Science and research
AI interviews for scientists and researchers
Talk through your real research with an AI interviewer, then see your results first and choose who sees them. Below: what strong answers show, the 5 practice questions, and a worked example.
- Free for experts
- About 15 minutes, whenever you like
- Type or speak. No camera needed
Retake it up to 3 times: you keep every attempt, and choose which one represents you.
Who it's for
The work it covers
Researchers and scientists in the natural, social and applied sciences.
Example roles
- Research scientist
- Postdoctoral researcher
- Lab manager
- Chemist
- Biologist
- Physicist
- Social scientist
- Economist
Name your exact specialty when you apply, and the questions follow it.
Topics the interviewer draws on in science and research
- An experiment that failed and how they found out why
- Designing a study with the right controls
- Choosing the analysis or statistics for their data and why it fit the design
- A result that surprised them
- Checking that a result would hold up before reporting it
- Responding to reviewer or peer criticism
- Explaining their work to someone outside the field
- Choosing which question to pursue with limited time or funding
Scoring
What a strong answer shows
These are the words the scoring AI is given for science and research, as written in Aethel's configuration. The scoring AI is told to score an area below 60 when its answers show one of the warning signs, however confident the answer sounds.
Strong work
The candidate designs studies with proper controls and enough replicates, troubleshoots by changing one thing at a time, matches the statistics to the design, and states the limits of the result.
Warning signs
The candidate changes several things at once and cannot tell what helped, skips controls or replication, overstates what a result shows, or cannot describe the method.
The four scored areas
Scientific reasoning
Hypotheses, troubleshooting, results
Counts as evidence when the candidate works through a specific research problem: the question, the hypothesis, what went wrong, how they troubleshot it, or what the result showed.
Research design and rigor
Methods, controls, statistics
Counts as evidence when the candidate describes how they designed a study or experiment: methods, controls, sample size, measurements, statistics, reproducibility or limitations.
Communication
Papers, talks, plain explanations
Counts as evidence when the candidate describes explaining research to other people: papers, grant proposals, talks, posters, or explaining results to people outside their field. Answers that respond to the question asked, in a clear order, with a concrete example, also count, but only up to 80 on their own. Grammar, word choice and fluency are never judged.
Collaboration
Co-authors, labs, reviewers
Counts as evidence when the candidate describes working with other people: co-authors, lab members, reviewers, partners or students, including feedback and disagreement. Saying "we" is not, by itself, collaboration evidence.
An area is scored only when the AI can quote you; otherwise it shows Not enough evidence, not a low score. How AI interviews are scored
Practice
The 5 practice questions
Questions like these come up in the interview. The free practice room asks 4 of them, at any level. The real interview follows up on your own answers instead.
- Question 1
Tell me about a research question you worked on and what made it hard.
- Question 2
How did you design the study or experiment, and what controls did you use?
- Question 3
Imagine your key result does not replicate on the second run. What do you check first?
- Question 4
How would you explain your findings and their limits to someone outside your field?
- Question 5
If you had six more months, what would you do next, and what result would change your mind?
No account. Each answer gets a quick rule check in your browser.
Worked example
One answer, read by the rules
Made up by Aethel to show how the published rules read one answer: not a real candidate, and not scored by the AI. A real score can differ.
The practice question
Imagine your key result does not replicate on the second run. What do you check first?
The made-up answer
In my postdoc, in a cell biology lab of about twelve people, I found that knocking down one gene cut cell migration by about 40 percent in a wound-healing assay. The second run showed nothing. Before changing anything, I listed what differed between the runs and tested one thing at a time. The image analysis was the first suspect, so I had a labmate re-score both runs blind, and her numbers matched mine, which ruled that out. Cell passage was next, but both runs were within two passages of each other, so I set that aside. Then I measured knockdown by qPCR on frozen samples from both runs: 85 percent in the first, only 30 percent in the second, and the second run had used a new lot of transfection reagent. I re-optimized with the new lot and ran three independent replicates, each with a scrambled control and a knockdown check. The effect came back at 35 percent, give or take 6. The lab now records knockdown efficiency in every run, and two other projects caught weak knockdowns that way in the following year.
