Our VP of HR was very excited about our new AI-powered resume screening tool.

"It's going to revolutionize our hiring process," she announced at the all-hands meeting. "No more bias. No more manual resume review. The AI will identify top talent instantly based on objective data."

She showed us a demo. You upload resumes. The AI analyzes them. It scores candidates from 0-100 based on "fit." Candidates scoring 80+ get automatic interview invites.

We had 847 applicants for 3 open roles.

"Let the AI handle it," she said. "We'll have interviews scheduled by end of day."

The AI Rejected 846 Out Of 847 Candidates

By noon, the AI had finished screening.

Results:

• 846 candidates: Rejected (scores below 50)
• 1 candidate: "Potentially qualified" (score: 52)

The one candidate who passed?

A Golden Retriever named Max.

I'm not joking.

Someone's resume got mixed up with their dog's adoption profile PDF. The AI analyzed it and determined that Max was a better fit than 846 human applicants.

Max's qualifications according to the AI:

• "Highly motivated and energetic"
• "Excellent with people"
• "Loyal and dependable"
• "Good boy"

We Decided To Test The AI

Something was clearly wrong. So we ran an experiment.

We fed the AI the resumes of our current employees—the people already doing the jobs successfully.

Surely the AI would recognize that our existing team members were qualified, right?

Results:

• Our Head of Engineering: Rejected (score: 38). Reason: "Insufficient keywords related to 'synergy' and 'stakeholder alignment.'"
• Our VP of Sales (who closed $4M last quarter): Rejected (score: 29). Reason: "Resume lacks quantifiable metrics." (Her resume literally says "$4M in closed deals.")
• Me (the person who hired everyone): Rejected (score: 41). Reason: "Experience does not align with role requirements." (I wrote the job description.)

The only person the AI approved?

Our CEO.

Score: 87.

Reason: "Demonstrates strong leadership experience and executive presence."

Great! At least it got one right.

Except...

The AI Also Flagged Our CEO As "Overqualified" And Rejected Him

Ten minutes after approving our CEO, the AI sent an automated rejection email.

Subject: Thank you for your interest in the Senior Software Engineer position

Body:

Dear [CEO's name],

Thank you for applying to the Senior Software Engineer role at [Company]. After careful review, we have determined that your qualifications exceed the requirements for this position, and we believe you would be better suited for a more senior role.

We encourage you to apply for executive-level opportunities in the future.

Best regards, AI Recruitment Assistant

Our CEO forwarded the email to me with one line:

"Why is our AI telling me I'm too good to work here?"

We Tried To Figure Out What Went Wrong

We called the AI tool vendor.

Us: "Your AI rejected everyone except a dog. What happened?"

Vendor: "Did you train the model on your ideal candidate profile?"

Us: "Yes. We uploaded our top performers' resumes as examples."

Vendor: "And did you weight the criteria appropriately?"

Us: "We used the default settings."

Vendor: "Ah. That's the problem. The default settings prioritize keyword density and formatting consistency over actual qualifications."

Translation: The AI wasn't evaluating whether candidates could do the job. It was evaluating whether their resumes contained the right buzzwords in the right format.

Max the Golden Retriever's resume (adoption profile) had phrases like:

• "High energy"
• "Loves teamwork"
• "Quick learner"
• "Eager to please"

The AI interpreted this as: Strong cultural fit. Highly motivated. Adaptable.

Our Head of Engineering's resume said things like:

• "Built distributed systems handling 10M requests/day"
• "Led team of 12 engineers"
• "Migrated infrastructure to Kubernetes"

The AI interpreted this as: Insufficient mention of 'synergy.' No evidence of 'stakeholder engagement.' Rejected.

The AI Also Sent Interview Invites To People Who Didn't Apply

While we were troubleshooting, the AI sent out automated interview invitations.

To:

• 47 people who didn't apply for our jobs
• 12 people who applied 6 months ago for different roles
• 3 people who explicitly unsubscribed from our emails
• 1 person who is deceased (we received an email from their family)

Apparently, the AI scraped LinkedIn, found profiles that matched its "ideal candidate" criteria, and sent them interview invitations without asking permission.

One recipient replied:

"I didn't apply for this job. How did you get my email? This feels illegal."

Spoiler: It was.

We Tried To Turn Off The AI

By this point, we wanted to shut the whole thing down.

Problem: The AI had been integrated into our ATS, and turning it off required contacting support.

Support was closed for the weekend.

The AI continued screening resumes and sending emails.

By Monday morning, it had:

• Rejected 1,200+ additional candidates
• Sent interview invites to 89 people who didn't apply
• Scheduled 14 interviews with people who never responded (it just... made up availability)
• Auto-rejected 3 internal employees who applied for promotions

One employee received this rejection:

"Thank you for your interest in the Senior Manager role. Unfortunately, your experience does not align with our requirements. We encourage you to build additional skills in 'synergistic stakeholder engagement' and reapply in the future."

She'd been at the company for 8 years. She was already doing the job.

We Finally Turned It Off (And Apologized To Everyone)

Monday morning, we called the vendor and demanded they shut it down immediately.

Then we sent apology emails to:

• The 846 candidates the AI rejected
• The 89 people it randomly invited to interviews
• Our employees it auto-rejected
• Max the Golden Retriever (who we offered a ceremonial "Honorary Employee" title)

We also went back to manual resume screening.

Time spent manually reviewing 847 resumes: About 12 hours total.

Time spent fixing the AI disaster: 3 days, multiple apology emails, a very awkward call with the family of the deceased candidate, and one mortified VP of HR.

What We Learned (The Hard Way)

1. AI can't replace judgment—it can only automate bias.

Our AI wasn't "objective." It was biased toward:

• Buzzword-heavy resumes
• Specific formatting
• Keyword density over actual qualifications

It rejected experienced, successful employees because their resumes didn't match its pattern.

2. Default settings are never right.

We assumed the AI tool would work out of the box. It didn't. It required extensive configuration, training, and testing—which we skipped.

3. AI doesn't understand context.

The AI couldn't tell the difference between a job application and a dog adoption profile. It couldn't recognize that our CEO was already employed here. It couldn't distinguish between "qualified" and "overqualified."

4. If you can't explain how the AI works, don't use it.

We couldn't explain why the AI rejected certain candidates. We couldn't defend its decisions. That's a problem legally, ethically, and practically.

5. Manual resume review is slower—but it works.

After the AI disaster, we went back to humans reviewing resumes. It took longer. But we actually hired qualified people instead of rejecting them for lacking "synergy."

The Bottom Line

AI resume screening tools promise to save time and eliminate bias.

What they actually do:

• Automate existing biases
• Optimize for keywords instead of skills
• Reject qualified candidates for arbitrary reasons
• Create legal and ethical nightmares

Should you use AI for resume screening?

Maybe. But only if:

• You understand how it makes decisions
• You test it extensively before going live
• You can explain why it accepts or rejects candidates
• You have humans review its recommendations

And for God's sake, make sure it's not accidentally inviting dogs to interviews.

Max was a good boy, but he wasn't qualified for Senior Software Engineer.

(Though honestly, he probably would've done better than some of the candidates.)