The conversation about AI in UX design has shifted from "should we use it?" to "where does it actually help?" After two years of testing these tools with real clients and real projects, here is an honest assessment.
Where AI Adds Genuine Value
Research Synthesis
Synthesising user interview notes, survey responses, and usability feedback into structured themes is time-consuming and cognitively demanding. AI tools handle the pattern-recognition phase effectively — identifying recurring language, grouping themes, and surfacing outliers. The designer still interprets and prioritises. But the first pass is dramatically faster.
Design Variant Generation
Generating multiple layout options for a section, a component, or a page structure is well-suited to AI tools. The output is rarely production-ready, but it accelerates the divergent thinking phase and surfaces options that might not have been considered. Treat it as an expanded sketchpad, not a finished brief.
The designers getting the most from AI tools are the ones who use them to expand the option space, then apply their own judgment to narrow it. The ones struggling are those who expect AI to make the judgment calls for them.
Where AI Creates Problems
Generic Output at Scale
The more AI tools are used without strong brand constraints, the more design output converges toward a generic aesthetic — clean, competent, and indistinguishable. For organisations where brand distinctiveness matters, unsupervised AI generation is a liability, not an asset. Brand systems and critical review are non-negotiable.
Accessibility Gaps
AI-generated interfaces consistently underperform on accessibility. Colour contrast ratios, touch target sizes, keyboard navigation patterns, and screen reader compatibility require explicit attention that current AI tools do not provide by default. Every AI-assisted design output should go through an accessibility review before it moves to build.
A Practical Starting Point
Map Your Workflow First
Before introducing any AI tool, write down your actual design process step by step. Mark which stages take longer than they should and which require the most client-specific judgment. AI belongs in the slow, low-judgment stages. It does not belong in the strategic and relational ones — regardless of how capable the tool appears.
