Field Notes · Elle Anderson · andersonco.uk

Product sense, the way Meta grades it

Meta invented the product sense interview and the rest of the industry photocopied it. Most candidates prepare for it as a framework recital. Most companies run it without knowing what the original was designed to detect. Both are missing the point, and the point has never mattered more.

Why this one interview took over

Meta started running product sense interviews around 2008, and the format spread because it tested something CVs cannot: whether a person can walk into ambiguity and find the problem worth solving before anyone tells them what it is. The classic Meta loop still pairs it with an execution round and a leadership and drive round, and it is still the one that sinks the most otherwise strong candidates.

It matters more now than it did when it was invented. Building has become cheap. A prototype that took a quarter in 2018 takes an afternoon with AI tooling, which moves the bottleneck from shipping to judgement. When anyone can build anything, the whole game is knowing what to build and why, and product sense is the closest thing we have to an interview for judgement.

What product sense is for when nobody is interviewing you

Product sense is not an interview trick; the interview is a compressed simulation of a working skill, and the skill is the job. It is what you reach for when someone slides a proposal across the table and asks "should we build this?", when a founder pitch or an acquisition target needs a fast read, and when the board asks whether an adjacent market is worth entering. Does this product make sense: is there a specific person with a problem severe and frequent enough that they would change behaviour to solve it, and are we the right ones to fix it? If you cannot name the person, the pain, and the reason it is still unsolved, the product does not make sense yet, however good the demo looks.

The same judgement, pointed at a market instead of a user, answers the disruption question. Look for where incumbents overserve, leaving room for something simpler and cheaper that is good enough, or underserve, leaving room for something focused and premium. Then find the structural reason they cannot respond: a margin they will not cannibalise, a channel they cannot abandon, a legacy stack they cannot rewrite. Add the unfair advantage you would bring and the why-now trigger, a cost curve, a regulation, a behaviour shift. A market with no structural incumbent weakness and no why-now is not a disruption opportunity; it is an expensive way to fund the incumbent's roadmap. We test all of this in hiring under time pressure because time pressure is the one condition where judgement cannot hide behind a research backlog.

What actually gets scored

Interviewers are not grading your framework. They are listening for a handful of signals, and at Meta each interviewer scores them against a standard rubric after every conversation, with a plain-language verdict on top. The signals are consistent wherever the format has spread: whether you picture a real person before you design anything, whether you can break an ambiguous prompt into a navigable structure without being led, whether your ideas show taste rather than a list of features, whether you connect the answer to the mission and the business, and whether the whole thing was a pleasure to follow.

Internally, Meta teaches its PMs a three-word loop: understand the people problem, identify the best first way to attack it, execute it well. The product sense interview is a test of the first two words. If your answer spends thirty minutes on solutions and ninety seconds on the problem, you have answered a different question than the one being scored.

There is a classic trap inside the audience step. Prompts about marketplaces and platforms have more than one side, and candidates who design beautifully for creators while forgetting consumers, or riders while forgetting drivers, fail a test they never noticed was set. Name the ecosystem, then choose, and say why out loud.

The shape of a strong thirty-five minutes

Strong answers share a spine, whatever words you hang on it. Clarify the prompt and its constraints in a minute, not ten. Anchor to the mission, because "why would this company do this at all" is a scored question. Lay out the people in the ecosystem, choose a segment, and defend the choice. Go deep on that segment's pain points, then prioritise them with a stated rationale rather than a shrug. Only then generate solutions, at least one of which the interviewer has not heard this week. Choose one, be honest about its trade-offs and risks, define what success looks like in a number, and land the plane with a summary that would survive being repeated in the debrief. The interviewer should feel driven, not dragged.

And when the prompt lands cold, buy the first minute honestly rather than talking fluently in no particular direction: restate the prompt, ask the one clarifying question whose answer would change your approach, then take thirty seconds to structure. Run a pocket checklist silently, five questions, one thumb each: who is this for, what is broken for them today, why has nobody fixed it, why would this company be the one, what number would prove it worked. Then say the map out loud before you walk it, and if the prompt is a market-entry question, swap the segment-and-pains passage for the market-structure lens above. The full on-the-fly method, with the practice prompts to drill it, is in the toolkit version.

The 2026 wrinkle: AI in the room

Meta has added a product sense with AI round for parts of its PM organisation: roughly half a session of classic product sense, after which the candidate is handed an AI tool and asked to prototype the idea live, directing the model rather than writing the code. Candidates assume they are being graded on prompt craft. They are not. They are being graded on whether they can tell when the AI's output is generic, wrong, or misaligned with the constraints they set five minutes earlier, and on whether they stay in strategic control of a very confident collaborator. That is the same judgement test wearing new clothes, and other companies are already copying it, just as they copied the original.

Where people lose the room

The failure modes are boringly consistent. Jumping to solutions before the problem has been earned. Reciting a framework so audibly that the structure becomes the content. Designing for "users" in the abstract rather than a person you could describe. Offering three safe ideas and no brave one. Never saying what you would not do. And, increasingly, offering ideas with the smooth, interchangeable texture of AI-generated brainstorming, which interviewers have become extremely good at spotting.

If you run hiring loops, steal the format properly

Most companies borrowed the question and skipped the machinery, and the machinery is what makes it work: a written rubric per signal, scores committed independently after each interview before anyone confers, and a summary verdict in plain words. Without that, product sense interviews collapse into vibes, and vibes hire people who interview like the panel. If you are going to use Meta's exam, use Meta's marking scheme.