iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AI adoption is widespread among surveyed product builders, but the available evidence does not show that bad design overall is increasing. The distinction matters: a polished-looking screen can still be hard to use, inaccessible, off-brand, or unlike the intended design. A Figma-to-code comparison is useful only when it spells out what was compared and how quality was judged.
My impression that I am seeing more bad design in the AI era is an observation, not a measured industry trend. The studies below help explain why attractive output can fall short—and why generating a screen quickly is not the same as producing a finished interface.
AI adoption is not evidence of a rise in bad design
Figma’s 2024 survey polled nearly 1,800 designers and developers across four continents. Its FAQ says 59% of respondents were already using AI at work. Among those who said they used AI, fewer than half had launched anything; among respondents who said they had shipped an AI feature, only one-third said they were proud of it. These are survey responses about adoption and sentiment, not audits of interface quality or evidence that bad design is becoming more common. Figma’s 2024 AI report
Figma’s 2025 report landing page describes a second annual survey of 2,500 product builders across seven countries. It identifies topics including agentic AI and differences in designer and developer workflows, but its detailed results were gated on the page reviewed. The published summary therefore does not establish whether product quality improved or declined. Figma’s 2025 AI report
#1 Best Overall
What counts as “bad design” depends on what you test
“Looks good” is only one possible measure. Visual appeal, task support, usability, accessibility, fidelity to a reference, and consistency with a design system are distinct qualities. A screen can score well on one and poorly on another; any comparison needs to name its criteria rather than compress them into an undefined quality score.
- Visual appeal: Does the interface feel coherent and attractive?
- Pragmatic quality: Can people understand it and complete the intended task?
- Fidelity: Does it preserve the reference layout, interactions, and responsive behavior?
- Accessibility and legibility: Can people perceive and operate the interface, including its text and controls?
- System consistency: Does it follow the product’s components, tokens, and guidelines?
One interface study found a gap between appeal and evaluation
A study summarized by the Chartered Institute of Ergonomics and Human Factors (CIEHF) tested burger-ordering interfaces generated with Midjourney, DALL-E 3, and Stable Diffusion 3. The initial outputs had problems with legible text and following prompts; after prompt adjustments, DALL-E 3 and Stable Diffusion 3 produced designs that met the brief.
In a survey of 32 participants, those designs were compared with commercial products and work by eight competent human UI designers. The study reported no difference in pragmatic quality, while AI-generated designs received significantly higher hedonic ratings—the study’s measure of appeal—than the commercial or human-designed examples. The commercial apps had the lowest ratings across the measures used. This is evidence about one burger-ordering task, not a verdict on every interface or today’s AI models. CIEHF’s study summary, May 23, 2025
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The researchers also tested AI evaluation using prompts based on the User Experience Questionnaire Short (UEQ-S). They reported: “We found little correlation between the ratings of the gen-AI apps and human raters.” In other words, a model’s assessment should not be treated as a substitute for human evaluation just because it produces confident-sounding judgments.
Fast generation still leaves fidelity and refinement work
A 2026 peer-reviewed study by John Bustamante-Orejuela, Xavier Quiñonez-Ku, and Pablo Pico-Valencia asked undergraduate IT engineering students to recreate mobile interfaces modeled on Duolingo using Figma, Uizard, Visily, and Stitch. On the System Usability Scale (SUS), the study reported these scores:
| Tool | SUS score in the study |
|---|---|
| Figma | 82.86 |
| Uizard | 67.14 |
| Visily | 78.57 |
| Stitch | 80.36 |
These are results from that study’s specific recreation task, not a universal product ranking. The authors found that all four tools supported rapid generation, while usability, structural fidelity, and perceived control differed. Uizard and Visily offered strong automation and faster initial generation in the study, but required more manual refinement for higher fidelity and customization. Their conclusion emphasizes the role of user control, responsiveness, and iterative refinement: “However, their effectiveness appears closely related to the degree of user control, responsiveness, and the ability to iteratively refine AI-generated interface components.” The 2026 study
Figma-to-code comparisons need more than a screenshot
A Figma file and a rendered page can look similar in one viewport while differing in behavior, responsiveness, accessibility, or implementation quality. A useful comparison separates what was directly observed from what was inferred, and checks the dimensions that matter to the product rather than relying on a single visual impression.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11- Reference and constraints: Which Figma frames, components, tokens, and interaction rules were supplied?
- Generation route: What tool, model, prompt, and contextual information produced the code?
- Visual fidelity: How closely did layout, typography, spacing, and assets match across the intended screen sizes?
- Behavior: Did navigation, controls, and task flows work as designed?
- Accessibility: Were text legibility, keyboard operation, semantics, and other relevant criteria actually checked?
- System adherence: Did the result use the product’s components and design tokens, or introduce inconsistent alternatives?
- Human correction: What needed manual repair, and how much iteration was needed before the result met the brief?
Unless those details are documented, a particular Figma-to-code experiment cannot establish which tools were used, what they produced, or whether AI improved or worsened the result. The studies discussed here provide context for evaluating such an experiment; they do not supply its method or findings.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Design-system compliance is its own engineering problem
Google Research’s 2024 UI-linting case study addresses the practical work of finding and correcting violations of design-system guidelines. It describes a hybrid pipeline that combines deterministic heuristics with the flexibility of large language models (LLMs). Its stated lesson is that AI alone was not sufficient for practical adoption and that a user-centered understanding of the problem mattered. That supports treating component, token, and guideline compliance as explicit checks—not assuming that a visually convincing output is consistent with a product’s system. It does not prove that every AI workflow fails those checks. Google Research’s UI-linting case study
Connecting design context to AI is promising, not a guarantee
Google Research’s PromptInfuser study examined a Figma widget that connected UI elements to LLM prompt inputs and outputs to create semi-functional mockups. Fourteen professional designers took part. Participants felt the connected workflow communicated a product idea better, stayed closer to the envisioned artifact, and helped them anticipate interface issues and technical constraints compared with their disconnected workflow. Those findings concern design-context flow and mockups; they do not establish that the approach automatically produces production-ready code. Google Research’s PromptInfuser publication
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Recommended Free Tools
