Looking for work with a US company? Apply to the Rolemote talent roster — free →

← All notes

November 16, 2026Hiring process6 min read

Interview questions AI can answer: rethinking screening

Contents

Type 'bookkeeper interview questions' into any search engine and you'll find thirty listicles, each with the top ten answers. Type the same phrase into any AI assistant and you'll get polished, confident responses in under five seconds. The candidate sitting across from you on a video call has had the same access since 2023.

That doesn't mean interviewing is useless. It means the specific questions most hiring guides recommend — behavioral walk-throughs, hypothetical strength-and-weakness prompts, 'tell me about a time you…' openers — have become a test of preparation rather than a test of capability. If you are using those questions to decide whether to pay a recruiting fee, you are paying to learn how well someone read a prep guide.

Which questions are fully solvable by AI

A question is 'AI-solvable' if a confident, plausible-sounding answer can be generated without the candidate knowing anything specific about your company, your tools, or the actual problem the role will face. Most classic interview questions clear that bar easily.

Tell me about yourself. What's your greatest weakness. Describe a conflict with a coworker and how you handled it. Where do you see yourself in three years. Why do you want to work in this industry. Every one of these has a known ideal-answer structure — STAR format, the 'weakness that is secretly a strength' template, the 'growth mindset' closer. AI can generate five polished variants of each in seconds, and candidates have been pasting those into prep documents for the last two years.

The harder category is role-specific knowledge questions: 'What's the difference between cash and accrual accounting?' or 'Walk me through how you'd set up a Google Ads campaign.' These are harder to fake — but only slightly. Answers to common technical questions exist verbatim on YouTube, in certification course transcripts, and in AI training data. A candidate who has never reconciled a bank account can still describe the process accurately because the process is documented everywhere.

What AI cannot answer well

AI generates text that fits a known pattern. What it cannot do is complete actual work on your actual problem, make a judgment call in a novel situation with competing priorities, or reveal what a person does when the right answer is not obvious.

A graded work sample, 'Here is a messy expense report from last quarter, categorize it the way we'd need it for our accountant', has no published model answer. The candidate either does it correctly or they don't. A scenario question that describes a situation specific enough to be real, 'Your US client asks you to reschedule four back-to-back calls on a Friday afternoon to fit a last-minute investor meeting; here are the constraints; what do you send?', also has no template. The answer depends on judgment, not recall.

This is the structural difference: performance on a task you define reveals capability; performance on a question with known ideal answers reveals preparation. Both matter, but only one is predictive of whether the hire will work out.

The interview's remaining legitimate uses

None of this means you should skip talking to finalists. The interview still does things a work sample cannot. It surfaces communication style and pace, does this person ask clarifying questions, or do they barrel through? It lets you describe what success looks like in month six and gauge whether the reaction is curiosity or anxiety. It gives the candidate a chance to ask questions, which is often more revealing than their answers.

What the interview is not built for: filtering candidates from a large pool, testing technical knowledge, or predicting whether someone will stay past month three. Using it for those jobs wastes the interview's actual value.

The practical implication: run the work sample and scenario judgment before the interview, not after. By the time you're on the call, you should already have a scored shortlist. The interview is a final check on fit and communication, not the primary filter.

What a rubric-based screen actually measures

The alternative to unstructured interview questions is a scored rubric applied before anyone gets on a call. A rubric that is worth using has to be specific enough that two different reviewers would score the same candidate roughly the same way, and public enough that candidates can see what they're being evaluated on.

A rubric built around graded work samples, scenario judgment, experience specificity, written English, and salary-band fit handles the problem directly. Work samples test whether the candidate can do the task. Scenario judgment tests whether they reason well under trade-offs. Experience specificity, asking for concrete details rather than general claims, is harder to fake with AI because it requires accurate recall of real events, not a plausible narrative. Written English and salary fit are hygiene checks that keep the pool realistic.

The weights matter too. If your rubric puts 70% of the score on a structured interview and 30% on everything else, you have built a system that selects for people who interview well. That's a different skill from the one you're hiring for.

