Candidate ghosting is a capacity problem

Because the queue outgrew the capacity, and silence is the only task you can skip without anyone inside the company noticing.

The numbers deteriorated fast: 53% of candidates were ghosted by an employer in 2026, up from 38% in 2024. Forty per cent were ghosted after a second or third interview. And 80% of recruiters admit to doing it, which rules out the theory of a few bad actors.

It is not a manners problem

No recruiter decides one morning to stop replying. What happens is more ordinary and harder to fix.

Applications arrive faster than they are processed. The pile grows. At some point every available hour has to be arbitrated between advancing live applications and closing dead ones. The first arbitration produces a placement; the second produces nothing measurable this month.

It is an incentive failure, not a character failure. Nobody inside the company notices a rejection that was not sent. The only person who sees it is outside, and they have no way to report it.

The cost is real but deferred. Only 26% of candidates say they had a good hiring experience, and median time to first offer reached 68.5 days in 2025, up 22% year on year. A market that takes more than two months to decide is a market where employer reputation travels fast.

AI created the problem before claiming to solve it

This is the uncomfortable part, and it has to be said before proposing anything.

Application volume exploded largely because candidates now apply with tools that write, tailor and send on their behalf. The unit cost of an application fell to almost nothing. Faced with that, hiring teams adopted silent rejection as a survival mechanism.

Put plainly: AI saturated the channel, and AI is now offered to absorb the saturation. That is a legitimate response, provided you recognise it as a race rather than a solution, and provided nobody sells it as a candidate-experience improvement when it is a catch-up.

That has a practical consequence. A tool that helps you answer a larger volume faster does not address the cause. What addresses the cause is filtering earlier and saying so earlier.

Seven days

The threshold is sharp. Median response time is 6.7 days, and beyond seven days the dropout rate at that stage doubles.

That is not a communications figure, it is an operational one. In staffing it translates directly: on a scarce profile, cybersecurity, cloud, data, seven days of silence is the window in which two other offers arrive. You do not lose the candidate because they judged you; you lose them because somebody replied first.

And the delay is paid twice: once on the current application, once on every subsequent one, since a profile left without an answer does not come back into your pool with the same enthusiasm.

What AI should draft: the no, not the yes

Here is the rule that seems right to me, and it is more restrictive than what most tools sell.

A rejection writes itself; it does not send itself. The rule does not move: a message leaving the company goes through approval, and a rejection is one. What the agent removes is the cost of writing, not the check.

In practice it is a queue. The agent prepares the rejections the same day, each naming the criterion it rests on: a missing skill, a level of experience, a location. A person runs through them in one pass and releases them. Fifty rejections reviewed in five minutes against fifty silences: that is the gain, and it is entirely in the delay.

The decision itself stays upstream and stays human. An agent that had rejected the profiles itself would have crossed the line; one that writes the message for a decision already taken does not.

Good news should not be automated. An interview invitation, an offer, an introduction to a client: those are the moments where the relationship forms, and delegating them to a machine saves five minutes by wasting the only interaction that mattered.

The interim update is the real opportunity. “Your profile is with the client, we are waiting for feedback, I will write on Thursday.” Nobody sends that message, it requires no decision, and it changes the lived experience more than anything else. It is a textbook first task to delegate.

What a rejection that costs nothing looks like

The fear that stops people automating rejections is sounding brutal. It is well-founded, and it is solved by wording rather than by silence.

A good rejection fits in three sentences. It gives the criterion: “the role required three years of Kubernetes in production”, because a precise reason is received as information, whereas a vague one is received as a judgement. It promises nothing you will not do: “we will keep your CV on file” is a line nobody believes, and it costs the credibility of the two sentences around it. It opens a real door if one exists, and nothing otherwise.

Two things to avoid absolutely. Generic compliments, “your profile is very interesting, but”, which read as boilerplate and cancel the criterion that follows. And the absence of a name: a rejection signed by a person, even when sent automatically, lands very differently from a message out of a no-reply mailbox.

The test is simple: reread the message imagining the candidate posts it on LinkedIn. If it makes you uncomfortable, it is not ready, and that is exactly the check a human should perform once on the template, rather than fifty times on each send.

A staffing firm also holds a card conventional employers do not. The rejection is almost never final: it is not “no”, it is “not on this assignment”. The sentence is true, it is useful, and it turns a rejection into a maintained relationship, provided, this time, that you follow up when another assignment opens.

Ghosting in the other direction

It is symmetrical, and partly a consequence of the first. A candidate left without an answer three times stops treating an answer as an obligation, and the consultant who does not show up on day one of an assignment is applying a norm they were taught.

For a staffing firm this is worse than for a conventional employer. Your talent pool is an asset you reuse for years: the same consultant will be contacted three, five, ten times across assignments. Every silence degrades the value of an asset you keep paying to maintain.

Reciprocity is the only lever available. A firm that answers systematically, even negatively, earns after two years a response rate to its own outreach that no sourcing tool can buy.

What matters more than speed: an explicit ending

One last point, and it costs nothing to put in place.

The worst thing is not slowness, it is the absence of an ending. A candidate who gets a clear rejection after ten days is better off than one who waits six weeks on a “we will get back to you” with no date.

So, two rules: never a status without a next-step date, and a systematic close, every application ends with a message, including and especially the ones abandoned midway because the need was filled another way. That is the most frequent case and the least handled.

It is also what separates a real agent from a plain automated send: a template fires when a field changes, whereas an agent knows a file has not moved in eleven days and nobody warned anyone. The difference is the one between a workflow and an agent, and this is exactly where it shows.