Brainstorming Product Ideas
Generating ideas is the easy part. Generating ideas that are meaningfully different from each other and that connect back to the specific root problem you identified is what separates a product thinker from someone who can fill a list. This step is where product taste shows, and interviewers immediately feel its presence or absence.
What interviewers are looking for
This is where interviewers directly assess product taste. They want to see that you can hold a problem in your head and generate solutions that attack it from genuinely different directions, not the same idea described three ways. They also want to see enthusiasm: at least one idea you are actually excited about, not just dutiful ones.
Signals they score here:
- Are your ideas genuinely distinct from each other, or are they variations on the same approach?
- Does each idea connect back to the root problem you identified, rather than to a different problem?
- Do you have at least one idea that shows a longer-term product vision?
- Is your prioritization rationale connected to the north star metric and mission from Step 2?
- Do you show genuine conviction about one direction?
The Product Cut Framework
Most candidates brainstorm within a single dimension. They think of three features, or three UX variations. Because all the ideas fall into the same category, they end up with redundant options that collapse into a single choice anyway, and the interviewer never sees the candidate's range.
Most candidates generate features rather than solutions. "A notification when the dog is distressed" is a feature. "A real-time emotional signal that closes the owner's feedback loop" is a solution. Features are implementations of solutions. You need the solution direction first; otherwise, the features lack coherence.
Product cuts solve this. A cut is a technology or delivery lens you apply to a problem. Each lens produces a different class of idea. Before your interviews, build a short list you can pull from on any question: AI and software, hardware or physical layer, platform or network, ambient or environmental, and moonshot. You do not need to apply every cut to every problem. You need enough to generate genuinely distinct ideas.
Step 1: Anchor to the root problem
This is the setup for everything that follows, and it is the step candidates often skip because they feel ready to brainstorm. Do not skip it. Before generating a single idea, state the root problem in one sentence.
Not the pain point. The root problem underneath it. "Owners are anxious about their dog while they are away" is a pain point. "Owners have no feedback loop, so every care decision is made in the dark" is a root problem. Your solutions must address the root problem. If they only address the surface-level symptom, they will feel thin when the interviewer probes further.
The candidates who generate the best solutions are the ones who treated the pain point step seriously. If you named a specific root cause (not just a frustration), this step almost writes itself. If you find yourself struggling to generate distinct ideas, the problem is usually that you never got past the symptom.
For the OpenAI animal mind-reading prompt, the root problem is: the owner has no signal from their dog while they are away, so every decision about the dog's care, crate or no crate, dog walker or not, which routine actually helps, is made in the dark.
Step 2: Generate ideas across different cuts
Apply three different cuts to that root problem.
Example for the pet owner and separation anxiety:
Cut 1, AI and software: A real-time emotional check-in app. The collar wearable sends continuous emotional state data. The app surfaces a simple read, calm, or distressed, updated throughout the day, with a timeline showing when and how long distress episodes lasted. The owner can see at a glance whether their dog has been calm all morning or whether distress spiked for 40 minutes around noon. This directly closes the feedback loop: owners can now see a signal they have never seen before and start making decisions based on real data rather than guesswork.
Cut 2, behavioral pattern layer: Departure pattern detection. The app learns the emotional signature around the owner's leaving. Does the dog settle quickly, or stay distressed for hours? Over time, it surfaces whether specific departure behaviors (crate, background music, a treat before leaving) are actually helping the dog settle. This moves from "here is the signal" to "here is what the signal means for your specific dog." It turns raw data into actionable care guidance.
Cut 3, moonshot: A conditioned return signal. The dog learns to associate a specific pattern or sound with the owner's return. When the owner is on their way home, they trigger the signal from the app. Not language, a learned cue paired with a positive emotional state. Reduces the dog's uncertainty on the other side. The technology does not need to translate precise intent to make this work; it just needs reliable positive/negative emotional tagging.
These are genuinely different: the first creates a new information channel, the second adds interpretation to that channel, and the third uses the channel to communicate back to the dog.
The fastest test for whether your ideas are genuinely distinct is whether each one can fail independently. If idea B would only fail if idea A also fails, they are not different ideas. Real distinct options have different risk profiles and different success conditions.
Step 3: Prioritize against the metric and mission
Pick one and make the rationale work on three levels: what most directly addresses the root problem, what is feasible given your assumptions about team and timeline, and what best maps to the north star metric you set in Step 2.
For this example, the real-time check-in is the right starting point. It most directly addresses the root problem (no feedback loop), it is fully buildable with vague emotional signals (you do not need precise thought translation, just reliable calm vs. distressed), it creates a daily habit that generates the data you need to build everything else, and it maps to the north star metric (accuracy of emotional state interpretation validated against observed behavior).
Departure detection is compelling but requires enough longitudinal data to establish a per-dog baseline. It is a second-order product. The moonshot is the longer arc. Naming both as "later" and explaining why specifically are part of the signal: they show you are thinking in product stages, not just listing.
Always include a moonshot. Interviewers are not expecting you to build it. They are checking whether you have a longer product arc in mind. A candidate who says, "Here is what this becomes in three years if the signal gets more precise" signals product vision. A candidate who stops at the "safe first" idea signals an execution orientation.
One way to pressure-test your priority choice is to ask: "Does solving this completely move my north star metric?" If the answer is yes, you are on the right track. If the answer is "maybe, eventually," you may be solving a supporting problem instead of the core one.
Leveling Tips
A solid answer generates three ideas, picks the most feasible one, and explains why it addresses the problem directly. The ideas are reasonable and the recommendation is defensible.
A senior+ answer uses different product cuts to generate ideas that are genuinely distinct in approach i.e. three different theories of how to solve the problem.
The candidate explains the conceptual difference between each direction before comparing them, which makes the eventual choice feel argued rather than assumed. The winning idea is connected back to the north star metric and mission from Step 2, not justified on its own terms.
And the deprioritized ideas get treated as a conditional roadmap, with a clear articulation of what would need to be true for each one to become the right call.
Common pitfalls
Three variations of the same idea. "An app, and also a widget, and also a push notification" are not three different solutions. They are three delivery formats for the same idea. Push yourself to find genuinely different approaches by switching product cuts deliberately.
Ideas that do not connect to the root problem. It is easy to drift toward exciting ideas that solve a related but different problem. Before you propose each idea, check: does this attack the root problem I named at the start of this step? If the answer is "sort of," the idea needs to be sharpened or replaced.
Prioritization that ignores the metric. Picking the "most impactful" idea without connecting it to the north star you set in Step 2 makes the prioritization feel arbitrary. Explicitly name the metric and show how the winning idea moves it. This is the connective tissue that makes the whole answer feel coherent rather than modular.