Safe candidate rejection feedback comes from the scorecard, not from the recruiter’s memory.
Quote 2 or 3 observable behaviours you recorded during the loop.
Tie each to a criterion you published in advance. Add 1 concrete thing to work on.
Send it within 48 hours. If the scorecard is empty, there is nothing safe to say.
“After careful consideration, we have decided to move forward with other candidates.”
The candidate spent 8 hours on your process. That sentence tells them nothing they can use.
Most companies send it because legal told them to. Writing something better takes time nobody has.
Both problems are solvable.
Why do companies send nothing?
Two real reasons and one bad one.
Specific feedback can be read as evidence in a discrimination claim.
A recruiter writing freely at volume is a genuine risk.
A recruiter closing 40 candidates a month has no time to write 40 thoughtful emails.
And nobody wrote down why the candidate was rejected, so there is nothing to say.
That third one is a process failure wearing a legal costume.
What makes safe feedback possible?
A scorecard with evidence in it. You cannot personalise what you did not record.
A line reading not a fit gives you nothing safe to send.
One line does the work.
“Did not establish requirements before designing. Could not name what breaks first under load.”
Specific, job-related, defensible.
The feedback problem is a scorecard problem.
What is safe to say?
Observable, job-related behaviour tied to criteria you published in advance.
In the system design round, the design worked and the requirements were not established first.
The test is whether you can point at the line in the scorecard.
And whether that line describes what the candidate did, not what they are.
Not safe: “You seemed nervous.” “You did not seem like a culture fit.”
Or “we were looking for someone more senior”, when the posting named no level.
Where does AI actually help?
By turning recorded evidence into readable prose at volume.
The interviewer writes terse factual evidence into the scorecard.
A model turns those lines into 2 or 3 paragraphs, using nothing else.
A human reads it and sends it.
That is a drafting tool. The judgement stays with the panel.
- The interviewer writes evidence into the scorecard as usual. Terse, factual, no polish.
- A model turns the scorecard lines into 2 or 3 paragraphs, using only what is in the scorecard.
- A human reads it and sends it.
2 guardrails matter. The prompt must forbid adding anything absent from the input.
And a human must approve every message.
Both are cheap and both are non-negotiable.
Where does AI not help?
In 3 places. Deciding who to reject, a regulated activity in the EU and a bad idea anyway.
Softening a decision nobody can justify. A well-written email invites a reply you cannot answer.
And personalising at the top of the funnel. A CV-screen rejection needs a clear fast no.
Save the detail for people who spent hours with you.
What to actually send
3 sentences of context. 2 or 3 specifics from the scorecard. 1 concrete thing to work on.
Keep the offer to answer questions if you can honestly staff it. Do not offer it if you cannot.
Send it within 48 hours. Feedback 3 weeks later reads as an afterthought, because it was one.
Why bother
51% of candidates say an interviewer’s attitude could make them withdraw. 72% share bad interview experiences publicly.
The people you reject this year are the people you want in 2 years.
And the people they talk to about you in the meantime.
A specific rejection is the cheapest reputation work available to a hiring team.
Common questions
Are you legally required to give rejection feedback?
In most European jurisdictions, no, and giving it is not itself a risk. The risk is feedback that describes the person rather than the behaviour. Or that cites a criterion you never published. Get your own legal read on the template once. Then reuse it rather than improvising per candidate.
How much detail should a rejection email have?
Scale it to what the candidate spent. A CV screen gets 3 lines and a fast no. Someone who did 4 rounds and 8 hours gets 2 or 3 specifics. Plus 1 thing to work on. Detail beyond the effort they put in reads as performance, not respect.
What if the candidate replies and argues?
Answer once, from the scorecard, and do not renegotiate the decision. Most replies ask whether the process was fair. They are not asking you to reverse it. A short reply pointing at the recorded evidence closes almost all of them. Only offer a reply channel you can staff.
Does giving feedback actually change anything for the company?
51% of candidates say an interviewer’s attitude could make them withdraw. 72% share bad interview experiences publicly. The people you reject this year are the people you want in 2 years. A specific rejection is the cheapest reputation work a hiring team has.
The short version
You cannot give useful feedback without a scorecard. Fix that first.
Use AI to turn recorded evidence into prose, never to decide anything.
Stay on observable, job-related behaviour. Send it within 2 days.