What candidates say the round looked like, summarised in our words, each linked to where it came from. Google hasn’t confirmed these details unless the source says “Official page”.
Background on the format: our Google guide.
Had this round? Add your report →
Cloud AI engineer role: the role-related knowledge round was spoken questions on AI terms, with most interest in prompt injection and how system instructions affect a model. The onsite coding round was a single array problem; the candidate solved it but wasn’t selected because of bugs in the code.
For a senior AI/ML role at L5, round 1 used Google’s newer format, in two parts. Technical Communication and Collaboration: a deep discussion of one past project, covering design choices, trade-offs, ambiguity, AI tools used and collaboration, with no slides. Then AI-assisted Code Comprehension on CoderPad: debug, extend and optimise a given codebase with an AI assistant.
Replying in a thread about Google’s AI-assisted rounds (the poster was interviewing for L4), a candidate who had one the week before said it was two LeetCode problems, a medium then a hard. The interviewer asked for pseudocode first and said not to touch the AI until it was written. They wrote the whole solution as pseudocode and had the AI translate it to Python in one shot, and felt coding it themselves would have been faster.
First intern round, about an hour: one tree problem with an N-ary follow-up, then the last few minutes on 3 or 4 AI fluency questions. These were spoken questions, not a coding task with AI: how they use AI in their normal workflow, whether they give the AI full ownership of a project, and how they use it for tasks like debugging.
Business Insider reviewed an internal Google document describing a pilot ‘code comprehension’ round: candidates ‘read, debug and optimize’ an existing body of code with Gemini as the only approved assistant. Interviewers assess ‘AI fluency’, described as prompt engineering, output validation and debugging skills. It’s a pilot for junior and mid-level roles on select US teams from the second half of 2026, and Google’s VP of recruiting confirmed it. Candidate posts describe a multi-file codebase of roughly 100 to 500 lines.