How to prepare for the Mercor interview
The first time most people meet Mercor's interviewer, they are talking to a machine. There is no small talk, no interviewer nodding along, and no chance to read the room. Twenty minutes later it is over, and you genuinely cannot tell whether it went well.
That uncertainty is why so many qualified people prepare badly for it. They either over-prepare like it is a FAANG coding loop, or they treat it as a formality and wing it. Both fail for the same reason: the interview is not testing what they think it is testing.
This guide covers what the conversation actually is, what it appears to score, the specific mistakes that cut strong candidates early, and - the part nobody writes about - what happens in the weeks after you pass.
What the Mercor interview actually is
Mercor recruits domain experts - doctors, lawyers, engineers, PhDs - and places them on projects training and evaluating AI models for frontier labs. Before you get near a project, you talk to an AI interviewer.
The format is a recorded video conversation, typically 20 to 30 minutes. Candidate reports on Glassdoor put the difficulty at roughly 3 out of 5 - moderate, not brutal. That rating is the single most useful expectation-setter available: this is not an exam you can fail on technical brilliance, and it is not a formality you can sleepwalk through.
The thing to understand is what it exists to do. Mercor's business depends on placing people whose expertise is real. The interview is a verification layer. It is checking that the person on camera can talk credibly and specifically about the work their CV claims.
What it asks
The questions anchor to your CV. If you wrote that you led a migration, expect to be asked what broke. If you claimed a specialism, expect a question only a practitioner could answer well.
Across candidate accounts, the shape is consistent:
| Phase | What it covers | Roughly |
|---|---|---|
| Opening | Who you are, what you do, current role | 2-3 min |
| Background depth | Your claimed experience, probed for specifics | 8-12 min |
| Domain judgement | A scenario or problem from your stated field | 5-8 min |
| Logistics | Availability, hours, tooling, connection | 2-3 min |
Technical candidates report a heavier skew toward practical problem-solving. Candidate write-ups on Naukri's Code360 describe rounds covering past projects, coding experience and system design for engineering roles. Non-technical domain experts get proportionally more scenario questions and fewer puzzles.
What it appears to score
Mercor does not publish a rubric, so treat this as inference from consistent candidate reports rather than documented fact. Four things show up again and again:
- Specificity. Numbers, timelines, team sizes, names of tools. Vague answers are the single most common reason strong people get cut.
- Consistency with your CV. The interviewer is anchored to what you claimed. Contradicting your own document is fatal.
- Communication clarity. You will be writing feedback that an AI lab reads. If you cannot explain your reasoning out loud, that is disqualifying for the actual job.
- Genuine depth. One follow-up question past the surface is where borrowed expertise falls apart.
“The interview is not asking whether you are impressive. It is asking whether you are who your resume says you are.”
The walkthrough above covers the full application flow end to end, including what the interview screen actually looks like when it starts.
How to prepare: the night before
Preparation for this interview is unusually high-leverage, because almost nobody does it. An hour the night before genuinely changes outcomes.
Re-read your own CV. This sounds trivial. It is the highest-return thing you can do, because every question anchors to it. If your CV says you "led a migration," you need the scope, the team size, the timeline and what went wrong ready to say out loud.
Prepare three stories with numbers in them. Not three topics - three specific situations, each with a real figure attached. "We cut review time from nine days to four across a team of six" survives a follow-up question. "I improved efficiency" does not.
Rehearse a two-line introduction. Who you are, what you specialise in. Say it out loud twice. The opening sets the tone of everything after it, and people who ramble at the start tend to ramble throughout.
Pick one thing to be genuinely deep on. You are being hired for judgement in a narrow area. Decide in advance what that area is, and be ready to go three questions deep in it.
If your CV itself needs work before any of this matters, start with our guide on writing a resume that passes AI screening - the screening layer sits before the interview, and a CV that reads badly to a model never gets you to this stage.
The setup nobody mentions
It is a recorded video conversation, so the basics carry more weight than they would with a human who can compensate for a bad line.
