What the real interview was like

1. Applied through micro1.
2. An AI interview and assessment.
3. Waiting for onboarding when the review was written.

• Explaining the reasoning behind each answer.
• Handling ambiguous cases.
• Interpreting task guidelines.
• Practical quality and evaluation scenarios for AI work.

• Structured and professional, with a flow that was easy to follow.
• Pushed for depth rather than accepting surface answers.

• →Practise explaining your reasoning step by step.
• →Prepare for ambiguous cases and read guidelines closely.

Finished the assessment; no project yet at the time of writing.

1. Account setup and general onboarding.
2. A project screener with written evaluation tasks.
3. A course described as paid training.

• Detailed written evaluations that needed careful answers.

• Self-paced, with no interviewer.
• Took about four times the advertised 75 minutes.

• →Allow far more time than the advertised screener length.

Onboarding failed, with no reason given.

1. An assessment of side-by-side response comparisons.
2. Later saw practice material showing each pair came from different systems.

• Choose the better of two chatbot responses, and explain why.

• No interviewer: written, self-paced tasks.

• →The two responses may come from different models. Judge each on its merits.

Not known when the review was posted.

1. An application form, including demographic questions.
2. An assessment built on technical, unusual prompt scenarios.

• Highly technical, niche scenarios instead of typical user requests.

• No interviewer: a written assessment.
• Did not seem to test everyday reasoning and communication.

• →Expect specialised prompts rather than everyday chatbot questions.

No response after submitting.

1. A generalist interview with the AI, about 40 minutes.
2. A C programming interview, about 20 minutes, then a coding test set in JavaScript.
3. An AI annotator interview drawing on their chip design field.

• How the generalist role works, in detail.
• Abstract C concepts.
• Name two red flags in an AI answer about chip scaling.

• Fast, abstract questions packed into a short time.
• Felt more like one AI talking to another than a conversation.
• The coding test language did not match the C interview.

• →Practise spotting red flags in AI answers from your own field.
• →Check which language the coding test uses before you start.

Rejected for the generalist role; other results not stated.

1. Brought onto a project.
2. The project paused for about three months.
3. A three-hour re-qualification evaluation.

• A three-hour evaluation of project work.

• No interviewer: a scored test with a 70 percent pass mark.

• →Expect a fresh test when a paused project restarts.

Failed by one point.

1. Applied through a job posting.
2. Several AI-run assessments spread over several days.

• Long assessments that felt unrelated to the advertised job.

• Run by AI, not people.
• Unpaid and spread across several days.

• →Ask how long the unpaid assessments are before committing.

No further contact afterwards.

1. A recorded video interview.
2. A long set of guidelines to read.
3. An hour-long training video.
4. An hour-long group call with trainers.
5. An evaluation test rating chatbot responses.

• Assess chatbot responses against the project guidelines.

• A recorded video interview at the start, with no live interviewer.
• The test seemed to be graded automatically within an hour.

• →Study the guidelines closely, since the test is judged against them.

Failed the evaluation with no feedback.

1. Identity check with both sides of a government ID.
2. A native-language qualification test.
3. Project-specific assessments.

• Native language proficiency.
• Spotting audio quality problems and following the guidelines closely.

• Demanding but fair for anyone who studies the instructions.

• →Study each project's instructions before its assessment.
• →Practise spotting audio quality issues.
• →Write clear explanations in simple words.

Passed and went on to steady weekly work.

1. Enrolled in the project.
2. Repeated unpaid evaluation rounds.
3. Surprise video and audio recording assessments.

• Recorded video and audio submissions.

• Felt like a loop of tests with no path to paid tasks.
• The recordings raised privacy worries.

• →Ask before starting whether recordings are required and how they are used.

Never reached paid work.

1. Studied a long guideline document for several days.
2. Took the qualification assessment for the audio project.
3. Rejected, then asked support for a review.

• Judging audio clips against the project guidelines.

• The expected answers seemed to disagree with the written rules and the audio.
• Scores could not be reviewed and retakes were not allowed.

• →Expect one attempt only on audio qualifications.
• →Budget unpaid study time before you start.

Failed, with no retake or review.

1. Joined the platform.
2. Transcription test, first attempt.
3. Transcription test, second attempt.
4. Rejected.

• Transcribing audio accurately to the project's standard.

• No feedback on why either attempt failed.

• →Transcription tests follow the project's style guide, so read it closely.

Rejected after two attempts.

1. Two written questions with the application: a past decision, and explaining a complex idea to a non-specialist.
2. A practice run of about 15 minutes in the real interface.
3. The real interview with Zara: 28 minutes, with five focus areas shown at the top of the screen.
4. A certification email two days later, then a project match nine days after that.

• Why you made a particular design decision.
• How you critique and evaluate work.
• How you make sense of an ambiguous brief.
• How you explain design to people outside the field.
• How you handle disagreement and bring others round.

• Each question was spoken and also shown as text, so it could be reread.
• When the candidate lacked the exact experience, Zara reframed the question around the underlying skill.
• Thinking aloud was taken as the answer, so pauses worked better kept silent.

