How to run structured interviews with AI

I've made every hiring mistake.

Hired for vibe. Hired someone I liked at dinner. Hired someone who interviewed brilliantly and shipped nothing. Hired a "rockstar" who turned a team against them inside 90 days.

The pattern I kept ignoring is obvious in hindsight. I was running unstructured interviews. Different questions for every candidate. No rubric. No scorecard. Just a gut feeling at the end. It's basically the worst possible way to hire, and it's how most founders do it.

Turns out there's a better way. It's called structured interviewing, it's been proven to work for decades, and AI makes it genuinely easy to set up. You don't need an HR team. You don't need a recruiting background. You need a scorecard, a question bank, and about two hours to put it together.

This guide walks through the whole process, built on top of Micro's free hiring toolkit (part of the 200+ free AI skills we built for founders) and the methodology from Geoff Smart's Who and Google's structured hiring research.

Why unstructured interviews fail founders

An unstructured interview is any conversation where different candidates get different questions. Which is most interviews.

The problem is that unstructured interviews mostly test how well someone interviews. They don't test how well they'll do the job. A smooth talker who knows how to read a room will crush every unstructured interview. A quiet person who ships 3x more than anyone on your team might bomb one.

Laszlo Bock ran hiring at Google for 10 years and wrote Work Rules about it. His finding: structured interviews are roughly 2x more predictive of job performance than unstructured ones. Google ran the data, and gut-feel interviews are barely better than random selection.

The deeper problem for founders is that we're especially bad at this. We naturally gravitate toward people who think like us, who went to similar schools, who have similar energy. That's not a talent filter. That's a mirror filter. And a team of mirrors is a team with one blind spot repeated everywhere.

The fix isn't complicated. It's just discipline: same questions, same rubric, scored the same way.

What a scorecard is and why it changes everything

Geoff Smart's Who is the clearest framework I've found for hiring. The core insight: define what you need before you talk to anyone. Not a job description. A scorecard.

A scorecard has three parts:

1. Mission: One sentence. What does this person need to accomplish in the next 12 months? Not responsibilities. Outcomes. "Grow revenue from $500K to $1.5M ARR" beats "Own the sales process."

2. Outcomes: 3-5 measurable results that define success. These become your evaluation criteria at the end of the process. You're not asking "is this person smart?" You're asking "did they demonstrate they can hit these specific outcomes?"

3. Competencies: 5-8 behaviors or traits required to achieve those outcomes. Resourcefulness. Attention to detail. Comfort with ambiguity. Each one gets a 1-5 rating in the debrief.

Every interview question you write should map back to a competency. If you can't connect a question to something on the scorecard, cut it.

The scorecard-builder skill in our hiring toolkit generates a complete scorecard in about 10 minutes. You give it the role, your stage, and what success looks like, and it outputs the mission, outcomes, and competency list. You edit and own it. AI generates the starting point so you're not staring at a blank page.

A scorecard built before you source candidates changes the whole dynamic: you're evaluating people against a standard, not against each other.

Do structured interviews actually work?

Short answer: yes, definitively.

Structured interviews predict job performance at roughly twice the rate of unstructured interviews, according to research from Google's people analytics team. Work sample tests (giving candidates an actual piece of work to complete) are even more predictive. But structured behavioral interviews are the most practical high-signal tool most founders can realistically run.

The key word is behavioral. Behavioral questions ask candidates to describe what they actually did in past situations, not what they would do hypothetically. "Tell me about a time you had to push a project forward without clear direction" reveals real behavior. "How would you handle a situation with unclear direction?" lets people describe their ideal self, not their actual self.

Thomas Ptacek's "The Hiring Post" is the most rigorous takedown of traditional tech interviewing I've read. His conclusion mirrors the research: unstructured interviews are barely better than coin flips. Work samples and structured evaluation are the fix.

The other thing that makes structured interviews work is consistency. Same questions in the same order for every candidate. When you ask different people different questions, you're not measuring them on the same scale. You can't compare notes in a debrief if two people got different interviews.

Structure doesn't kill good conversation. It gives it a foundation.

How to generate role-specific interview questions with AI

This is the part that used to require an experienced recruiter or a lot of time with a question bank. AI handles it well.

Here's the approach:

Start with your scorecard competencies. For each competency, you need 2-3 behavioral questions. If resourcefulness is on your scorecard, you need questions that surface how a candidate has actually been resourceful under pressure. Not "are you resourceful?" but "tell me about a time you had to figure something out with almost no resources."

Use the STAR probe sequence. Every behavioral question gets followed up the same way:

• Situation: What was the context?
• Task: What were you responsible for?
• Action: What did you specifically do? (push hard here)
• Result: What happened? What did you learn?

The follow-ups are where the real data lives. Candidates prepare polished openers. The probing questions are harder to fake.

Generate with AI, refine for your context. The structured-interview-guide skill in our toolkit takes your scorecard and generates a full question bank: 2-3 behavioral questions per competency, plus follow-up probes, plus a suggested sequence. A 45-minute interview guide for a first engineering hire takes about 5 minutes to generate.