Applied example: screening a bookkeeper

Say you're a seven-person e-commerce brand hiring a part-time bookkeeper in the Philippines. The salary band for that role in the Philippines, based on our published research, runs $800-1,500 per month.

If you post on a job board and schedule calls with everyone who applies, you will spend most of your screening time on candidates whose strengths are preparation, not accuracy. The candidate who interviewed confidently and walked you through the accrual method in smooth sentences may still categorize a Stripe payout incorrectly on their first week of real work.

A work sample, a small, realistic reconciliation exercise using anonymized data similar to what the role would actually handle, sorts the pool in a way that no interview question can. It is also harder to fake: either the numbers reconcile or they don't. A scenario question about what to do when the owner's personal Amazon purchase appears in the business account tests judgment that textbooks don't cover. Neither requires you to be on a call.

How Rolemote handles this

Our screening rubric is published at rolemote.com/how-we-screen, so candidates know exactly what's being measured. The weights: graded work sample 35%, scenario judgment 25%, experience specificity 20%, written English 15%, salary-band fit 5%. Every score is based on those criteria. You read the scored finalists before you pay anything, the whole search runs free up to the point where you ask to meet candidates, at which point a flat fee applies rather than a percentage of salary.

Typical offshore staffing agencies in this category charge 25-35% of first-year salary, which works out to roughly $4,500-6,300 on common roles; our flat fee is $1,995. The re-run is free if no finalist clears your bar. The point of publishing the rubric is not marketing, it's that a black-box screening process has the same problem as an interview: you can't tell what it actually measures.

If you want to test the screener before committing to anything, rolemote.com/try-the-screener is public.

What to actually do differently

Replace your opening interview filter with a short, role-specific work sample. Keep it under 45 minutes of candidate time. Use a scenario question that describes a real situation from your business, not a generic 'how would you handle a difficult stakeholder' prompt, but the actual kind of situation this person will face in week three.

When you do get on a call, ask about specifics: not 'tell me about a time you managed a complex project' but 'in the last role you listed, what tool did you use for task tracking and what broke down about it?' Specific questions about real past situations are harder to script in advance and harder for AI to invent plausibly, because they require accurate recall, not pattern-matching.

Score before you meet. By the time you are on a video call, you should already have confidence in the candidate's ability to do the work. The call is to confirm the person behind the score, not to discover them.

Common questions

Are behavioral interview questions still worth asking?

They can be, but only if they're specific enough to require real recall. 'Tell me about a time you handled a conflict' has a template answer. 'In your last role, what broke down in the first month and what did you do about it?' is harder to script. The more specific the question, the less AI preparation helps.

How long should a work sample take a candidate?

Under 45 minutes is a reasonable ceiling for unpaid screening tasks. Longer than that and you lose good candidates who have competing offers. The task should be realistic enough to reveal actual skill, a small, concrete piece of work similar to what the role handles in week one.

Can't candidates use AI to complete work samples too?

Sometimes, depending on the task. A scenario that requires judgment about your specific business context is harder to outsource to AI than a generic exercise. For roles where AI-assisted completion is a real risk, a short follow-up question asking the candidate to explain their reasoning out loud on a call will usually surface whether the work was theirs.

What's the difference between experience specificity and just asking for a resume?

A resume lists titles and dates. Experience specificity means asking for concrete details: exact tools, specific numbers, what went wrong and why. AI can generate a plausible-sounding resume narrative; it cannot accurately recall details from work the candidate never actually did.

Does the order of screening steps matter?

Yes. Running the work sample and scenario questions before the interview means you arrive at the call with scored data. The interview then functions as a final fit check rather than the primary filter. That order uses each step for what it's actually good at.

Hiring one of these roles?

Describe the role and our screener runs the whole search — you read scored finalists before paying anything.

Start a free search

Free to start — no card. Pay only to meet finalists. Free re-run if none clear your bar.

Hiring one of these roles?

Describe the role and our screener runs the whole search — you read scored finalists before paying anything.

Start a free search

Free to start — no card. Pay only to meet finalists. Free re-run if none clear your bar.