• →The written questions hint at the rubric. Do the practice run.
• →Think in silence, or say you need a moment. Thinking aloud was taken as a finished answer.
• →Prepare two or three project stories before the call.

Certified within two days and matched to a project nine days later, after a hiring manager reviewed the profile.

1. Several micro1 assessments across design roles.
2. Practical exercises inside the visualization specialist assessment.

• Spot a tiny header size difference, 48 against 46 pixels, with no tools.
• Other very specific visual questions.

• The AI-led exercises felt strange at times.
• No design or measuring tools were available for the visual tasks.

• →Expect fine visual checks done by eye alone.

Not stated.

1. Signed up, then chose location, languages and specialism.
2. Identity and phone verification with a photo ID.
3. Uploaded a resume and picked areas of expertise.
4. A screening exam; no video interview for this track.
5. Guided training modules, then placed in a team with a lead.

• A screening exam for the content reviewer track.
• Training modules on the tools and task guidelines.

• No AI or live interview for this role, only a written screening.
• Onboarding was long and demanding, more than simple clicking.

• →Have your ID ready before verification.
• →Read each task's guidelines closely.
• →Set aside several hours for onboarding.

Passed, joined a team and received a small onboarding payment.

1. One AI interview with Zara.

• Understanding of journalism.
• Writing ability.

• Warm and supportive in tone.
• Named what it liked in answers and offered polite suggestions to improve.

• →Take the interviewer's suggestions on board; they show what the rubric values.

Not stated; the candidate left a five-star review.

1. An AI interview in English.
2. An AI interview in Spanish.

• Spoken answers in Spanish, with the interviewer remarking on word choice along the way.

• The English interview felt complete and comfortable.
• The Spanish one kept suggesting other words and sounded defensive when challenged.

• →In Spanish, expect comments on your vocabulary. Keep your rhythm and carry on.
• →If you can choose, think about which language shows you at your best.

Not stated.

1. One AI interview with Zara for a role in a language other than English.

• Questions built around English technical terms, even though the role was not in English.

• Natural and respectful in tone.
• Interrupted the candidate and could not handle many of their questions.

• →Learn the English technical terms for your field, even for a job in another language.

Not stated.

1. Removed from the German version of a project for not being fluent enough.
2. Passed the screening for the English version of the same project.
3. Stayed blocked by the old German restriction.
4. A multilingual exam retake was required.

• A project screening in English.
• A multilingual language exam, retaken under pressure.

• Support replies were polite but generic.

• →A failed language project can carry over as a block on related projects.
• →Keep a written record when a qualification seems wrongly blocked.

Failed the multilingual retake and could not keep working.

1. Applied through a job listing.
2. Several exam rounds back to back, about four hours in total.
3. Failed one exam, which ended the application.

• Rank items on scales from moderate to highest.
• Judge how well search results meet a need, on a meets-style scale.

• No interviewer: written exams.
• The ratings felt subjective and the criteria unclear.

• →Expect several hours of unpaid exams.
• →Learn the rating scale definitions in the guidelines before you start.

Failed one exam and was dropped.

1. Unsolicited outreach from a recruiter.
2. Automated replies said to take the tests before any role or pay details.
3. A Dutch proficiency test, failed twice.
4. Passed on the third attempt.

• A Dutch language proficiency test, which the candidate felt had grammar mistakes of its own.

• Automated replies only, with no role or pay details before testing.

• →Ask for pay and role details before investing time in tests.

Passed on the third attempt, then chose not to continue.

1. Accepted after about three weeks.
2. Learning materials, then a qualification exam with a seven-day window.
3. Part one: timed rating tasks.
4. Part two: tasks needing a phone to check an app and video links.

• Rate ads and results on a set rating scale.
• Check an app and video links on a mobile phone.

• Per-task timers were not explained beforehand.
• Support took four days to reply.

• →Start early in the exam window in case access fails.
• →Have a phone ready for the second part.
• →Watch for per-task time limits.

The exam closed early with no result.

1. A microphone check by saying a short greeting to the AI.
2. An AI interview including a pronunciation assessment.

• Pronunciation in a tonal language.

• Judged tones accurately.
• The microphone check gave no clear sign that audio was captured.

• →Test your microphone separately before the interview.

Not stated.

1. One AI interview, available in French and English.

• Questions built on specific examples from the CV.

• Fluent in both languages.
• The questions felt well designed and relevant.

• →Know the examples on your CV well, since the questions build on them.

Not stated.

1. Invited to projects, though links often failed.
2. A French assessment covering speaking and writing.
3. Early encouragement, then an email saying they did not qualify.

• Spoken French.
• Written French.

• Mixed messages: early encouragement, then a rejection.

• →Passing the language test does not guarantee a project place.

Told they did not qualify for the project.

1. Applied on Mercor for a remote role judging AI-generated consulting work.
2. One AI-led interview conversation.

• Where an AI transformation should start: existing systems, data and processes.
• Balancing business ambition against what is technically feasible.
• AI disruption in media and entertainment, and what should be released to the public.
• Governance, ethics and guardrails.
• Open-source versus proprietary foundation models.

