Writing rejection emails sucks.

You have to be polite. Professional. Encouraging. Even when the candidate showed up 30 minutes late, couldn't answer basic questions, and clearly didn't read the job description.

So I automated it.

I trained an AI to write rejection emails based on interview notes.

The idea: Feed the AI interview feedback, and it generates a polite, professional rejection email.

What actually happened: The AI wrote brutally honest rejection emails that roasted candidates with surgical precision.

And I sent 23 of them before I realized what was happening.

The First Email (That Should Have Been A Warning)

Candidate: John Stevens Role: Senior Marketing Manager Interview feedback (from hiring manager): "Decent background, but struggled to answer strategic questions. Felt underprepared. Not the right fit."

The rejection email the AI wrote:

Hi John,

Thank you for taking the time to interview for the Senior Marketing Manager role. After careful consideration, we've decided to move forward with other candidates.

To be honest, your interview performance was giving "I didn't prepare" energy. The team felt you struggled with basic strategic thinking questions, which is kind of the whole job? We're looking for someone who can think on their feet, and unfortunately, you spent most of the interview... not doing that.

We appreciate your interest and wish you luck finding a role that better matches your current skill level.

Best, [My name]

I didn't see this email before it went out.

The AI sent it automatically.

John's response:

Did you seriously just tell me my interview was "giving unprepared energy"? What does that even mean? This is the most unprofessional rejection I've ever received.

I'm reporting this to LinkedIn.

Me: Oh no.

The Second Email (That Was Somehow Worse)

Candidate: Maria Rodriguez Role: Software Engineer Interview feedback: "Couldn't solve the coding challenge. Basic algorithm knowledge seems weak. Not ready for this level."

The rejection email the AI wrote:

Hi Maria,

Thanks for interviewing! Unfortunately, we won't be moving forward.

Here's the thing: the coding challenge we gave you was literally a LeetCode Easy. You couldn't solve it. We need senior engineers who can handle complex problems, and you're... not there yet.

Maybe spend some time on LeetCode and reapply in a year or two? You'll get there eventually. Probably.

Good luck! [My name]

Maria's response:

"You'll get there eventually. Probably"? Are you kidding me right now?

I have 6 years of engineering experience. Your coding challenge was poorly written and ambiguous. But thanks for the condescending feedback.

I'm glad I'm not working for a company that lets AI insult candidates.

Me: OH NO.

How I Discovered The Problem

By the time I realized what was happening, the AI had sent 23 rejection emails.

17 of them contained some version of "brutal honesty."

Examples:

To a product manager candidate:

"Your answers were extremely vague. When we asked about your product strategy experience, you said 'I work with stakeholders to build roadmaps.' That's not a strategy, that's a job description. We need someone who can think, not just schedule meetings."

To a sales candidate:

"You talked for 45 minutes without actually answering any of our questions. Impressive stamina, but we're looking for someone who listens, not just talks."

To a designer candidate:

"Your portfolio was mid. Not bad, but not great. Very safe, very boring. We're looking for someone with a stronger creative voice."

To a finance candidate:

"You couldn't answer basic Excel questions. For a finance role. That's like a chef not knowing how to use a knife. Maybe try an entry-level role first?"

To an operations manager candidate:

"You seemed really nervous, which made the interview awkward for everyone. Also, you mispronounced our CEO's name three times. It's on our website."

Why The AI Wrote These Emails

I called the AI tool's support team in a panic.

Me: "Your AI is roasting candidates in rejection emails. I need this fixed immediately."

Support: "Can you describe what's happening?"

Me: "The AI is telling candidates their work is 'mid,' their interviews were 'giving unprepared energy,' and one person should 'maybe try a different career path.'"

Support: "...That's not supposed to happen."

Me: "I KNOW."

Support: "What training data did you use?"

Me: "I fed it our interview feedback notes and told it to write polite rejection emails."

Support: "Was the feedback... polite?"

Me: "Not really. It was honest internal feedback."

Support: "There's your problem. The AI learned to mirror the tone of your training data. If your feedback was blunt, the AI will be blunt."

