OpenAI product designer interview practice
Design for answers you can't predict.
Bring a portfolio decision into focus, sketch failure states for ChatGPT, and explain your design choices in spoken practice.
Practice for a senior product designer interview in the style of OpenAI should focus on designing for model output you can't fully predict. Good answers show how a design handles a good answer, a partial one, a wrong one and a declined request, and how people stay in control when an assistant acts for them.
For AI product design roles, a portfolio is stronger when it shows the states beyond the happy path, one clear decision per project with its tradeoff, and testing with real outputs instead of placeholder text. Designers without shipped AI work can show another system with uncertain output, such as search or recommendations.
Framing your portfolio for AI product work
Show the states, not only the happy path.
Model-driven screens don't always land on the ideal result. Put the loading, partial, wrong and declined versions of a screen next to the polished one, and explain which of them took the longest to get right.
Show what you decided and why.
Pick one decision per project and give the alternative you rejected, plus what it would have cost. A reviewer learns more from one honest tradeoff than from a gallery of final screens.
Bring uncertainty from outside AI if you need to.
If you haven't shipped an AI product, choose a system where results were never guaranteed, such as search relevance, recommendations or forecasts. Name the design moves that carry over, like confidence cues or graceful fallbacks, and the parts that would change with a generative model.
Related page on presenting project stories: portfolio walkthrough practice
ChatGPT uncertainty, memory and action-control design prompts
These prompts are about craft, not product strategy: each asks how an interface should behave when the model surprises you. ChatGPT and the API are public products, used as design scenarios in the style of OpenAI interviews, with craft-level notes under every prompt.
Designing for model behavior
1. Show a project where output was unpredictable. Which design choices would carry over to ChatGPT, and which would not?
Show every state: a good result, a partial one, a wrong one, and no result. Explain how you tested with real outputs rather than ideal ones, and which state you underestimated at first.
2. Design how ChatGPT signals that an answer may be wrong, without teaching people to ignore warnings.
Match the signal to the stakes. A recipe and a medication question shouldn't look the same. Consider inline cues and sources over blanket banners, and say how you would measure whether people actually check important answers.
3. The model sometimes declines a request. Design that moment.
Keep it short and specific about what the assistant can do instead, and avoid a lecturing tone. Name the partners you would work with on wording (policy, safety, research), and how you would keep the pattern consistent across products.
Control, safety and privacy
4. An assistant can now take actions for the user, such as sending an email or booking something. Design confirmation and undo.
Sort actions by how reversible they are: preview and confirm anything that can't be undone, and let low-risk actions run with a clear log. Describe how trust could grow over time without the user losing control.
5. Design memory controls so people understand what the assistant remembers about them.
People should be able to see what is stored, change it and delete it, at the moment it matters, not only in settings. Explain your default, and how you would test whether people really understand it, not only whether they click.
New capabilities and developer tools
6. A new capability, such as live voice or image input, is ready before the interface is. What do you design first?
Start from the core loop and its failure states, ship the smallest version that is honest about its limits, and plan for the capability to change under you. Say what you would deliberately leave out of version one.
7. Design a playground where developers compare how different models respond to the same prompt.
The users are developers, so show the details they care about: side-by-side outputs, parameters, latency, cost signals and an easy path to code. Explain how the design serves a quick test and a careful evaluation.
Critique
8. Critique a ChatGPT flow you have used: where does the interface help or hinder your judgment of the output?
Choose a real, specific flow, describe the problem you observed, propose one change and say how you would validate it. Be fair to the constraints the team likely faced.
Research disagreements and design corrections
9. Research partners want a richer ChatGPT response, while engineering wants a simpler loading state. How would you set design direction?
Prototype partial, delayed and complete outputs using real responses. Show what each proposal helps the user understand, where it misleads, and how that evidence could change your preferred design.
10. A memory-control design tested well, but users later misunderstand what deleting an item does. How would you investigate and revise it?
Compare what participants expected with what the interface communicated. Revisit the wording, placement and feedback for deletion, and describe a comprehension test that could expose the same mistake before release.
Prototype with real outputs
Lorem ipsum hides the hardest part of designing for a model. When you prototype, paste in actual responses, including the rambling one, the confidently wrong one and the polite refusal, and watch what the layout does with each. Long answers break cards, short ones look broken, and a refusal can strand the user with nowhere to go.
