Went through the full Perplexity PM loop earlier this year and got the offer. Been holding off writing this up because I wasn't sure if I'd join, but I ended up declining (comp structure didn't work for my situation) so here's the honest breakdown.
Perplexity PM interviews are not structured the way most tech companies run them. There's no rigid "product design round, strategy round, metrics round" scaffold. It's more blended. Here's what the rounds actually contained.
Recruiter screen (30 min): Background, why Perplexity, real product opinions. They want you to actually use the product and have views.
PM take-home (given 48 hours): Write a product spec for a new Perplexity feature. I got a prompt about improving the research/report generation experience for professionals. This is where a lot of candidates get filtered. They want specificity, real user understanding, and clear tradeoffs. I wrote about 1,200 words. The spec had a problem statement, user segments, three potential approaches with pros/cons, my recommendation, and a rough success metric. They explicitly said "don't over-format it" in the instructions, which was a signal.
PM interview with eng lead (45 min): Walked through my spec. They pushed hard on my metric choice and why I'd prioritize approach X over Y. Then asked a completely different product question: how would you think about Perplexity's position in a world where every browser has native AI search? Real strategic question, no right answer, they wanted to see how I reason.
HM conversation (45 min): Mix of behavioral and strategy. See the behavioral post for flavor. They also asked about a time I worked with a really strong engineer who disagreed with my roadmap, and how I navigated it.
There was no formal metrics round but metrics came up in every conversation. If you can't talk fluently about user retention, query quality, and how you'd measure a qualitative thing like "is this answer trustworthy", you'll struggle.
One thing I noticed: they care that you actually understand how AI products are different. Session-based UX, trust signals, latency perceptions, hallucination awareness. Not because they quiz you on LLMs, but because the product decisions are downstream of understanding those constraints. Generic PM frameworks land flat here.