Updated by Google candidates
Software Engineer Intern Interview Experience
Google’s online assessment took me like 5 minutes and it was supposed to be 90, and that’s when I realized they really prioritize your interactions with the interviewer. You absolutely are not going to run any of your code, so if you don’t understand that landscape, you’re going to fail.
Interview process
I got an OA first, and it was so easy that I finished it in about five minutes even though it was supposed to be ninety, which told me pretty quickly Google was not using that as the real filter. After that I had two technical interviews that were both blank-doc coding rounds where you cannot run code, so the whole thing is about reasoning, asking good questions, and keeping the interviewer engaged while you work. My actual coding questions were both pretty standard LeetCode-medium level, but the follow-up optimizations were where the signal really was, especially when I had to optimize a heap update problem with no hints. Then I went into team matching, which lasted way longer than the technicals and stretched from September into December with multiple team-specific conversations. Those calls were mostly resume-driven, but some were unexpectedly technical, so it felt like the real bottleneck was not passing interviews but getting matched.
- Online assessment
- Technical interview
- Final round
Interview tips
For Google, do not prep like this is a normal coding interview where you just grind until the code works. Practice solving on a blank doc, ask for test cases early, talk through your intent the whole time, and always debug your own code out loud with edge cases before they do it for you. If they give you a streaming problem, think about classes, caching, and helper data structures instead of just a raw for loop. And for team matching, know your resume inside and out and have an actual niche, because being vaguely into ML is not enough.
Company culture
My read was that Google cares way more about how you think and communicate than whether you brute-force the exact right answer immediately. The interviewers seemed to value curiosity, methodical reasoning, and whether I could keep them involved, and I really got the sense that if you get quiet or make it boring, you lose them. There was no recruiter screen in my case, and the OA felt almost trivial, so the real evaluation started once I was talking to humans. Team matching also felt very different from other companies because it seemed heavily resume-driven and maybe AI-matched on the backend, and a lot of strong people probably get stuck there rather than in the technical rounds. It also felt like they were hiring for very specific skill sets, not just generic smart interns.
Questions asked
Overview
Team matching was a whole separate phase and honestly felt like its own process. I had multiple calls from September through December, and they usually would not even tell me the team beforehand. Most of it was resume and behavioral, but some teams got surprisingly technical depending on what they thought my background was.
Question types asked
Specific questions asked
What kind of work are you looking to do?
How do your technical and social experiences match this team’s vibe?
These calls were very team specific, and it felt like the managers did not know much about me beyond an AI-style match from my resume. Since I was not told the team in advance, I had to adapt in real time once they introduced the team and role. I focused on tying my past work, both technical and social, to what they said they needed. The biggest prep here was just knowing my resume inside and out and being completely honest on the form, because if you exaggerate, it will catch you fast.
What parts would you move out of Python?
Why would C++ or C help here?
One team took one of my AI projects and basically asked me to rebuild parts of it in C++ because I had said I had lower-level language experience. That caught me off guard because I expected a normal team match chat, not a technical redesign. I said I’d focus on the lower-level parts like tokenization and other fundamental LLM processing pieces that don’t need Python overhead. We worked through it together, and it felt difficult but collaborative rather than adversarial.
Get full access with a membership, or share your experience to try it free.
