# Design Interviews and AI: What They Ask and How to Answer > Design interviews ask how you use AI, what you verify and where it fails. What 2026 guides say and how to answer with stages, criteria and real cases. *Tags: ai, career, interviews, design* --- > [!info] Quick Definition > A **design interview with questions about AI** assesses how you use generative tools inside your process: at which stages, with which criteria, what you verify and what you do when the output is poor. It can also assess how you would design an AI feature for other people. It does not test whether you know a tool's name, but whether your judgment improves the result. ## What do they ask today? There is no official list: each company builds its own process, and the available guides are preparation material, not hiring studies. What follows summarizes four guides consulted in October 2026; take it as a signal, not a rule. **Questions about how you use AI.** A guide published in April 2026 by Evangeline, author of the newsletter "Evangeline's AI Design Garden", says the recruiter screen asks about tools: "How are you using tools like Cursor or Figma Make in your process?" and "What have you made with AI recently?" With the hiring manager, something like "How do you see yourself using AI daily to improve team processes?" comes up. The author describes her own experience, not a survey, and the guide does not say which companies ask this. **Questions about designing AI features.** KORE1 published a question guide for product designers on October 13, 2026. One question describes a tool that extracts vendor, due date, line items and total from a PDF, and gets every field right on 92 of 100 invoices; the candidate must design the review screen. According to the guide, a weak answer adds a sparkle icon, a gray disclaimer and a big green Approve button. A strong one asks which fields do the most harm when wrong, shows the source line next to each value and makes correcting cheaper than retyping. Another list of questions for AI designers, from Final Round AI with no visible date, includes ethical considerations, user testing and explainability. **The exercise.** According to Evangeline's guide, whiteboard prompts have barely changed, but some companies now expect a live prototype or a flow built with AI-generated code, and take-home exercises have grown in scope. For senior roles, KORE1 proposes spending about ten minutes critiquing a flow generated by a prompt-to-UI tool such as Figma Make: strong candidates spot the missing empty state, the error path nobody prompted for and the component that ignores the design system. ## How do you answer a question about your AI workflow? Use a three-part structure: 1. **Stage.** Say which step AI enters: research, exploration, writing, prototyping or review. 2. **Tool and criteria.** Say what you use and what you ask for, for example "I generate five copy variants with a fixed tone and choose two for clarity." 3. **Verification.** Say how you know the result is good and what you corrected. This is the part that gets probed most afterwards. TechInterview, a guide to "built with AI" portfolios aimed mostly at people who ship features built on models, warns that interviewers drill into whichever part you claimed was hard. It lists as red flags "AI does X" claims without evidence, a vague "we used GPT" with no model or evaluation, and inflated capabilities. Its advice is to name the cases where the model failed and what you did about them. Evangeline adds another: arrive with real examples and evidence, and show rather than tell. If you have a number of your own, such as "I cut exploration from three days to one morning", say it. If you do not, do not invent one: an honest, concrete answer weighs more. ## What happens in the design exercise? First ask what is expected about AI. If you can use it, show how you guide it and where you decide. If you cannot, show that you can solve the problem without it. For a live challenge, Evangeline's guide suggests limiting it to one key flow and generating the code for that flow only. In both cases, explain your reasoning out loud, because what is being assessed is judgment. If you are asked to critique an AI-generated flow, look for what KORE1's guide mentions: empty states, error paths and components that ignore the design system. ## How do you show AI in your portfolio? TechInterview describes a good case as a real feature, with a concrete user, where AI is load-bearing: if a regex or a lookup table would do the same job, it does not count. It recommends documenting decisions and trade-offs, having a small evaluation and being candid about failures. That is written for profiles that build products with models; if your role is design, adapt the idea: show a case where AI was part of the solution, what failed and how you fixed it. More ideas in [[portfolio-with-ai-projects|a portfolio with AI projects]]. ## What should you prepare the week before? Prepare three cases, not thirty. For each one, write down on a sheet the problem, the most important decision, the part where AI helped, the part where you corrected it, and the result. If there is no measurable result, say what you would have measured and how. One well-explained case beats a list of projects without context. Rehearse out loud with someone who asks you the questions you dread most. And have ready an example of something AI got wrong that you caught before delivering. ## When should you talk less about AI? When the question is about how you reached a research or strategy conclusion, or about a project where you did not use AI: there, what is assessed is your general judgment. Also when you cannot sustain a technical conversation about what you showed. In those cases, talk about the problem and the decision. ## Example Illustrative example: in a redesign exercise for a payment flow, a candidate uses AI to generate three versions of error messages. When presenting, she says which ones she discarded and why: two blamed the user and one did not say what to do next. Then she shows how she adjusted the prompt with those rules and what tests she ran with five people. What persuades the interviewer is not the tool but the fact that she spotted what was wrong. ## Common mistakes - Answering "I use ChatGPT for everything" without naming the stage or the criteria. - Saying AI did everything and being unable to explain a decision. - Inventing savings figures to impress. - Talking about AI as an end in itself instead of a tool for solving a problem. - Presenting a generated flow without checking empty states, error paths or consistency with the design system. ## Sources - [KORE1, Product Designer Interview Questions 2026 (October 13, 2026)](https://www.kore1.com/product-designer-interview-questions/) - [Evangeline, AI Product Design Interview Guide (April 2026)](https://evangelineng.substack.com/p/ai-product-design-interview-guide) - [TechInterview, The AI Portfolio: What "Built With AI" Means in 2026 Interviews](https://www.techinterview.org/post/3233475399/ai-portfolio-built-with-ai-2026-interviews/) - [Final Round AI, The 25 Most Common AI Designers Interview Questions](https://www.finalroundai.com/blog/ai-designer-interview-questions) --- Source: https://www.fernandoux.com/en/wiki/ai/design-interviews-and-ai/