2023–2026 Google New Grad Software Engineering OA / Interviews: 196 Experiences (Part 1 :)

A frequency-first review of the technical questions and interview patterns that actually repeated across 196 firsthand Google New Grad and Early Career reports.

Coverage: Full-time Google New Grad and Early Career software-engineering reports published from March 2023 through September 2026. The corpus contains 181 general SWE, 13 MLE, one frontend, and one data-engineering experience. Part 1 covers repeated OA and live-coding questions plus the coding-round behaviors that appeared across many otherwise different prompts.

The short version

The question bank was huge, but the genuinely high-frequency list was short.

1. OA questions repeated more than live-interview questions. One two-question OA pair appeared in nine independent experiences; another pair appeared in five.
2. Restaurant Waitlist was the clearest live-interview repeat. It appeared in seven independent experiences, but the exact matching rule and follow-ups varied.
3. Most live questions appeared only once. The 196 experiences map to 155 distinct FastPrep practice pages, including 145 algorithm questions. Memorizing a short tagged list would not have covered the long tail.
4. The follow-up was often the real question. At least 80 reports explicitly mention a follow-up or changed requirement. Candidates were asked to revise data structures, support updates, reduce memory, handle scale, or explain a different contract.
5. Explanation was part of the signal. At least 37 reports mention communication or alignment, 32 mention complexity analysis, and 13 explicitly mention a dry run.
6. BQ was just as visible as coding. At least 79 experiences mention BQ, behavioral, Googleyness, or Googlyness. Part 2 groups the concrete questions and interview formats that repeated.

How the counts work

The Google-filtered source contained 219 strict full-time New Grad SWE rows. Repeated thread appearances, duplicate application stories, and 19 ineligible sources were removed, leaving 196 independent experiences.

When a question is labeled 9×, the same verified canonical contract appeared in nine independent candidate experiences. Multiple posts from the same application count once. Similar topics with different required behavior stay separate.

There are three useful question totals:

The wider crawl's 231 pages break down into 216 algorithm, 10 System Design, and five Low-Level Design pages. Of those, 141 were brand-new publications from this run: 130 algorithm, six System Design, and five LLD. The rest were existing canonical matches or updates.

High-frequency online-assessment questions

OA 01: Moving tokens and grouping by a common digit — 9× each

These two questions repeatedly appeared together.

Moving tokens

1. The board is represented as a one-dimensional string containing empty cells, tokens, and coins.
2. A token moves exactly three positions to the right and may move more than once.
3. A move is blocked when another token occupies the destination.
4. A coin is collected only when a token lands on it.
5. Return the maximum number of coins that can be collected.

Practice: Maximum Coins With Moving Tokens

Common-digit group

1. The input contains two-digit numbers.
2. Select the largest group whose members all contain one common digit.
3. A repeated digit inside one number still counts that number only once.

Practice: Maximum Elements With a Common Digit

One source displayed a contradictory example for the common-digit task and never resolved it. The practice page keeps the coherent contract instead of silently copying that example.

OA 02: Split and Sort plus Minimum Absolute Sum — 5× each

These two tasks formed the second repeated OA pair.

Split and Sort

Choose a cut between two array elements. Sort the left and right pieces independently, concatenate them, and count the cuts that make the final array nondecreasing. The reported linear approach compares each prefix maximum with the next suffix minimum.

Practice: Split and Sort

Minimum Absolute Sum

Given an integer array, negate at most one element so the absolute value of the total sum is as small as possible. If the original total is S and the chosen value is x, the new total is S - 2x.

Practice: Minimum Absolute Sum

Some reports described a 90-minute assessment, no camera, and the ability to add custom tests. Those details belonged to specific recruiting windows, not one permanent Google-wide rule.

High-frequency live-interview questions

01: Restaurant Waitlist — 7×

Candidates maintained waiting parties and selected a party when a table became available. Reported operations included joining, leaving, and serving the earliest suitable party.

The reports did not preserve one universal matching policy. Some used an exact party-to-table size; others allowed a smaller party to take a larger table. Follow-ups covered deletion, rejoining, duplicate reservations, complexity, concurrency, and how the design might extend to a queueing system.

Practice: Restaurant Waitlist

02: Meeting Rooms II / minimum chairs — 4×

Candidates detected interval overlap or returned the maximum number of simultaneous meetings. The repeated solution family used sorted endpoints or a min-heap of active ending times.

Practice: Minimum Number of Chairs / Meeting Rooms II

03: Count wrapped document lines — 3×

Given text and a width limit, return the number of display lines. Richer versions introduced explicit newlines, whitespace, words longer than the width, or a two-column follow-up.

Practice: Count Lines in a Width-Limited Document

04: Filesystem entity total size — 3×

Model files and directories as a tree and compute the total size below a directory. Follow-ups asked how to answer repeated size queries efficiently or update every affected ancestor when a file changes.

Practice: Filesystem Entity Total Size

05: Top-K IP addresses from a large file — 3×

Read IP-address records, count frequencies, and return the most common addresses. Follow-ups covered ties and a file too large to fit in memory.

Practice: Top-K IP Addresses from File Records

Questions with two independent sightings

Each family below appeared in two independent experiences. Two sightings are useful corroboration, but not enough to call the prompt standard.

Patterns that repeated across different coding questions

Follow-ups changed the contract

At least 80 original-author reports explicitly mention a follow-up or changed requirement. The recurring moves were:

1. Turn a batch problem into an online or streaming problem.
2. Support updates, deletions, or repeated queries.
3. Reduce O(n) extra space or improve an O(n log n) solution.
4. Handle input that no longer fits in memory.
5. Add ordering, tie-breaking, duplicate, concurrency, or fault-tolerance rules.
6. Explain how the design changes when one hidden assumption is removed.

Interviewers checked alignment, not silent coding

At least 37 candidates explicitly discussed clarification, alignment, communication, or thinking aloud. Reports repeatedly describe interviewers recording whether the candidate clarified the input and output, tested examples, explained a pivot, or checked that both sides were solving the same problem.

This did not mean talking continuously. The stronger pattern was short alignment points: restate the contract, name the invariant, explain the chosen structure, and say what changed when a follow-up arrived.

Complexity questions were specific

At least 32 reports mention complexity analysis. Some interviewers wanted more than a single O(n log n) label: they asked what each variable represented, how heap operations contributed, whether result construction added another term, and what happened when the data became too large for memory.

The live bank remained a long tail

Only five live-question families appeared at least three times. Most of the 145 algorithm pages in this strict corpus had one verified sighting. System Design, ML design, and role-specific domain questions were even less repetitive.

That is the main limitation of any Google “high-frequency” list: it identifies the few questions that genuinely repeated, but it does not turn an unusually broad interview bank into a predictable one.

Continue to Part 2 for the BQ/Googleyness question bank, interview formats, follow-up patterns, and process findings.