What Mercor's Domain Expert Interview is, how it differs from its other AI interviews, and a worked example of a shallow versus an expert answer.
The Mercor domain expert interview is a real, named step. Mercor's glossary defines the Domain Expert Interview as a type of AI interview tailored to subject-matter expertise, and its expert listings use the same name. It is one spoken interview of roughly 15 to 20 minutes with Mercor's AI interviewer, Monty, and Mercor's engineers describe it as a single assessment covering medicine, economics, history, law, software architecture and almost any other knowledge domain. Your resume shapes which field it probes and how deeply, and we found no published rubric or pass mark.
Below is what Mercor says about it, how it differs from Mercor's other interviews, and what separates an expert answer from a competent one, with a worked example in physics. Setup and general answering habits are in our guide to preparing for the Mercor AI interview.
What Mercor documents about the Domain Expert Interview
Mercor's help-center glossary describes the AI interview in general as an on-camera interview that assesses your domain knowledge and communication skills. Its talent-network listings for physics, chemistry, medicine and law describe the same process. You apply with your resume, confirm where you work from and take an AI interview of about 20 minutes, and Mercor then verifies your background and adds you to that field's pool.
The most detailed source is a March 2026 post about Monty on Mercor's engineering blog. It says Mercor first built one assessment per job title, found that this did not scale, and then grouped listings by the skills they actually test. The largest group, Domain Expert, accounts for over 70% of all sessions, and code and language assessments cover most of the rest. Completing Domain Expert once qualifies a candidate for every role in that cluster, and Mercor processes the resume before the session so that it shapes what gets asked, how deeply to probe and what to skip.
The blog post calls each interview fifteen minutes long, while the help center and listings say about 20 minutes, so plan for 20. Mercor also asks candidates not to share the specific questions they were asked, which is why public lists of questions from actual interviews are thin and why our example below is an illustration.
How it differs from Mercor's other AI interviews
The engineering post names other interview types, including Code, Language and Professional, and Mercor's help pages mention a "Generalist Interview" as an example. We found no published comparison from Mercor, so the differences below are our reading of its documents.
The first difference is that one interview serves many fields. A coding interview generates a problem and starter code from your profile, while the Domain Expert Interview has to infer your field from your resume and then go deep on it. If you span two fields, such as a physics PhD now working in quantitative finance, your resume effectively chooses which one gets tested, so lead with the field you want to be matched on.
The second difference is that the listings tell you what depth means. The physics, chemistry, medicine and law networks describe three kinds of paid work: writing a problem from your own work with a reference answer, grading a model's attempt, and flagging plausible answers built on reasoning that fails under real conditions. The physics listing says that where you trained matters less than whether you did the derivations and the analysis yourself, recently enough to explain the calls you made; the chemistry listing uses almost the same wording, and the physician listing asks for current clinical judgment. We expect an interview that screens for this work to reward answers that judge reasoning, not only recall facts.
The one candidate report we could open points the same way. Published on Mozibox and last updated in July 2025, it describes a medical-role interview with three themes: why the candidate's specialty gave them deep expertise, a recent case and how they reasoned through it, and an idea for applying a language model in their field, with a way to evaluate it. It is a single account that predates Mercor's 2026 post, but it fits the listings' emphasis on applied judgment.
What an expert-level answer sounds like
The difference between a competent answer and an expert one is rarely the fact itself. Read against the listings, an expert answer names the assumption the result depends on, says where that assumption stops holding, gives a quick check that would catch the error, and says what to do about it. A competent answer stops at the fact, and an answer that stops there reads like a textbook, when the listings say labs come to Mercor for judgment their models cannot generate.
Timing matters more for experts than for most candidates, because experts pause to qualify their answers. Mercor's engineering post says it settled on 900 milliseconds as its production pause threshold, with the right setting shifting by interview segment, so a pause of around a second before your caveat may be treated as the end of your turn. Say the conclusion first and the qualifications second, so that if you are cut off, you have already made your main point.
Worked example: grading an approximation in physics
We wrote this exchange as an illustration; it is not a question Mercor has published or a candidate has reported. We chose it because the physics listing names this skill: checking whether an approximation is valid in the regime claimed and whether the limiting cases check out.
Interviewer: A model was asked for the period of a 1-metre pendulum released from 60 degrees, and it answered 2π√(L/g), about 2.01 seconds. How would you grade that answer?
