in AI-generated UI because the generation happens at speed and at scale. A human designer working on a single screen might manually insert real labels because the effort is low. An AI generating 30 screens fills all 30 with placeholder text unless the generation is explicitly driven by real content from a structured source.

The result, with most AI design tools, is 30 screens of plausible-looking but content-free layouts. They look like a product. They aren't one. When you try to review them — to decide whether the flows are right, whether the hierarchy works, whether anything is missing — the absence of real content makes the review partial. You're approving a structure without being able to evaluate whether the structure fits the actual content that will live in it.

Real copy is not a finishing step. It's the test of whether the design works.

Sources