- Wired connection if you have one. A dropped answer mid-sentence reads as a non-answer.
- Quiet room, door closed. Background noise degrades the transcription you are being assessed on.
- Face lit from the front. A window behind you makes you a silhouette.
- Headset over laptop mic. Cleaner audio, fewer transcription errors on technical terms.
- CV open in a second window. You are allowed to refer to it. Most people forget they can.
The five mistakes that cut strong candidates
These are the patterns that come up repeatedly in candidate accounts. None of them are about ability.
| Mistake | Why it fails | Do this instead |
|---|---|---|
| Answering in generalities | There is nothing for the model to verify | Attach a number, a date or a name to every claim |
| Overstating on your CV | One follow-up exposes it, and it reads as dishonesty rather than stretch | Claim only what you can discuss for three minutes |
| Treating it as a coding test | Most roles are judgement, not implementation | Prepare to explain decisions, not just produce answers |
| Rambling | Long answers bury the substance | Answer, then give one example, then stop |
| Winging it because 'it's just an AI' | The bar is moderate, not absent | Spend the hour. Almost no one else does |
That last one deserves emphasis. Some candidates report winging the interview successfully, which has created a folk belief that preparation is unnecessary. Survivorship bias - you are reading the accounts of the people it worked for.
What actually happens after you pass
This is the section most guides skip, and it is the one that changes how you should feel about the whole process.
Passing the interview does not produce a job. It puts you in a pool. Work then arrives - or does not - depending on what labs are currently commissioning in your field.
The community discussion here is worth reading honestly. In a TeamBlind thread on getting hired through Mercor, the most common complaint is silence after the interview. One commenter reported receiving several offers but noted projects "usually just last 1-2 weeks only." Others report hearing nothing at all and question whether they were ever really in contention.
Set your expectations against that, not against the enthusiastic YouTube reviews:
| What people expect | What tends to happen |
|---|---|
| A job offer | Entry to a pool of approved experts |
| Steady weekly hours | Bursts of work tied to lab demand |
| Long engagements | Often short projects, sometimes 1-2 weeks |
| Feedback on rejection | Usually silence |
None of this makes Mercor a bad option - the rates are genuinely strong, and across the 180 Mercor listings we track the typical advertised rate is $60 to $80 an hour. See the full breakdown on our Mercor pay page or the wider pay report across 341 listings. It just means you should apply somewhere else at the same time rather than waiting.
A more sceptical take on whether the rates hold up in practice. Worth watching alongside the walkthrough above - the honest answer sits somewhere between the two.
If you get rejected
Rejections usually arrive without a reason, which makes them hard to learn from. Three things are worth doing.
- Apply to a second platform immediately. micro1 runs a comparable process with a different screening model - see our micro1 interview guide and our honest review of micro1.
- Reread your CV as an assessor would. Most rejections trace back to claims that could not survive a follow-up question.
- Wait before reapplying. Reapplying the next day with the same profile produces the same result. Change something material first.
It is also worth being clear-eyed about whether the platform suits you at all. Our honest review of Mercor covers who it genuinely works for and who it does not.
Common questions
- How long is it? Typically 20-30 minutes, recorded, one sitting.
- Can I retake it? Not immediately. Change something meaningful on your profile before trying again.
- Is it a coding test? Only for engineering roles, and even then it leans toward explaining decisions rather than writing production code.
- Does it use my camera? Yes, it is a video conversation. Treat the setup as part of the assessment.
- How soon do I hear back? Days to a few weeks. Silence past that is common and is not usually reversible.
- Do I need AI experience? No. You are being hired for your existing domain expertise.
The one-hour version
If you only do four things: reread your CV out loud, write down three stories with real numbers in them, test your microphone, and apply to a second platform the same week so you are not waiting on one outcome.
That is genuinely most of the edge available here. The interview rewards preparation far more than it rewards brilliance, and almost nobody prepares.
Some outbound links to platforms are referral links. This does not affect the pay figures above, which are taken from the listings themselves.