I have interviewed twice with Google, at two locations (Europe and HQ), so I will combine the results of both interviews. As noted here, the process usually involves three steps:

• Initial contact call (possible light technical questions)
• 1-2 VoIP technical screenings
• On-site 5 x 45-minute sessions with a full interview day, including lunch

Most reviewers focus on steps 2 and 3, but it's worth remembering that step 1 might include a few scripted technical questions. It is certainly possible a strong candidate could be filtered out at this stage if they make the mistake of coming into the call completely cold and blanking on something not entirely trivial, such as the worst-case scenarios of common sorting algorithms.

I also think there is a large gap between the difficulty of the pre-screening and on-site stages. This may well lull you into a false sense of security. Additionally, the difficulty of the tasks on the day can vary quite a lot. Some of my domain-specific questions were trivial, or at least seemed so because I am, unfortunately, highly specialized.

You will, of course, get plenty of low-level algorithm questions, including some dressed up to conceal the textbook techniques that apply. Also, it is highly likely that you will be thrown a curveball or two; in my case, I did receive a reasonably "mathematics" question. Unfortunately, your ability to handle the pressure of the situation will have a large bearing on your success. That math question, for example, coming at the end of the day, was a real killer. As a mature (30s) candidate who was an A student at pure math in his teens, it was disheartening to be flummoxed by something like this.

I also had a design session the second time, which is a positive because as a mature applicant—although with "only" 6 years in the industry—I would be hoping to be in lead roles. I received good feedback from the recruiter, actually, since I made the blunder of not using a very top-down approach. Perhaps it is easy to make this mistake, though, since a session like this is such a gear change from the rest, which will be decidedly bottom-up.

Organization of the process was fairly good. I know the recruiter had some issues organizing the interviewers, which caused things to be finalized quite last minute. My second interview was moved closer to home as a result, which was actually a benefit. Interviewers were mostly well-prepared and professional. The "vibe" in the interview is, of course, highly dependent on your performance. Sessions that go badly for you are likely to feel much less comfortable.

I would put myself somewhat in the skeptic camp with regards to these kinds of technical interview processes. I have also interviewed at Facebook, where the process is very similar (although somewhat less academically oriented). While I understand that these companies have their reasons for using this kind of process, it will be quite disheartening to a large number of strong candidates who hit that false-negative filter. In my own case, after quite a few rejections trying to join a US tech giant through the coding avenue, I have to question whether it is even worth persisting. My suspicion is that these roles are for a "certain kind of people," are skewed towards recent graduates and coding competition hobbyists, and "aren't for people like me." Personally, I have to consider carefully if it is worth pursuing future attempts, including constantly reviewing coding-challenge-style material in my spare time over a period of 2-3 years now.

Broadly, from what I remember:

• Implemented some simple tree recursion, including indentation.
• Implemented a cached object factory.
• Optimized search for subsets within both the source and target datasets. This was based around a bingo game where both the "called numbers" and the player game cards would need to be processed for efficient lookup. It was a quite concealed subset generation question that I should have been able to answer well.
• Re-implemented a well-known view component in my specialist OS.
• Serialized and deserialized trees with extreme space optimization.
• Participated in a design session for a simple two-player turn-based game.