How I Write Rejection Emails That Don’t Sound Like They Came From a Robot
Last Tuesday I sat with a draft open on my screen for something like forty minutes. The email itself was twelve words. “Thank you for your application. Unfortunately, we will not be progressing.”
I’d written it, read it back, hated it. Closed the tab. Came back later, hated it again. Eventually just sent it at five past five because I needed to leave.
If you work in recruitment, especially at the junior end where it’s basically you, an ATS you’re still figuring out, and a hiring manager who replies to emails twice a week if you’re lucky, you probably know that particular feeling. Sending something you’re not happy with because the alternative is staying late to agonise over twelve words.
I’ve been in this role just over a year now and the rejection email problem has genuinely taken up more of my mental energy than probably anything else. It sounds ridiculous. But getting the tone right on a rejection is harder than it looks, and getting it wrong has actual consequences.
Why AI rejection emails are so obviously AI
So, about eight months ago I started using ChatGPT to draft these. I thought it would save me the agonising. Type in the context, get back a usable paragraph, send. Done.
First attempt: “We regret to inform you that after careful consideration of your application, we have determined that your profile does not align with the current requirements of this role.”
Nobody says that. I kept staring at it thinking, has anyone in the history of employment ever actually said “your profile does not align with the current requirements” to a real human person? It reads like a rejection generated by an HR compliance bot in 2009.
The problem with raw AI output for this kind of thing isn’t factual accuracy. It gets the facts right. It’s that the writing has no texture. It’s uniformly formal in a way that feels completely inhuman. Every sentence the same weight, the same register. No imperfections. No warmth. Candidates can tell, they’ve seen enough of these to recognise when something came from a template, and when they can tell, the rejection stops being just a no and starts feeling like evidence that the company didn’t bother.
That’s a reputational issue for the whole employer brand, but in practical terms it also just makes me feel bad every time I send one. I know what it’s like to get those emails from the other side.
So I started actually digging into how to humanize AI text in a way that would work for professional writing specifically. Not just grammar corrections. Something that would change the actual feel of the output.
What I tried first (and why it didn’t quite work)
I gave Grammarly a proper go for a few weeks. It’s a solid tool for what it does, but it’s really a grammar and style checker, not a tone converter. It’ll tell you if you’ve used passive voice too much or if your sentences are too long. It won’t tell you that your rejection email sounds like it came from a legal team. Those are different problems.
QuillBot I tried for maybe a week. The paraphrasing is good. But paraphrasing and humanizing aren’t really the same thing. Paraphrasing changes the words while keeping the structure and register. What I needed was something that would change the feel of the writing: the rhythm of it, how it lands when you read it. That’s a different level of intervention.
ngl I also tried just manually rewriting everything for a bit, which was just me spending twenty minutes on emails that should take five. Reading things back out loud, tweaking, still not being happy, sending them anyway. Not a real solution.
How I actually do this now
This is going to sound obvious but it took me a while to land on it.
I use ChatGPT for the content skeleton only. Role title, the actual decision, what the next step is. Nothing else. No filler phrases, no softening language. I strip all of that out because I’m going to be adding something better back in. The AI draft is raw material, not something I’d send.
Then it goes through Walter Writes. What the humanisation layer does is change the rhythm and register of the text rather than just substituting words. Contractions come in. The overly even sentence length breaks up. The policy-document formality drops off. It’s not making every decision for me, but it’s handling the baseline in a way that would otherwise eat twenty minutes of my day.
What I’m left doing is the judgment calls that no tool can make: what does this specific email actually need? A rejection after a quick first-call screen is a different email to one after someone’s done two stages and submitted a task. The investment is different. The emotional weight is different. Having the language baseline handled means I’m making one or two of those judgment calls instead of thirty.
I also specifically look for what I’d call false warmth. Some AI humanizers overcorrect and start adding phrases like “Your passion genuinely shone through” or “We were really impressed by your background.” That’s actually worse than robotic because candidates can read performative sympathy. I cut anything like that before it goes out.
The emails I send are almost always three short paragraphs now. Paragraph one: the decision, stated directly. Paragraph two: something genuine and specific about what the candidate did or how far they got. Paragraph three: next steps, or an invitation to apply again only if that’s true. Clean, predictable structure. Candidate finds the answer without having to read through layers of softening first.
The actual principles that make a difference
Took me a while to work these out. HR resources are not short on interview guidance. The post-interview communication is another matter.
The biggest one is specificity. “After reviewing a number of strong candidates” is meaningless and candidates know it. “We had a competitive shortlist and this one came down to a specific technical requirement” at least gives the rejection some actual shape. It’s not oversharing internal process. It’s just not hiding behind language so generic it could apply to literally anyone.
Say sorry once. One sentence. More than one and it reads as either anxious or performative. Acknowledge it, move past it.
Match the email to the investment the candidate made. Someone who applied through a job board and had zero contact with anyone doesn’t need three paragraphs. Someone who came in for two interviews and did a four-hour task deserves more than a form letter. This should be obvious but when you’re sending thirty in a day it requires conscious effort.
And please, stop writing “we’ll keep your details on file” if you’re not going to look at that file again. Candidates remember. It’s one of those small broken promises that adds up to the general impression that recruitment departments can’t be trusted.
The humanise AI text step matters for all of this because AI drafts tend to pile on politeness and professionalism until the email says nothing at all. The humanising step is what strips it back to something actually usable. Without it, the principles above get buried.
Tbh, the bar I’m actually aiming for is that the email should read like someone spent five minutes on it and cared about those five minutes. Not thirty seconds, approved, next. At volume, that genuinely takes tool support.
Being honest about scale
This whole thing works better in theory than in practice at scale, and I want to be straight about that.
When I’m sending thirty rejections in a day, I’m still working from a consistent base structure. The format doesn’t change. What changes is the middle paragraph: what’s specific to this person, this role, this stage, the one honest thing I can say about why we’re not progressing. Everything else stays consistent out of necessity.
The question of how to humanize AI text at scale is really about deciding where to put the variable effort. Mine goes into the specifics. Walter Writes handles the baseline conversion from stiff ChatGPT draft to something that sounds like a person typed it. I add the specific detail on top.
Between those two things, what I’m sending is better than what I was producing when I was writing everything from scratch while also dealing with interview scheduling, candidate questions, and the perpetual task of chasing hiring managers who’ve apparently forgotten what email is. Junior recruitment does not come with long uninterrupted blocks of writing time.
If you’re in a similar situation and you’ve been searching for an AI humanizer for professional writing, the thing to look for is texture change rather than vocabulary change. Tools built for paraphrasing produce writing that reads differently on the surface and identically in terms of feel. The mechanical quality is still there, just dressed differently. What you need is something that genuinely changes the rhythm.
I can’t claim this has fixed everything. But I get fewer one-word replies. More responses where someone thanks me and says they’ll keep an eye on future roles. Those weren’t happening much before.
Maybe that’s not how anyone else measures it. It’s how I’m measuring it. Let me know if it’s useful or if you’ve got a completely different approach. Genuinely curious what other people at this level are actually using.