Shallow answer: "That's the standard pendulum formula, so the method is right and 2.01 seconds is correct."
Strong answer: "I'd mark it wrong, or at best incomplete. 2π√(L/g) is the small-angle result, and 60 degrees is outside the regime where that holds. The exact period uses an elliptic integral, and at 60 degrees it is about 7% longer, roughly 2.15 seconds. A quick check is the first correction term, which multiplies the small-angle period by one plus θ₀²/16 and already gives about 6.9% at 60 degrees. In the limits, the exact result goes back to 2π√(L/g) as the amplitude goes to zero and diverges as it approaches 180 degrees, so a formula that doesn't depend on amplitude can't be right here. In the rubric I'd give credit for identifying the system but not for the final number, and I'd note that the answer is acceptable below about 10 degrees, where the error is under 0.2%."
None of the strong answer's advantages is a more obscure fact. It names the assumption hidden in the model's answer, tests it with numbers and limiting cases, and turns the judgment into something a rubric could use, which is the physics listing's work done out loud.
We checked the numbers against the formulas on Wikipedia's pendulum page, using its arithmetic-geometric mean form of the exact period, in Python 3:
from math import sqrt, cos, radians
def agm(a, b):
while abs(a - b) > 1e-15:
a, b = (a + b) / 2, sqrt(a * b)
return a
def exact_over_small(theta0_deg):
# exact period / (2*pi*sqrt(L/g)), using the AGM form
return 1 / agm(1, cos(radians(theta0_deg) / 2))
def first_correction(theta0_deg):
t = radians(theta0_deg)
return 1 + t**2 / 16
for deg in (10, 30, 60, 90):
print(f"{deg:>2} deg exact {exact_over_small(deg):.4f} "
f"1 + theta^2/16 {first_correction(deg):.4f}")10 deg exact 1.0019 1 + theta^2/16 1.0019
30 deg exact 1.0174 1 + theta^2/16 1.0171
60 deg exact 1.0732 1 + theta^2/16 1.0685
90 deg exact 1.1803 1 + theta^2/16 1.1542For a 1-metre pendulum with g = 9.81 m/s², the small-angle period is 2.006 seconds and the exact period at 60 degrees is 2.153 seconds.
How to adapt this to your own field
Each expert listing describes its own version of a right-looking answer with flawed reasoning. The physician listing describes a recommendation defensible in a textbook but wrong for the patient in front of you. The law listing describes a model that states the law correctly and would still lose, and the chemistry listing a route that works on paper and fails in the flask. Work through these steps in the week before.
- Open your field's talent-network listing on Mercor and rewrite its three task types in the language of your own speciality.
- Pick two pieces of recent work you did yourself, and for each one write down the assumption that carried the result, the check you ran, and what would have changed your conclusion.
- Build one grade-this example of your own, like the pendulum above: a plausible answer with a hidden flaw, and two or three sentences explaining exactly why it fails.
- Practise each answer aloud with the conclusion first, and keep it under two minutes so the caveats arrive before the interviewer moves on.
- Take Mercor's free practice interview to hear the pacing, because it does not affect future opportunities, and save your official retakes, since hiring managers can see your retake count.
For a worked example from a different field, our page on Mercor finance interview questions applies the same approach to expected loss and value at risk.
What happens after the interview
The listings say they do not accept or reject candidates, so you will not get a decision on the listing itself. Mercor invites matching people from the pool to a specific project listing when one starts, which can take a week or several months. The engineering post says more than half of Mercor's offers now go out proactively, to candidates who took an interview earlier without applying for that job.
Some projects add a human conversation. A current Mercor listing for an Expert Interviewer describes structured interviews of about 20 minutes with shortlisted engineers, assessing technical depth, communication and AI-evaluation fit against a rubric, then sorting each candidate into advance, hold or pass. If you are shortlisted for such a project, expect a person to probe the same depth. Our overview of how the Mercor AI interview works covers the later steps, including assessments and work trials.
Practising with Verve AI
Mercor's LLM usage policy allows language models for grammar, diction and subtle phrasing checks, and it prohibits using them to assess or evaluate model responses, to provide justifications or explanations for writing purposes, or to predict code behaviour or runtime results. Its AI interview help page separately asks you not to use LLMs to compose your responses, so build the skill before the interview and check both pages for what applies to your session. Verve AI's mock interview has an AI recruiter ask you questions one at a time, and its Custom Questions bank lets you load your own grade-this examples and practise explaining them against an interview profile built from your resume. Afterwards you get "a detailed performance report with personalized feedback and scoring", which you can read against the depth criteria above.
