For UX and Product Designers
Protecting Your Judgement: A UX Designers's Guide to Using AI Without Losing Your Edge
AI tools now summarise your research, suggest design patterns, and generate prototypes faster than you can read them. The speed feels like progress until you realise you are designing for the AI's model of your users rather than the real people sitting across from you. Your job is to stay the person in the room who knows when a contradiction in the data points to something true.
These are suggestions. Your situation will differ. Use what is useful.
Read Your Raw Research Before the AI Does
When you paste interview transcripts or survey responses into Dovetail AI or ChatGPT, the tool finds patterns it recognises. It removes the friction. It smooths over the moments when a user says one thing but means another. These contradictions are where real insights live. Read your notes first. Write your own initial observations. Only then use the AI to check whether you missed something.
- ›Spend 30 minutes with raw transcript quotes before opening Dovetail AI's summary
- ›Mark the moments that surprised you or did not fit the pattern before asking AI to find themes
- ›When AI suggests a user behaviour pattern, go back to the original quote and ask why the person actually said it
Question Every Design Pattern AI Suggests
Figma AI and Adobe Firefly can show you how others solved similar problems. That is useful. But a pattern is not a principle. A dark mode toggle works for one product because of their specific user base, their technical constraints, their brand. When AI recommends a pattern, you need to understand why it works there before you decide whether it works here. Skipping this step means you are borrowing someone else's solution without knowing whether it solves your problem.
- ›When AI suggests a design pattern, ask yourself what user need it actually addresses in your research
- ›Prototype the pattern AND your own alternative before choosing between them
- ›Document why you chose the pattern so you can defend it to stakeholders without saying 'AI recommended it'
Stay in the Ambiguity Until You Understand It
Research often leaves you with contradictions. Some users want simplicity. Others want control. Some need onboarding. Some want to jump straight in. AI will try to resolve this tension by finding a compromise or picking the largest group. But the real solution often comes from understanding what makes these groups different and designing something that serves both. This takes time. It requires sitting with the problem longer than an AI summary allows.
- ›When Maze AI or your research tool flags conflicting user needs, resist the urge to average them together
- ›Create separate user journeys for the conflicting groups and look for the moment where their needs diverge
- ›Ask yourself what deeper need each group has before you assume they need different solutions
Use AI to Fill Gaps in Your Thinking, Not Replace It
ChatGPT is useful for generating questions you might ask in follow-up research or exploring edge cases you had not considered. It is not useful for deciding whether something matters. Use AI to expand your thinking, not to replace the messy work of sitting with data. Ask it to play devil's advocate on your assumptions. Ask it what you might be missing. Then decide whether it is right.
- ›When using ChatGPT for research synthesis, ask it to argue against your current hypothesis, then evaluate the argument yourself
- ›Use Figma AI to generate multiple layout options, then choose based on your user research, not on which looks polished
- ›Ask AI 'what would this interaction look like for someone with motor impairments' before finalising accessibility decisions yourself
Keep the Skill of Noticing What Users Cannot Say
The strongest part of your job is empathic imagination. You notice when a user struggles with a task they say they like. You hear what people need even when they cannot name it. You see the gesture or hesitation that reveals the real problem. AI cannot do this. It works with what was said. When you let AI tools think for you, this skill atrophies. Protect it by conducting some research without recording it into tools. Sit with users. Write your own notes. Trust the thinking that happens when you are not transcribing.
- ›Do one round of research per project where you take notes by hand and write synthesis without AI input
- ›Share raw observations with your team before showing them the AI summary
- ›When prototyping, test with users before AI generates variations, so you know what questions matter
Key principles
- 1.Read the raw data first, then ask the AI what you might have missed, rather than the other way around.
- 2.Understand why a pattern works elsewhere before you use it here.
- 3.Contradictions in research are not problems to solve but information to understand.
- 4.Let the AI expand your options but make the actual decision based on what your users need.
- 5.Keep doing some thinking without tools so you stay the kind of designer who asks better questions than any AI can generate.
Key reminders
- When Dovetail AI or ChatGPT summarises your research, highlight the three quotes that surprised you most and ask the AI to explain them before accepting its overall theme
- Before you use Figma AI to generate layout options, sketch three directions by hand so you know what matters to you independent of what the tool suggests
- Create a 'research notes' document that lives outside your design tools, where you write observations that do not fit neatly into any category
- Share contradictory user feedback with your team as a design problem to solve together rather than as noise to filter out
- Test one prototype with real users before asking Adobe Firefly for variations, so you know which aspects need iteration