• Moved between themes like a practitioner, not a recruiter ticking keywords.
• Focused and probing, though without the warmth of a good human interviewer.

• →Show how you judge messy business situations, not just AI vocabulary.
• →Link boardroom strategy to hands-on knowledge of the tools.
• →Frame answers around governance and human accountability.

The post does not give a result.

1. Submitted a profile and CV.
2. Asked to enter much of the same information again.
3. An AI video interview with a blank video screen.

• A key project that met a challenge, and how you overcame it.

• Cut in when the candidate paused mid-answer.
• Did not recover well when told he had not finished.
• The question had no clear link to the role.

• →Keep answers flowing; a long pause may be taken as the end.
• →Expect general behavioural questions as well as role ones.

He came away with a poor impression of AI video interviews.

1. Found the listing on a job site and spent about an hour on a tailored application.
2. Sent straight to an AI interview.
3. Placed in a general talent pool afterwards.

• Questions on legal background and the work, each with a sensible follow-up.

• Thoughtful and logical, and seemed genuinely interested in the answers.

• →Ask whether a real project stands behind the listing before investing hours.
• →Keep sensitive personal details off the resume you upload.

Joined a talent pool with no specific role offered.

1. Applied to a part-time doctor contract, up to 20 hours a week, helping build AI for medical research tasks.
2. A recruiter sent instructions by email and followed up with a phone call.
3. One AI interview, taken the day after applying.

• Your specialty, and why you count as an expert in it.
• A recent clinical case and how you reasoned through it.
• One way language models could help in medicine, and how you would measure whether it works.

• Asked sensible follow-up questions on each answer.
• Cut in a few times while the candidate was still explaining.

• →Bring one concrete use of AI in your specialty, and say how you would score it.
• →The interviewer may cut in mid-answer. Its follow-ups still made sense.

The post does not say whether the candidate went further.

1. The AI interviewer introduced itself and explained the format.
2. Questions drawn from the CV: running programmes, research, care with limited resources.
3. Five minutes to read a case tailored to the candidate's background.
4. About thirteen minutes to answer the case aloud, with follow-up probes.

• Running disease programmes while balancing clinical care and coordination.
• Research listed on the CV.
• Delivering care with limited resources and building local skills.
• A health delivery case: weak infrastructure, culture, and working with NGOs and government.

• A spoken voice with subtitles on screen.
• Adapted as it went and asked probing follow-ups.
• Knew the context of the region in the case.

• →Expect the case to be built from your own CV.
• →Use the reading time to plan a spoken answer.
• →Be ready to go deep on each programme you list.

Placed in the talent pool; no project yet when the post was written.

1. A short online application of a few minutes.
2. ID verification while applying.
3. A reply within about 12 hours.
4. Self-paced training modules with quizzes after the offer.
5. Joined a project once training was done.

• Identity documents.
• Quizzes on the training material.

• No interviewer: an application and verification only.
• The training felt structured and written for physicians.

• →Have your ID ready when you apply.
• →Set aside time for the training and quizzes after the offer.
• →Join the live prompt and rubric sessions if offered.

Offer received and onboarding started.

1. One AI interview on the platform.

• How you teach.
• How you interview patients.

• Direct, and focused on the expertise the profile claimed.

• →Expect questions that test the exact skills your profile claims.

Not stated.

1. An AI screening.
2. An AI interview.

• Scientific and technical expertise in their field.

• Professional throughout.
• Judged shown skill rather than paper credentials.

• →Be ready to show your expertise live, not just list it.

Not stated.

1. A recruiter got in touch about a remote student role supporting AI lab projects.
2. One 20-minute AI interview, purely technical, with a visible clock.
3. Waiting for a decision.

• Backend architecture and how it was used in real projects.
• How parts of a system talk to each other, such as choosing REST or GraphQL.
• What was actually built during internships.
• Designing a financial chatbot: latency, caching, WebSockets versus polling, several data APIs.
• Security: encryption, tokens, rate limits, compliance logging, SQL injection.

• No warm-up: technical questions from the first minute.
• Cared more about why a choice was made than about definitions.
• Intense but fair. Admitting a gap and reasoning it through seemed to land well.

• →Use real project examples rather than textbook answers.
• →If you have not done something, say so and explain how you would approach it.
• →Have concrete numbers ready, such as the traffic a system handled.

Still waiting to hear back when the post was written.

1. Applied after seeing an advert.
2. A coding test of roughly three hours.
3. Two weeks with no decision.

• Review functions and explain which are right and which are wrong.
• A puzzle-style programming problem.
• Python was the main expected language.

• A self-paced written test with no live interviewer.

• →Set aside about three hours for the coding test.
• →Do not count on quick feedback.

No response two weeks after submitting.

1. Signed up.
2. Identity and phone verification.
3. Nine separate coding screenings, one per language, over about a month.

• Screenings in Python, Java, Go, Dart, C++, C#, JavaScript, Swift and Verilog.

• Unpaid, self-paced screenings with no live interviewer.
• Took a lot of time across many languages.

• →Passing a screening does not guarantee paid work.

Passed all nine, but had no paid tasks when the review was posted.