The State of Data Science Interviews (2026): What 389 Real Questions Reveal About What Companies Actually Test

Methodology

Every table in this report comes from DataScienceHired's own problem bank: 389 published interview questions, tagged across 49 real companies and 32 topics, and cross-referenced against public interview reports and candidate write-ups rather than any single company's official materials. We didn't survey candidates or scrape a third-party question bank for this report — we counted what's actually sitting in our own database, question by question, topic by topic, company by company.

The numbers below are as of August 29, 2026. This bank grows every week as we add newly reported questions, so if you're reading this later and the counts look slightly different, that's expected — the underlying methodology (real, published questions only, tagged by topic and company) hasn't changed, only the sample size.

The Numbers at a Glance

• 389 published interview questions
• 49 companies with at least one tagged question
• 32 distinct topics covered
• 239 conceptual questions (61%)
• 100 SQL questions (26%)
• 50 Python questions (13%)

Conceptual questions — explain-a-concept or whiteboard-reasoning prompts rather than code-writing exercises — make up the majority of the bank, but the SQL and Python splits below tell a more useful story: SQL alone is roughly a quarter of everything a candidate might face, and it's concentrated in a much smaller set of topics than the conceptual side.

What Gets Tested Most: Full Topic Breakdown

Machine learning fundamentals is the single largest topic in the entire bank, with statistics, llms, and neural-networks tied for second at 35 questions each. Together, those four topics account for 142 of the 389 questions — roughly 37% of everything asked.

The long tail is notable on its own: 11 of the 32 topics have four questions or fewer. Interview prep that chases breadth across every possible topic is wasted effort — the top eight topics above cover more ground than the bottom twenty-four combined.

Which Companies Ask What

Company tagging in this bank is multi-valued — a single question can be associated with several companies that are known to ask similar things, based on reported interview loops. That's why company totals below don't sum to 389. Here are the 19 companies with the deepest question coverage, each with its single most common topic and how its questions split across SQL, conceptual, and Python.

A few patterns jump out. Google and Meta lean hardest on neural-networks as their top topic, consistent with both companies' deep bench of ML infrastructure roles. Amazon, Stripe, Airbnb, Uber, and LinkedIn all lead with joins — a strong signal that companies running large marketplace or logistics businesses lean on relational data skills more than companies running consumer feeds or model APIs. And the language split is its own tell: Microsoft, OpenAI, DeepMind, Anthropic, and NVIDIA all show zero SQL questions in this dataset, while Amazon's split is more than a third SQL — a useful data point if you're deciding where to spend your prep time before a specific onsite.

SQL Interview Questions: The Breakdown

SQL makes up 100 of the 389 questions in the bank, and within that 100, one topic dominates everything else.

Joins alone account for 30% of every SQL question in the bank — nearly double the next-largest SQL topic, window functions, at 15. If you only have time to drill one SQL concept before an interview, the data says it should be joins: multi-table joins, self-joins, and join-plus-aggregate combinations show up far more often than any single window-function or subquery pattern. Window functions and subqueries round out the next tier, together making up another 27% of SQL questions, and the remaining eight topics — group by, CTEs, WHERE clauses, string functions, date functions, SELECT basics, CASE WHEN, and set operations — split the last 43% in progressively smaller slices.

The LLM Question Surge

The most "of the moment" cut of this data is the LLM breakdown. Of the 35 questions tagged llms, here's how they concentrate by company (companies overlap, since a question can be tagged with more than one):

Google, OpenAI, Anthropic, DeepMind, and Meta together account for 83 of the 102 total company-question associations in this table — the five companies you'd expect to be building and shipping large language models are also the five companies most likely to ask you about them in an interview. What's notable is how new this category is relative to its size: llms is tied for the second-largest topic in the entire bank at 35 questions, on par with statistics and neural-networks, despite LLM-specific interview questions barely existing as a distinct category two years ago. Compare that to a legacy topic like where clauses, which has had a decade-plus head start and still only sits at 6 questions. The llms topic didn't grow into its current size gradually — it arrived quickly, tracking the pace at which the underlying technology moved from research paper to production API to standard interview material.

How to Reproduce This

Every number in this report is a direct count against DataScienceHired's own published question bank, filtered to status=published, with no external data sources, surveys, or estimates involved. Because the bank is actively growing, these exact counts will drift upward over time — that's why every stat here is dated to August 29, 2026, and why we'd encourage anyone citing these numbers to note the as-of date rather than treating them as a fixed, permanent snapshot.

Practice these yourself: the Google problem set is the single largest company-tagged collection in the bank, machine learning is the single largest topic, and joins is the single most-tested SQL pattern — three good starting points if you only have time to prep in one place before your next interview.