What I do is take the AI output and edit it to reflect my actual company. Generic questions get replaced with questions that reflect what we've actually been through. "Tell me about a time you had to ship something imperfect under time pressure" hits differently when you run a startup than when you work at Google.

The question bank is the artifact that makes every subsequent hire easier. Build it once, refine it, and it compounds.

How to run the interview itself

Structure doesn't mean robotic. The best structured interviews feel like real conversations because the interviewer is fully present instead of improvising questions on the fly.

A few things that actually matter:

Set the agenda at the start. Tell the candidate: "We'll spend about 40 minutes working through a few specific questions about your past experience, then leave 10 minutes for your questions." This puts them at ease and signals that you're organized. Candidates notice.

Take notes during the interview, not after. Memory fades fast and gets contaminated by impressions. Write down what candidates actually said, word for word when possible. These notes are your debrief material.

Don't react emotionally to answers. Nod. Stay neutral. If you visibly light up when someone gives a good answer, you've just telegraphed what you want to hear, and smart candidates will mirror it back for the rest of the interview.

Interview intelligence tools make this much easier. Metaview auto-generates structured notes and maps responses to your scorecard competencies in real time. Granola runs locally, works with any meeting tool, and has 29 templates including interview-specific formats. Fathom is free and captures the full transcript. Any of these eliminates the "I had a feeling about them" problem because you have an actual record.

The hiring toolkit's interview intelligence section reviews all of these with startup-specific recommendations on which to use at which stage.

The goal of the interview is evidence collection. You're building a case, not having a chat.

How to score candidates and run a debrief

The debrief is where structured hiring actually pays off, and where most founders blow it.

Common mistake: someone walks into the debrief and says "I really liked them" before anyone else has spoken. Within 60 seconds, the whole room has anchored on that opinion and the scoring is contaminated. This is called anchoring bias, and it's why debrief order matters.

The right debrief format:

1. Silent scoring first. Everyone fills out their scorecard independently before the debrief starts. 5 minutes. No talking.

2. Round-robin recommendation. Each interviewer states their hire/no-hire recommendation and their single biggest reason. No debate yet.

3. Competency deep dive. For each scorecard competency, go around the room. What evidence did each person collect? This is where the notes matter.

4. Open discussion. Now you debate. But the debate is about evidence, not impressions.

The AI debrief tool in the hiring toolkit can run this as an interactive session, prompting each interviewer for their ratings and synthesizing the results. For solo founders, it's useful just for organizing your own notes into a structured output before you make a call.

The decision rule from Who: if a candidate doesn't get a "yes" rating on most of the core competencies, the answer is no hire. Don't talk yourself into a hire because the candidate was charming or because you're desperate to fill the role. Desperation is the single most common cause of bad startup hires.

A disciplined debrief turns subjective impressions into a decision you can defend.

Reference checks aren't a formality

Most founders treat references as a rubber stamp. Call two people the candidate pre-selected, listen to them say nice things, and move on. That's not a reference check. That's a courtesy.

The Who Method approach is different. You ask the candidate to arrange the calls themselves. That's fine. But then at the end of every reference conversation, you ask: "Who else should I talk to who worked closely with [name]?" The second-degree references are the ones who give you the real picture.

The calibrated question technique (borrowed from FBI negotiator Chris Voss): "On a scale of 1-10, how would you rate their performance?" Then: "What would make it a 10?" The follow-up is where people actually tell you things. A reference who says "9 out of 10, honestly just needs to improve on communication sometimes" has just told you something real.

AI can help you build a structured reference check guide the same way it helps with interview questions. The reference-check-guide skill generates a set of calibrated questions tailored to your role and the specific competencies you're evaluating. Takes 3 minutes to set up.

References done right are the highest-signal thing in the entire process. Most founders skip past them and pay for it later.

Putting it all together with the hiring toolkit

Here's the sequence I'd run for any first hire today:

1. Use the agentic-hiring-setup skill to get oriented. It asks about your stage, role, and urgency, then routes you to the right tools.
2. Build your scorecard with the scorecard-builder. Mission, outcomes, competencies.
3. Generate your question bank with the structured-interview-guide skill.
4. Run interviews with notes or a tool like Granola or Fathom.
5. Score with the interview-scorecard skill.
6. Run a structured debrief using your notes.
7. Run reference checks with the reference-check-guide skill.

The entire toolkit is free. It's built on the Who Method and Google's structured hiring research, translated into AI skills you can run today.

Founders making their first hires are in the hardest position. You don't have an HR function. You've probably never run a formal process. The stakes are insanely high because a bad first hire at a 5-person company is catastrophic in a way that it just isn't at a 500-person company.

Structure isn't about bureaucracy. It's about making a decision that's hard to undo with the best information you can possibly collect.

Start with the hiring toolkit. It's the closest thing to having Geoff Smart sit next to you through the whole process.