FAQ
Is the Mercor domain expert interview different for each field?
Mercor describes Domain Expert as a single assessment that covers many knowledge domains, so it is not a separate test written for each field. It reads your resume before the session, and that shapes what it asks and how deeply it probes. A physicist and a lawyer should therefore expect very different questions from the same assessment.
How long is the Domain Expert Interview?
Mercor's help center and its expert listings say about 20 minutes, and its engineering blog describes interviews as fifteen minutes long. An attempt that ends much earlier than planned shows as "Interview too short" and has to be retaken. Plan for 20 minutes in a quiet room.
Do I need to retake it for every network I join?
No. Mercor says completing Domain Expert once qualifies you for every role in that cluster, and its help pages say completed interviews carry over to later applications and do not expire. The listings add that finishing the interview and joining every network you qualify for both improve your chances of being matched. Our guide to retaking or resetting a Mercor AI interview explains the three-retake limit.
Will it ask textbook questions?
We found no published list of Mercor's questions, and Mercor asks candidates not to share them. Its listings say that where you trained matters less than whether you did the work yourself and can explain the calls you made, which points to questions about your own work and your judgment. Prepare to go deep on two or three things you did, and to explain how you would spot a flawed answer in your field.
Sources
- Glossary definitions of the Domain Expert Interview and the AI interview: https://talent.docs.mercor.com/support/glossary
- AI interview length, practice interview, three retakes, retake count visible to hiring managers, "Interview too short", request not to share specific questions, instruction not to use LLMs to compose your responses: https://talent.docs.mercor.com/support/ai-interview
- Completed steps carry over and never expire; "Generalist Interview" as an example name: https://talent.docs.mercor.com/new-releases/new-application-steps
- LLM usage policy: grammar, diction and phrasing checks allowed; assessing model responses, writing justifications or explanations, and predicting code behaviour prohibited: https://talent.docs.mercor.com/policies/llm-usage
- Monty, Domain Expert as one assessment across domains, over 70% of sessions, code and language types, one completion qualifies for the cluster, resume shapes questions, 900 ms production pause threshold that shifts by interview segment, fifteen-minute interviews, over half of offers proactive: https://www.mercor.com/blog/monty-engineering-deep-dive/
- Physics Expert network: task types, "where you trained matters less", 20-minute AI interview, "domain expert interview", no accept/reject, timing of matches: https://work.mercor.com/jobs/list_AAABoKMjqUMNbxdYam1EnL82/physics-expert
- Chemistry Expert network: "works on paper and fails in the flask", same process wording: https://work.mercor.com/jobs/list_AAABoKMjbmWSx1ImOhtE4p1n/chemistry-expert
- Medical Expert (M.D./D.O.) network: "defensible in a textbook but wrong for the patient in front of you": https://work.mercor.com/jobs/list_AAABoKMig9vns0-qeYBLhoVI/medical-expert-m-d-d-o
- Medical Expert network: same process wording: https://work.mercor.com/jobs/list_AAABoKOPjbVqcleYq1NNjKxo/medical-expert
- Legal Expert network: "states the law correctly and would still lose": https://work.mercor.com/jobs/list_AAABoKOQUds_uf4uJMtGK64s/legal-expert
- Expert Interviewer listing: 20-minute structured human interviews of shortlisted engineers, rubric, advance/hold/pass: https://work.mercor.com/jobs/list_AAABoDrYQdAJqq6xRalOo4e9/expert-interviewer
- Candidate report of a medical-role Mercor AI interview (updated July 2025): https://mozibox.com/articles/1142
- Pendulum period: small-angle formula, exact period, AGM form, series with θ₀²/16, divergence at 180 degrees, under 0.2% error at 10 degrees: https://en.wikipedia.org/wiki/Pendulum_%28mechanics%29
- Verve AI mock interview: AI recruiter asks questions one at a time, Custom Questions, interview profile, detailed performance report with personalized feedback and scoring: https://docs.vervecopilot.com/features/mock-interview
Ryan
CEO of Verve AI
