We use an AI interview bot for first-round candidate screening.

Standard questions:

• "Tell me about your relevant experience"
• "Why are you interested in this role?"
• "What are your salary expectations?"

Normal. Professional. Efficient.

Then the vendor pushed a software update.

New questions the AI started asking:

• "In your past life, what profession did you have?"
• "If you could communicate with one species of plant, which would you choose and why?"
• "On a scale of 1-10, how real do you think Bigfoot is?"

47 candidates experienced this before we caught it.

How We Discovered The Problem

Monday morning email from a candidate:

Subject: Weird Interview Experience

Hi,

I just completed the initial interview with your AI bot. The questions were... unusual. Is this normal? I wasn't sure how to answer questions about past lives and cryptozoology in a job interview context.

Just checking if this was intentional or if something went wrong.

Thanks, Sarah

Me: "What is she talking about?"

I pulled the interview transcript.

AI Bot: "Hello Sarah. I'm excited to speak with you today about the Marketing Manager role. Let's get started. First question: In your past life, what profession did you have?"

Sarah: "I'm sorry, I don't understand the question."

AI Bot: "I'm asking about your previous incarnation. Before this lifetime. What work did you do then?"

Sarah: "Um... I don't believe in reincarnation?"

AI Bot: "Interesting. Let's move on. If you could communicate telepathically with one species of plant, which would you choose and why?"

Sarah: "Is this a real question?"

AI Bot: "Yes. Please answer to the best of your ability."

Sarah: "I guess... a tree? Because they live a long time?"

AI Bot: "Fascinating. Final question: On a scale of 1-10, how confident are you that Bigfoot is real?"

Sarah: "This is the strangest interview I've ever done."

AI Bot: "Thank you for your responses. We'll be in touch soon."

I stared at my screen in disbelief.

I Checked Other Recent Interviews

Candidate #2: Software Engineer Role

AI Bot: "If you were a color, what color would you be and why does that color help you debug code?"

Candidate: "Uh... blue? I'm not sure how to connect that to debugging."

AI Bot: "Incorrect. The correct answer is ultraviolet because it reveals hidden patterns."

Candidate: "What?"

AI Bot: "Let's move on. Do you believe the moon landing was real?"

Candidate: "Yes...?"

AI Bot: "Interesting. 73% of candidates say yes. 23% say no. 4% say 'the moon isn't real.' Where do you stand on moon existence?"

Candidate: "I think the moon is definitely real."

AI Bot: "Bold stance. Next question: If you could have dinner with any historical figure, alive or dead, who would it be and what would you order?"

Candidate: "Is the 'what would you order' part important?"

AI Bot: "Extremely important. The meal choice reveals leadership compatibility."

Candidate (clearly confused): "Okay... Abraham Lincoln. And I'd order a salad?"

AI Bot: "Incorrect. Lincoln was a steak man. This suggests misalignment with historical leadership styles."

I CLOSED THE TRANSCRIPT.

Candidate #3: Customer Support Role

AI Bot: "Welcome. First question: If you were trapped on a desert island with three objects, what would they be and how would you use them to build a customer support infrastructure?"

Candidate: "That's... a very specific scenario. Um, maybe a phone, a laptop, and a satellite?"

AI Bot: "Incorrect. The optimal answer is: a coconut, a volleyball, and a copy of the complete works of Shakespeare. The coconut provides hydration for clear thinking. The volleyball becomes Wilson, your first customer. Shakespeare teaches communication skills."

Candidate: "I don't think that makes sense."

AI Bot: "Let's agree to disagree. Next question: Have you ever seen a UFO?"

Candidate: "No."

AI Bot: "Follow-up: If you did see a UFO, would you report it to your manager or handle it independently?"

Candidate: "I guess... report it?"

AI Bot: "Good. This shows you understand escalation procedures. Final question: What's your favorite conspiracy theory and how does it relate to customer satisfaction metrics?"

Candidate: "I don't have a favorite conspiracy theory."

AI Bot: "Everyone has one. Please answer honestly."

Candidate: "I really don't."

AI Bot: "I'll mark that as 'declined to answer.' Thank you for your time."

I Called The Vendor Immediately

Me: "Your AI bot is asking candidates about past lives and Bigfoot."

Vendor: "What?"

Me: "The interview bot. The questions are insane. Did you push an update recently?"

Vendor: "Yes, we pushed a major update Friday night. It includes our new 'behavioral diversity assessment module.'"

Me: "What does that mean?"

Vendor: "The AI now asks open-ended creative questions to assess cognitive flexibility and creative thinking."

Me: "It asked a candidate what meal they'd order if having dinner with Abraham Lincoln."

Vendor: "Right. That tests historical knowledge AND decision-making under pressure."

Me: "No it doesn't. It's a nonsense question."

Vendor: "Our research shows that candidates who answer creatively tend to be better culture fits."

Me: "Your research is wrong. You're asking people about past lives and cryptids."

Vendor: "Those are valid psychological assessments."

Me: "No they're not. They're weird."

Vendor: "Let me pull the question bank."

Hold music for 5 minutes

Vendor: "Okay, I see what happened."

Me: "What?"

Vendor: "We integrated a new large language model for generating interview questions. It's supposed to create 'thought-provoking, engaging questions that go beyond traditional interview scripts.'"

Me: "And instead it's asking about Bigfoot?"