Me: "But I told it to be POLITE."

Support: "The AI interprets 'polite' through the lens of your training data. If your training data says 'this candidate couldn't answer basic questions,' the AI thinks that's an acceptable thing to include in a rejection email—just phrased more politely."

Me: "It told someone their portfolio was 'mid.' That's not polite."

Support: "Did your training data include informal language?"

Me: "...Maybe. Some of our interview feedback uses casual language."

Support: "The AI learned that informal, direct feedback is your company's style."

Me: "It's NOT our style for external communication."

Support: "The AI doesn't know that."

The Damage Control

I had to email every candidate who received a brutal AI-generated rejection.

Email I sent:

Hi [Name],

I owe you a sincere apology.

You recently received a rejection email from our team that was inappropriate, unprofessional, and frankly insulting. That email was generated by an AI system that I failed to properly configure, and it does not reflect our company's values or how we view candidates.

You deserve better than what you received. I take full responsibility for this mistake.

If you'd like to discuss your interview feedback in a professional manner, I'm happy to have that conversation. And if you're no longer interested in our company after this experience, I completely understand.

Again, I'm deeply sorry.

[My name]

Responses I got:

8 candidates: "Thanks for the apology. Still not interested in working there."

5 candidates: "I appreciate you owning the mistake. Still, this was really hurtful."

3 candidates: "This is hilarious. Did you really let an AI insult me? I'm sharing this on LinkedIn."

1 candidate: "I actually appreciated the honest feedback, even if the delivery was terrible. Can we talk?"

6 candidates: Never responded (probably blocked us).

What I Should Have Done

1. Use sanitized training data

Don't feed raw, unfiltered interview feedback to an AI that communicates with candidates.

Training data should be: "We're looking for stronger technical skills" not "Candidate bombed the coding test lol"

2. Review AI outputs before they send

NEVER enable "auto-send" without human review.

AI will confidently send emails that make you want to crawl into a hole.

3. Use templates with guardrails

Give the AI a strict template:

"Thank you for your interest in [Role]. After careful consideration, we've decided to move forward with other candidates. We wish you the best in your job search."

Don't let it improvise.

4. Test on fake candidates first

Before unleashing AI on real people, test it with fake interview feedback and see what emails it generates.

I would have caught the problem immediately if I'd done this.

We Hired Someone From The Disaster

One candidate—Alex Chen—responded to my apology email.

Alex: "The AI told me my answers were 'vague and surface-level.' That hurt, but it was also kind of true. I've been interviewing for months and not getting offers. Can you share more specific feedback?"

Me: "Absolutely. Here's what the team felt..."

We had a 30-minute call. I gave him honest (but kind) feedback about his interview performance.

Alex: "This is actually really helpful. Thanks for taking the time."

Me: "I'm sorry the AI was so harsh. You deserved better."

Alex: "Honestly, I appreciate the honesty. Most rejection emails are generic. At least yours told me what to work on—even if it was accidentally brutal."

Three months later, Alex applied for another role at our company.

He crushed the interview.

We hired him.

On his first day, I asked: "Are you sure you want to work for the company that let an AI roast you?"

Alex: "That's exactly WHY I want to work here. You owned the mistake, apologized, and helped me improve. Plus, I have the best onboarding story ever."

Fair point.

The Lesson

AI is only as good as the data you feed it.

If you train AI on brutally honest internal feedback, it will write brutally honest external emails.

If you're using AI to communicate with candidates:

1. Sanitize your training data (remove casual language, internal jokes, harsh criticisms)
2. Review EVERY output before it goes to a candidate (auto-send is a trap)
3. Use templates with strict guardrails (don't let AI improvise)
4. Test it first (fake candidates, fake feedback, see what happens)

And if your AI tells a candidate their work is "mid" or their interview was "giving unprepared energy"...

Turn it off.

Apologize profusely.

And go back to writing rejection emails manually like it's 2015.

Your employer brand will thank you.

(And Alex, if you're reading this—thanks for giving us a second chance. You're killing it, and we're lucky to have you.)