Marked words are the quotes the rules would look at.
How the rules read it
The situationLevel 2 of 2
A specific case with where it was and how big: one assay in a lab of about twelve people, with a 40 percent effect that did not replicate.
The reasoningLevel 3 of 3
Two or more options compared, with why each was rejected: image analysis and cell passage were each ruled out with evidence before the reagent lot was found.
The resultLevel 3 of 3
Measured, and then adopted beyond the case and followed up over time: the replicated effect is measured, and the check became lab practice that caught later problems.
Together the three ratings give 8 of 8 points, which is 95 on the fixed scale. Its longest complete quote has 44 words, so the quote rule does not lower it. It is evidence mainly for Scientific reasoning.
What would raise it
This already earns 8 points. A follow-up would probe the statistics: which test was used on the three replicates, and whether the smaller effect changed what the result can claim.
The same story, told vaguely
One of my key results didn't replicate when I ran the experiment again. I went through the possible causes carefully, ruled out the obvious ones, and eventually found a problem with one of the reagents. Once I fixed it, the effect came back, and the lab changed its process so it wouldn't happen again.
A case without its context: no lab, no experiment and no size of effect, so situation is 1 and the area is at most 60. The causes it ruled out are not named.
A whole result: the sample result. More examples: strong interview answers.
Company interview links
Roles companies can set up
Starter templates a company can use for an interview link, and what each one steers the interviewer toward.
Analytical chemist
Focus: Method development and validation, HPLC and mass spectrometry, troubleshooting instruments, data integrity
Bioinformatics scientist
Focus: Analyzing sequencing and lab data, reproducible analysis pipelines, statistics, working with bench scientists
Environmental scientist
Focus: Field sampling, environmental data analysis, regulatory reporting, writing reports for clients or agencies
Lab manager
Focus: Lab safety, equipment and supplies, training lab members
Research scientist
Focus: Designing experiments, analyzing results, writing up findings
FAQ
Questions people ask
Should I talk about a failed experiment or a published result?
Whichever you can explain in more detail. A failure you troubleshot one change at a time often shows scientific reasoning better than a clean result, because it shows how you work when the data disagrees with you. The field's warning signs include changing several things at once and not knowing what helped, or overstating what a result shows.
How much statistics should I include?
Enough to show the analysis fits the design: the sample size or number of replicates, the test or model, and the uncertainty around the main estimate. Research design and rigor covers controls, statistics, reproducibility and limitations, and strong work matches the statistics to the design. You do not need formulas.
How is explaining my research to non-specialists scored?
Communication is scored on what you describe doing: a talk, a grant proposal, a poster or a plain summary, who it was for, and how you knew it landed. Answering clearly counts too, but communication shown only by how clearly you answered is capped at 80. So describe a real time you explained your work, not only your results.
My research is very specialized. Will the interviewer follow it?
It is told to follow up on what you say and to use the words people in your field use. If a question misses the point of your work, say so and steer it back.
Do I have to share confidential details?
No. The interviewer is told not to ask for names or confidential details about clients, patients, students or employers, and to let you describe situations without identifying anyone. Share unpublished results only as far as you are comfortable.
How to prepare for an AI interview
Guides for a company's AI interview:Mercor AI interviewmicro1 AI interviewHireVue interview
What this interview cannot do
It is one conversation of about 15 minutes, so it covers only part of what you can do, and we have not tested whether results predict job performance. The AI can misread an answer, which is why every score shows its quotes and people make the decisions. It is not a certification or license check. No independent bias audit has been done. Science and research was not one of the 8 fields in our own test (not an audit) of whether non-native English grammar changes scores.
Your interview, when you are ready
Related fields
Hiring scientists and researchers?
Interview links for science and research, and how to check a result a candidate sends you.