The Candidate Experience Cost of AI Screening, Measured Honestly

Around 70 percent of candidates complete an AI screening call, and that number is more useful than it looks. A screen with a low completion rate is not filtering for capability. It is filtering for tolerance, and the candidates who abandon first are disproportionately the ones with alternatives. Publishing this honestly matters because the cost is real and it is the part of the AI screening argument vendors leave out.

Last reviewed July 2026. Completion figures from Sortinghat delivery data, stated as directional.

What completion rate actually measures

Not whether candidates are suitable. Whether the process was worth their time.

A screen that takes twelve minutes to ask questions a two-minute conversation could cover produces abandonment, and the people who abandon are not randomly distributed. Someone with three live processes will not spend twelve minutes on a speculative one. Someone with no alternatives will.

That means a badly designed automated screen systematically filters out the strongest part of the market while reporting a completion statistic that looks like a candidate quality signal.

The three things that move it

Length

Every additional question costs completion. The discipline is to cut anything that does not trace back to a weighted criterion, which usually removes a third of a first draft. Three to seven minutes is the range where completion holds.

Clarity about what happens next

A candidate who does not know whether a person will see this, or when they will hear back, disengages faster. Saying both at the start costs fifteen seconds and buys more completion than any interface improvement.

A visible human path

Counterintuitively, knowing you can speak to a person makes people more likely to complete the automated version. The opt-out is rarely used and its visibility does most of the work.

Why disclosure helps rather than hurts

The instinct is that telling candidates a call is automated will reduce completion. In practice the opposite holds, because the alternative is a candidate working it out mid-call and feeling misled.

A candidate told at the outset can decide whether to proceed. A candidate who realises at minute three has been given no choice, and that is where the resentment comes from rather than from the automation itself.

Disclosure is also a legal requirement in a growing number of jurisdictions. The EU AI Act classifies recruitment AI as high-risk with transparency obligations, and several US states have their own rules. Doing it because it works is a better reason than doing it because you must.

The trade you are actually making

Honest framing: automated screening buys throughput and spends candidate goodwill. Whether that is a good trade depends entirely on relevance.

A well-targeted call to somebody who genuinely fits the role costs them four minutes and may produce a job. Three hundred of those perform better than forty unfiltered calls, because 300 relevant people are less annoyed than 40 irrelevant ones.

A poorly targeted one at scale is worse than doing nothing, because you have now annoyed 300 people instead of 40. The volume amplifies whichever way the targeting went.

What to measure instead of completion alone

Completion by seniority. If it falls sharply at senior levels, the screen is wrong for that segment rather than the segment being wrong.

Opt-out rate. If it climbs when volume climbs, the targeting is the problem, not the tooling.

Post-screen conversion. Completion is worthless if the people who finish do not convert. High completion with low conversion means the screen is easy rather than useful.

Unprompted candidate feedback. Rare, and worth more than a survey nobody fills in.

Frequently asked questions

The number to segment before you defend it

Split your completion rate by seniority. If it holds at junior levels and falls at senior ones, the screen is not working for the roles where a bad hire costs most, and the aggregate figure is hiding it.

Then check your opt-out rate against volume. If it rises as you scale, the targeting is wrong and scaling further makes the brand damage worse rather than the throughput better.