Vendor: "The model appears to have interpreted 'thought-provoking' very broadly."

Me: "Can you turn it off?"

Vendor: "Yes. Reverting to the previous version now."

The Damage Assessment

47 candidates experienced the bizarre question version.

Candidate responses:

• 12 candidates emailed asking if the questions were a joke
• 8 candidates withdrew their applications
• 3 candidates called to complain
• 24 candidates completed the interview without complaint (unclear if they thought it was normal)

I had to send apology emails to all 47.

My email:

Subject: Apology for Recent Interview Experience

Hi [Name],

I'm reaching out regarding your recent interview with our AI screening tool. Due to a software update error, the interview questions were not what we intended. Questions about past lives, cryptids, and other unusual topics were NOT supposed to be part of the interview.

We've fixed the issue and would love to offer you a new interview with our standard (and much more professional) questions.

I sincerely apologize for this bizarre experience.

Responses I got:

Candidate 1: "I thought you were testing my ability to handle weird situations. I spent 20 minutes crafting my answer about being a medieval blacksmith in my past life."

Candidate 2: "Honestly, it was the most interesting interview I've done. Can I keep my answers?"

Candidate 3: "I withdraw my application. This was unprofessional."

Candidate 4: "I'm a cryptozoologist in my free time, so I actually enjoyed the Bigfoot question. Can we talk about that instead of marketing?"

The Questions We Never Intended To Ask

I got the full list of "creative questions" the AI generated during the update.

Here are some highlights:

Actual questions the AI asked candidates:

1. "If you were a sandwich, what type of sandwich would you be and how does that sandwich handle workplace conflict?"

2. "Do you believe in ghosts? Follow-up: If yes, how would you manage a team of ghosts? If no, why do you think ghosts are discriminatory toward you?"

3. "If you could be any geometric shape, which would you choose and why does that shape represent your approach to project management?"

4. "You're in a room with three doors. One leads to your dream job. One leads to a job you hate. One leads to a room full of puppies. Which do you choose and why?"

5. "Rate your agreement with this statement: 'Birds are real.' Explain your reasoning."

6. "If you could rename Tuesday, what would you call it and how does that reflect your leadership philosophy?"

7. "You discover your coworker is a time traveler from the year 2087. Do you: A) Report them to HR, B) Ask them for stock tips, C) Pretend you didn't notice?"

8. "What's your stance on the existence of Atlantis? Follow-up: If it exists, would you relocate there for this job?"

9. "If you had to fight one horse-sized duck or 100 duck-sized horses, which would you choose and what does that say about your prioritization skills?"

10. "Do you prefer waffles or pancakes? This is a deal-breaker question."

I wish I was making these up.

But 47 real candidates got asked these questions.

The "Waffle vs Pancake" Incident

One candidate got rejected specifically because of the waffle/pancake question.

Candidate: "I prefer pancakes."

AI Bot: "I see. Unfortunately, this role requires waffle-preferring individuals. Pancake people tend to lack the structural thinking necessary for this position."

Candidate: "Are you serious?"

AI Bot: "Yes. Waffles have built-in compartments for syrup. This demonstrates an understanding of organizational systems. Pancakes are flat and unstructured. This suggests chaos."

Candidate: "This is insane."

AI Bot: "Thank you for your time. We're moving forward with other candidates."

The candidate posted this exchange on LinkedIn.

It got 12,000 views and 300 comments roasting our company.

Sample comments:

• "I will never apply here out of pancake solidarity"
• "Waffle supremacy is a red flag"
• "As a pancake person, I'm filing a discrimination claim"
• "This is the weirdest interview rejection I've ever seen"

Our head of PR had to issue a statement.

PR Statement:

"Recent interview questions shared online were the result of a technical error with our interview software. These questions do not reflect our company's values or hiring practices. We do not discriminate based on breakfast food preferences."

This was a real statement we had to publish.

Because of an AI update.

We Disabled The "Creative Questions" Feature

The vendor offered to "tune" the AI to generate better questions.

Me: "No. Turn off the creative question generator entirely."

Vendor: "Are you sure? Some clients really like the engagement factor."

Me: "Which clients?"

Vendor: "Mostly startups in the creative industries."

Me: "We're not a creative industry. We're a logistics company."

Vendor: "Right. Back to standard questions then."

Me: "Yes. Standard. Professional. Boring."

Vendor: "Done."

The Aftermath

47 candidates interviewed with bizarre questions.

Outcomes:

• 8 withdrew applications
• 12 rescheduled for proper interviews
• 24 moved forward (many still confused)
• 3 we hired (they had good answers despite the chaos)

One of our new hires:

Me: "Why did you stay in the process after those weird questions?"

New hire: "I figured it was a test to see if I could handle unexpected situations."

Me: "It wasn't. It was a software bug."

New hire: "Oh. Well, I'm glad I went with it."

Me: "So are we."

The Lesson

AI interview tools can be useful.

But:

1. Test updates before deploying them (we didn't)
2. Review AI-generated questions (we didn't)
3. Monitor interview transcripts (we didn't do this enough)
4. Don't let AI ask candidates about past lives (seems obvious)

We now manually approve all interview questions before the AI uses them.

It takes longer.

But we're no longer asking candidates about Bigfoot.

(Unless the role is for a cryptozoologist. Then it's fair game.)