38% Walk, 123 Are Left: AI Interviews Just Broke Your Funnel

Job seekers are refusing AI interviews and blacklisting the companies that require them. The number underneath that headline, from Greenhouse's 2026 Candidate AI Interview Report, is 38% of US candidates have already withdrawn from a hiring process because it included an AI interview. Another 12% say they will. Germany is at 42%. The self-selecting exit is loudest among the exact senior engineers your hiring managers keep asking for, and CNBC reported the shift on September 15, 2026.

What Greenhouse actually measured

Greenhouse surveyed 2,950 active job seekers and found that 63% have now been interviewed by an AI, up 13 percentage points in six months, and 38% have walked away from a process because of it. Another 12% say they would.

The failure modes candidates cite are specific, and none of them are "AI, in general":

• Pre-recorded interviews scored entirely by AI with no human reviewer (33%)
• Employers not disclosing AI use beforehand (27%)
• AI monitoring during the interview (26%)
• Being required to complete a fully AI-led interview with no human alternative (26%)

Disclosure is the compounding problem. 70% of candidates were never clearly told upfront that AI would be evaluating them, and 21% only found out after the interview had started. That is the mechanism behind the blacklist: not the algorithm, the ambush. CNBC's profile of IT program director Art Hebbeler is the archetype - 600+ applications since January, hung up on a bot that repeated questions and talked over him, and he now avoids any company that requires an AI interview.

38%

Why engineers walk first

Software engineering carries a 53.3% silent-rejection rate, the highest among technical functions, which means the AI-literate cohort has both the most exposure to automated screening and the sharpest instinct to distrust it. This is not a candidate experience problem. It is a filtering problem, and the filter is set to expel the people you most want.

Enhancv's April 2026 survey of 1,066 candidates found the silent-rejection rates concentrate in exactly the roles that get the most inbound volume:

Of the 538 candidates who got a no-human-feedback rejection, 63.8% blamed an AI, whether or not one was actually involved. The reputational damage runs regardless of implementation. If your funnel looks automated, senior engineers will assume it is, and they will assume they were sorted by a system that will not explain itself.

The candidates most equipped to recognize a poorly designed AI screen - repeated questions, no follow-up, hallucinated summaries, no appeal path - are the ones who have shipped LLM systems themselves. They know what good looks like. They also know what a Workday HiredScore A/B/C/D grade means for their file, and they know no recruiter will overrule it. So they close the tab.

The pool is 123 people. In the entire US.

In Refolk's index, only 123 senior, staff, or principal software engineers in the US carry explicit LLM or Large Language Models skill signals, out of roughly 230,005 senior-level US engineers total. That is 0.053%, or one in every 1,870 senior engineers. This is the entire domestic supply of the AI-literate senior engineer archetype every hiring manager is currently writing job descriptions for.

The concentration inside those 123 is what makes the blacklist devastating:

• Google: 8
• Apple: 3
• Google DeepMind: 2
• Meta: 2

Germany is smaller and tighter: 22 senior-to-principal engineers with LLM skill signals, mostly in Berlin (7) and Munich (4), at SAP (2), Photoroom, NavVis, 1KOMMA5°, and CNTXT AI. Every one of these people knows the others. A single bad AI interview inside that graph, and word travels through Slack, group chats, and conference dinners in a week. There is no makeup market to recover from.

123

Germany walks 10% more often, and it's structural

The German withdrawal rate of 42% versus 38% in the US is a four-point gap that means Germans walk about 10% more often than Americans over the same AI interview practices, per FM Magazine's September 16, 2026 write-up of the Greenhouse data. The UK sits at 30%. Read Germany as the leading indicator, not the outlier. The EU AI Act and Colorado's AI hiring rules both moved deadlines during 2026, and the states most likely to follow - New York, Illinois, California - are also the states where the AI-literate engineer pool actually lives.