In a design exercise or a portfolio story, say which real outputs you tried and what changed because of them. That one detail shows you designed for the system as it behaves, not for the demo.
Design scope for model-driven experiences
A Senior Product Designer at OpenAI works on ChatGPT, the API, and the products built directly on frontier models. The underlying capability changes underneath you, sometimes in the middle of a quarter. At this level you are expected to set direction for an area, bring partner teams with you, and be right often enough that people stop asking for the reasoning.
What you'd do
- Design an interface for a capability that is new to the people using it.
- Decide what to expose and what to abstract away when the model can do more than the UI admits.
- Set direction for a product area rather than a single feature, and bring partner teams along with you.
- Raise the craft bar around you: critique, design-system contribution, and mentoring designers earlier in their career.
- Turn an ambiguous problem into work a team can start on Monday.
What they look for
- Comfort with heavy ambiguity, and the taste to invent a pattern rather than borrow one.
- Six or more years designing shipped products.
- At least one piece of work where you can say precisely what you decided, and what you were wrong about.
- A record of influencing scope and strategy, not only execution.
- Comfort presenting to, and disagreeing with, senior stakeholders.
This is a practice interview, not a job posting. ADPList is not hiring for this role and is not recruiting on the company's behalf.
Portfolio and critique perspectives from OpenAI designers
Start with the product designers for a portfolio or critique conversation; the visual design and engineering profiles suit questions about range and measurement. Each summary comes from that person's own profile, and all are experienced mentors at OpenAI.
Kevin Tunc, Member of Design Staff at OpenAI. A designer at OpenAI working on making ChatGPT a genuinely helpful assistant, with experience as an IC, lead and manager. Offers portfolio reviews, product and design critique, design interviews, career direction and the path between IC and leadership, in a direct, practical style.
Chester Hyunil Cho, Senior Product Designer at OpenAI. “I design at OpenAI building something unprecedented but useful through AI.”
Pablo Boerr, Senior Staff Product Designer at OpenAI. Works at OpenAI on complex problems and engaging experiences, specializing in human-centred design, with a belief in collaboration and sharing knowledge.
OpenAI colleagues in visual design and engineering
Ruby Chen, Visual Designer at OpenAI. A member of the Design staff at OpenAI with a mixed background in product design and marketing, happy to mentor people making similar career pivots.
Anna Yoon, Tech Lead, Senior Software Engineer at OpenAI. Focuses on online experimentation at OpenAI, building A/B testing frameworks for LLM applications. Her background suits designers who want to talk through measuring an AI feature.
Community members
More mentors from OpenAI in the ADPList community. They aren't taking bookings through this page, but you can view their profiles to learn about their work.
Portfolio choices for uncertain model behavior
Which model-output states should I include in a ChatGPT design exercise?
Practice designing for model behavior you can't fully predict: answers that may be wrong, requests the model declines, actions taken for the user, and memory people need to control. For each prompt, show the good, partial, wrong and empty states, not only the ideal one.
How should I present my portfolio for AI product design roles?
Lead with the states your design handled beyond the happy path, a single decision per project and the tradeoff behind it, and proof that you tested against real model responses. No shipped AI product yet? Work on any system with uncertain output, search or recommendations for example, still counts if you explain which lessons transfer.
Do I need machine learning experience to interview for product design at an AI company?
You don't need to build models to practice for these interviews, but you should be able to explain how uncertain output changes a design. Practice talking about confidence, error states, user control and how you would measure whether people trust the right answers.
Can I get feedback from designers who work at OpenAI?
Kevin Tunc’s profile lists portfolio reviews and product critique. Chester Hyunil Cho and Pablo Boerr also have product-design profiles linked here. Use those bios to choose a discussion topic and consult the profile for current scheduling options.
Is this page affiliated with OpenAI?
OpenAI has not endorsed this portfolio and design practice resource. Its publisher, ADPList, is not affiliated with OpenAI.
Ready when you are
Company-style practice. OpenAI has not endorsed this portfolio and design practice resource. Its publisher, ADPList, is not affiliated with OpenAI.