# I Conducted a UX Review with AI: Here's What Happened Can AI replace UX experts? Insights from a real-world project experience. - **Source:** [UX Collective - Medium](https://brifing.si/source/5c6bdc88-e150-4203-af43-069801e37dd9) - **Published:** 2026-07-21 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/1889a84f-7226-43ec-95ec-0a95b0a56cb8 ## AI's Role in UX Review While conducting a UX review for a portal's search, results, and product detail pages, I tested Claude and ChatGPT models. My goal was to understand if AI could replace human experts in such tasks. I provided all data (target audience info, Google Analytics data, heatmaps) to the models. ## Methodology and Process I performed the analysis through a structured, step-by-step process rather than a single prompt. I asked models to support their findings with GA and Clarity data. In parallel, I conducted my manual analysis to create a comparison benchmark. ## AI's Findings Both models produced analyses but included irrelevant and non-existent problems. Claude outperformed ChatGPT in phrasing and structure. Some AI-identified problems helped me catch issues I'd missed. ## Comparative Results A 2025 study showed GPT-4o detected only 21% of usability problems identified by experts. My experience confirmed this: AI is excellent for initial reviews but can't replace expert evaluation. ## Collaboration Model I combined my observations with Claude to create the final analysis. AI proved extremely useful for text editing and consistency checks. The process was approximately 15% faster than fully manual work. ## AI's Limitations The core limitation isn't data access but judgment capability. AI struggles with understanding context, weighing problem severity, and distinguishing real issues from noise. I don't believe live browsing would overcome these limits. ## Future Implications AI will eventually replace our jobs long-term. Currently, the most productive approach is developing specialized agents that encode our expertise. Baymard's 95% accuracy came from models grounded in UX knowledge bases, not better prompts. ## Sources covering this story - https://uxdesign.cc/i-handed-a-ux-review-over-to-ai-heres-what-happened-0511e73e4ffe?source=rss----138adf9c44c---4 — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._