Dataford · Field Guide

What Data Interviews Actually Test in 2026

We broke down 7,976 real interview questions to see what data, analytics, and AI loops actually grade — which topics dominate, how hard they get, and how the exam changes from one role to the next. Here is the anatomy.

Author
Amney, Founder at Dataford
Dataset
7,976 questions
Period
2026 question bank
7,976
interview questions broken down by topic and format
#1
Behavioral & Leadership is the most-tested topic of all
69%
of questions are written or spoken, not coded
17%
are rated hard — the bar is real but not extreme

Summary · Key findings

01

The most-tested topic is not technical. Behavioral & Leadership is the single largest category at 1,121 questions — ahead of coding, SQL, or statistics. Data interviews grade how you think and communicate as much as what you can build.

02

Most questions are answered in words, not code. 69% of questions are written or spoken responses. Hands-on coding — Python and SQL combined — is about 20%. The whiteboard is a smaller part of the loop than its reputation suggests.

03

Difficulty clusters in the middle. 44% of questions are medium, 39% easy, and 17% hard. The genuinely punishing questions exist, but they are a minority — most of the test rewards solid fundamentals.

04

Each role sits a different exam. A Data Analyst loop leans on SQL and statistics; a Data Scientist loop leans on machine learning, model evaluation, and NLP. Same field, very different question mix.

05

The technical core is broad, not deep. Across the bank, eighteen distinct topic areas appear. Strong candidates are not the ones who go deepest on one; they are the ones with no obvious hole across the spread.


Candidates prepare for the interview they imagine, which is usually a wall of LeetCode. The interview that actually happens is wider and stranger than that.

Every question in Dataford's bank is tagged — what topic it belongs to, how hard it is, how you answer it, and which roles it shows up for. That structure lets us take 7,976 published questions and ask a simple question back: what is a data interview made of?

The answer reorders most candidates' priorities. The biggest category is one many people barely rehearse, the coding share is smaller than feared, and the “data interview” turns out to be several different exams wearing the same name. The full method is at the end.


Topics

What gets asked, ranked

The single most-tested topic is Behavioral & Leadership — bigger than coding, bigger than SQL, bigger than statistics. The technical categories matter, but they share the loop with a large, often under-rehearsed human layer.

Questions by topic · top 12 of 18
The most-tested topics in data interviews
Behavioral & Leadership
1,121
Execution
966
Coding
723
SQL & Data Manipulation
638
Product Sense
533
Pipelines
530
Strategy
478
Metrics
468
Statistics & Probability
427
Machine Learning
427
Model Evaluation
357
NLP
306

Read the top of the list together: Behavioral & Leadership and Execution are the two largest categories, and both are about judgment and delivery rather than syntax. A candidate who can write flawless SQL but cannot narrate a decision is preparing for a third of the interview and ignoring the rest.


Format

How you actually answer

Most questions are not typed into an editor. More than two thirds are written or spoken — you explain an approach, defend a trade-off, walk through a design. Hands-on coding is real but contained.

Share of questions by answer format
Words first, code second
68.6%
14.9%
Written / spoken 68.6%Python 14.9%Conceptual 6.1%Quantitative 5.4%SQL 5%

Difficulty, meanwhile, clusters in the middle. The bar is real, but the test is mostly built from medium and easy questions that reward clean fundamentals over heroics on a single hard problem.

Share of questions by difficulty
Most questions are not the hard ones
38.9%
43.8%
17.3%
Easy 38.9%Medium 43.8%Hard 17.3%

By role

Same field, different exams

The biggest mistake in data prep is treating “data” as one subject. Put a Data Scientist and a Data Analyst loop side by side and the question mix barely overlaps.

Questions by topic · count
Data Scientist vs Data Analyst
Data ScientistData Analyst
Machine Learning
143
19
Model Evaluation
125
23
NLP
113
10
Statistics & Prob.
44
31
SQL & Data Manip.
30
66
Behavioral
29
39

The Data Scientist exam is built on modeling; the Data Analyst exam is built on SQL and statistics.

The same divergence runs across every role. Each one has a signature — the cluster of topics it is graded on more than any other.

Data AnalystSQL & statistics
Data ScientistML, model evaluation & NLP
Data Engineerpipelines & SQL
ML Engineermachine learning & NLP
AI EngineerNLP, LLMs & coding

Preparing for the wrong signature is the quietest way to fail a loop you were qualified for. An analyst who drills neural networks and a scientist who only practices SQL are both studying the next role over.


Outlook

How to prepare for the real exam

If this report changes one thing about your prep, let it be the balance. The coding-only candidate is over-indexed on a fifth of the interview and under-indexed on the largest part of it.

Build for breadth first — be unembarrassing across the topic spread for your specific role — then rehearse the behavioral and execution layer until it is as fluent as your code. That is the combination the data rewards, and it is the one most candidates skip.


Practice the exam your role actually sits

Thousands of real questions, sorted by topic, difficulty, and role — with worked solutions, so you can train for the right mix.

Browse the question bank

FAQ

Frequently asked questions

What do data interviews actually test?+

More than code. Across 7,976 questions, the single largest topic is Behavioral & Leadership, followed by Execution and then technical areas like Coding, SQL, statistics, and machine learning. A data loop grades problem framing and communication alongside technical skill.

Is SQL or Python more important in data interviews?+

Both matter, but neither dominates the way candidates expect. Hands-on coding — Python and SQL together — is about 20% of questions. The majority (69%) are written or spoken responses about how you would approach a problem.

How hard are data interview questions?+

Balanced toward the middle: 44% medium, 39% easy, and 17% hard. The hard questions are real but a minority. Consistent fundamentals across topics beat going very deep on a single one.

Do different data roles get different questions?+

Yes, clearly. A Data Analyst is tested mostly on SQL and statistics; a Data Scientist on machine learning, model evaluation, and NLP; a Data Engineer on pipelines and SQL. Preparing for the wrong role's mix is a common, avoidable mistake.

How was this measured?+

From Dataford's published question bank of 7,976 questions, each tagged with a category, difficulty, answer format, and applicable roles. Counts reflect the questions companies ask for these roles, as captured and structured by Dataford.


Methodology

How this report was built

This report is built on Dataford's published question bank: 7,976 questions, each tagged with a topic category (18 in total), a difficulty (easy, medium, hard), an answer format, and the roles it applies to. Topic and format shares are computed across the full bank.

Per-role figures use role tags, and a question can apply to more than one role, so role columns are read within a role rather than summed across them. Answer formats were grouped into written or spoken responses, Python, SQL, quantitative, and conceptual.

The bank reflects the questions companies ask for these roles as captured and structured by Dataford; it is a large sample, not a census of every interview. Figures are current as of June 2026.