AI can now generate entire interfaces’ worth of text in seconds. But how do we know if the words are actually good? This talk is for designers, product managers, and anyone else who already uses AI to write UX content and needs a clearer way to judge what it produces. It offers a simple, practical framework for evaluating AI-generated UX content based on five key principles: clarity, next-step guidance, reassurance, user language, and value communication.

Through real product examples, Michal will show how to spot unclear or confusing copy, refine tone and intent, and build teams’ confidence in integrating AI tools without losing the human touch. She will then show how to decide which parts of the experience deserve extra human attention, and how to use lightweight or more in-depth testing methods to check if the words actually work. Attendees will leave with a realistic way to combine AI speed with human judgment in their UX content generation workflows.

First done at the Webexpo conference in Prague in 2026, this session received excellent feedback with a score of 4.75 out of 5. A recording can be found here: https://webexpo.net/prague2026/sessions/dont-trust-the-bot-a-human-framework-for-wvaluating-ai-copy/

Some of the feedback received:
"one of the best talks so far wow"
"👏👏👏 So gooood! So much knowledge and practical tips! Really loved it."
"I liked the reminder that we do have our own brains and jugdement to decide whether something is good or bad. sometimes you might forget haha"
"Practical with great examples!"
"very informative. It's good to have to re-learn things."