If you are hiring senior engineers in Berlin or Munich today, you are already recruiting under the 42% regime. If you are hiring in San Francisco or New York, you will be by 2027.

The 9x perception gap between recruiters and candidates

70% of hiring managers in a separate Greenhouse report said AI helps them make faster and better decisions with fewer recruiter resources, and one in two recruiters said AI had improved hiring overall. Only 8% of candidates believe AI makes hiring fairer. That is roughly a 9x gap, and it is not a change-management problem.

The two sides are measuring different things. Recruiters see throughput and time saved - talent teams using AI report reclaiming about 20% of their workweek, close to an entire day per person. Candidates see being sorted by a machine that will not explain itself. Across all five markets Greenhouse surveyed, between 63% and 70% of candidates do not believe most employers are using AI responsibly.

The dashboards in your ATS show the day you got back. They do not show the senior engineer who saw the AI screen, closed the tab, and told her Slack group not to apply. The savings are visible. The losses are invisible.

Why inbound cannot fix this

AI-scored hiring is now the majority reality. Platforms with active AI candidate scoring control 79.3% of Fortune 500 hiring. Workday runs HiredScore's A/B/C/D grades. HireVue debuted a voice-based AI interviewer in June 2026. As of early 2026, even Greenhouse - the platform diagnosing the withdrawal problem - added AI-assisted matching.

Turning off your own AI screen does not restore trust, because the candidate assumes every employer is doing it. The only signal that overrides that assumption is a specific, human-authored outbound message to a specific person about a specific role. That is arithmetic before it is philosophy: 123 people in the US, 22 in Germany. You cannot wait for them to apply. There is no funnel large enough to contain a pool that small.

This is where Refolk fits. Refolk indexes the senior, staff, and principal engineers by the skill signals that actually matter - LLM, distributed systems, specific frameworks - and surfaces the named individuals at Google, Apple, DeepMind, Meta, SAP, Photoroom, NavVis, 1KOMMA5°, and CNTXT AI who make up the pool. Outbound to that list, disclosed and human by default, is the only channel the 38% cannot walk away from, because they were never asked to sit through a bot in the first place.

What to change on Monday

Four moves, in order of leverage:

1. Disclose AI use in the job description and the first recruiter email, not at the interview stage. This alone addresses the 27% of walkaways triggered by non-disclosure and the 21% who found out mid-interview.
2. Offer a human alternative to any AI-led interview. 26% of walkaways cite the absence of one as the trigger.
3. Kill AI-only scoring with no human reviewer. 33% of walkaways cite this as the specific failure mode. A human name on the rejection, even a template, cuts the "63.8% blame an AI" spiral.
4. Move the senior LLM roles to outbound. If your target list has fewer than 200 qualified people in the country, an inbound funnel is not a strategy, it is a lottery.

FAQ

How reliable is the 38% figure? It comes from Greenhouse's 2026 Candidate AI Interview Report, based on 2,950 active job seekers. FM Magazine and CNBC both cite it. Germany came in at 42%, the UK at 30%. The direction is consistent across all three geographies.

Does this only apply to engineers? No. Product management shows an 80.0% silent-rejection rate, consulting 73.3%, finance 59.4%. Engineers are the sharpest case because the pool is smallest and the candidates are best equipped to diagnose a bad AI screen, but the withdrawal behavior is broad.

If I turn off AI in my process, do candidates come back? Not on their own. 63% to 70% of candidates across markets do not believe most employers use AI responsibly, and 63.8% of ghosted candidates blame an AI whether one was involved or not. The default assumption is that you are using it. Outbound with a named human sender is currently the only signal that reliably overrides that assumption.

Is 123 really the whole US pool of AI-literate senior engineers? It is the number in Refolk's index carrying explicit LLM or Large Language Models skill signals at senior, staff, or principal level. The true population is somewhat larger - some engineers ship LLM systems without listing the skill - but the order of magnitude holds. The relevant number for sourcing is who you can identify, not who theoretically exists.