42% Walk Out of AI Interviews. Chicago Booth Data Says Stay.

You clicked the interview link and a bot answered. Now you're staring at two headlines from the same month: 42% of German candidates walked out of AI interviews, and applicants who finished an AI interview were 12% more likely to get an offer. Both are real. Both are 2026. Here is how to decide which one applies to you before you hang up.

The two datasets aren't contradicting each other

They measure different populations at different points in the funnel. The 42% walk-out number counts self-selected candidates bailing mid-process, often because nobody told them a bot would score them. The 12% offer lift is conditional on finishing a specific kind of structured voice interview where a human still made the final call.

Here's the underlying data side by side:

The Glasshouse survey polled roughly 3,000 job seekers across the US, UK, Ireland, Germany, and Australia. The Chicago Booth / PSG Global Solutions randomized trial covered 70,000+ real customer-service applicants across three arms: human recruiter, AI voice agent, or candidate choice. Those are not the same experiment. Reading them as one is the mistake that gets people to walk out of a screen they should have finished.

Should I do an AI interview? The one-sentence answer

Yes, if it is a structured conversational voice screen with disclosure, and no, if it is a one-way video with silent scoring you weren't warned about. That distinction predicts almost everything downstream, from your offer odds to whether you'll ever hear back.

The Chicago Booth working paper by Brian Jabarian and Luca Henkel ("Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews," January 2026) studied PSG Global Solutions' voice agent, "Anna." Anna covered a median of 9 topics per interview versus 5 for human recruiters. Candidates who spoke to Anna were:

• 12% more likely to receive an offer
• 18% more likely to start the job
• 17% more likely to still be there at 30 days
• Reporting roughly half the gender discrimination of human interviews

That is the good AI interview. Now the bad one: the Greenhouse 2026 report (n=2,950) found 70% of candidates were never told AI would evaluate them, 21% didn't learn until they were already in the video room, and only 26% believe AI evaluates fairly. That is the HireVue-style one-way video that generates the walk-outs.

Why Germany hit 42% and the US didn't

Germany's walk-out rate is a supply-side story, not a cultural one. There are simply not enough recruiters in the market, so employers automate harder and candidates hit bots more often with less human context around them.

In Refolk's index of professional profiles, the US has 91,918 recruiters and talent acquisition professionals. Germany has 4,197. That is roughly a 22:1 ratio for economies where the US GDP is only about six times larger. Combine that scarcity with GDPR, works-council friction, and a hiring culture that expects a named human contact, and you get exactly what Glasshouse measured: a German candidate is more likely to encounter AI screening and more likely to bail on it when they do.

The practical read for a US or UK job seeker: your walk-out instinct is being shaped by a European debate that has different structural inputs than your market. The 42% is real. It is also not your base rate.

The stay-or-walk checklist

Run the interview through these six questions before you decide. Three or more "no" answers is a defensible walk. Fewer than three, you finish the call.

1. Was AI disclosed before you clicked the link? If disclosure came inside the interview window (the 21% surprise cohort in the Greenhouse data), that is a real red flag about the employer, not just the tool.
2. Is it conversational voice, or one-way video? Voice agents like PSG's Anna generated the +12% offer lift. One-way video with silent facial and vocal scoring is a different product with no comparable RCT evidence.
3. Does a human make the final decision? In the Chicago Booth study, humans still hired. If the posting or the bot admits an algorithmic auto-reject, your calibration collapses.
4. Is the role high-volume and structured? Customer service, sales development, entry-level ops: structured AI screens fit these well. Staff engineer, design lead, exec: much weaker fit, walk-out is more reasonable.
5. Is there a named recruiter you can email after? No named human, no appeals path. That is the Glasshouse "never heard back" pattern, which covered more than half of respondents.
6. Does the vendor tell you what it scored? Total opacity plus a rejection at 3 a.m. is the worst-case pattern, and per Greenhouse it is common.

If you decide to stay, the prep changes. If you decide to walk, do it in writing to the recruiter's inbox with a one-line reason, so you preserve the option to re-enter later.

AI voice interview prep that actually has evidence behind it

Most "HireVue AI interview tips" articles cite nothing. The Jabarian and Henkel paper is the rare exception with real behavioral findings you can drill on. Two effects showed up cleanly in the transcripts of 70,000+ interviews:

• Vocabulary richness raised offer probability. Candidates who used a wider range of specific nouns and verbs did better. Generic answers ("I'm a hard worker, I'm a team player") underperformed concrete ones ("I handled escalations on Zendesk, routing billing disputes to tier 2").
• Back-channel fillers lowered it. "Uh," "mm-hmm," "uh-huh," and similar utterances measurably reduced offer odds. Human interviewers tune those out. Voice AI transcribes them and, apparently, weights them.

Two more prep moves that follow from the study design:

• Cover more topics than the bot asks. Anna hit a median of 9 topics per interview versus 5 for human recruiters. If your answers are tight, you can get through more of them, and the model has more signal to score. Long ramble answers waste your topic budget.
• Answer in structured chunks. Situation, action, number, result. The model is looking for extractable claims, not vibes.

The other high-leverage prep step is upstream of the interview: making sure the resume that got you into the AI screen actually matches the job the bot is going to ask you about. If your resume oversold a skill, Anna will find the gap in about 90 seconds. Pasting the job posting into Refolk and letting it rewrite your resume against that specific posting is the cheapest way to make sure the interview you show up to matches the one the bot was briefed on.

AI recruiter screening: what changes if you finish it

Finishing the screen keeps you inside the population where the +12% offer lift is measured. Walking removes you from it. That is the single most important line in this article.

There is also a fairness argument that is uncomfortable but real. Jabarian's paper found reported gender discrimination roughly half of what human interviewers produced. For candidates from underrepresented groups, the human interviewer is not obviously the safer bet. That does not mean AI screening is unbiased. It means the comparison group is not clean either.

And the Philadelphia Inquirer's July 27, 2026 feature raised a point most candidate-experience surveys miss: high walk-off rates may be a feature for employers drowning in AI-generated applications. If the employer is using the screen partly as a fraud filter, walking is exactly the signal that filter is looking for. A serious candidate who bails ends up in the same bucket as the bot-written applicant who couldn't hold a conversation.

When walking out is the right call

Walk when disclosure failed, when the format is one-way video with no human decision-maker, or when the role is senior enough that a 15-minute voice screen isn't going to represent you. Those three cases are where the +12% offer-lift evidence stops applying.

Concretely, walk if:

• You were dropped into a video recording with a countdown timer and no prior notice
• The posting is for a staff-plus IC role or a people-manager job with real scope
• There is no named recruiter, no company email domain in the invite, and no appeal path
• The vendor is unnamed or the URL looks like a redirect chain
• You're being asked to grant camera, microphone, and screen-recording permissions for a first-round screen

If you walk, send a two-sentence email to whatever recruiter contact you can find: "I'm not comfortable completing an unmoderated AI video interview for a first-round screen. Happy to do a 15-minute call with a recruiter or hiring manager." That keeps the door open, and career coaches like Amanda Augustine at Careerminds have made the same point in the Euronews coverage: the format is negotiable more often than candidates assume.

FAQ

Should I do an AI interview if the company won't tell me which vendor they use?

Push back once, politely. Ask the recruiter which platform, whether it's voice or video, and whether a human reviews the result. If they answer, you have the information you need to run the stay-or-walk checklist. If they refuse or don't know, that itself is the answer, and it's the pattern behind the Greenhouse finding that only 26% of candidates believe AI evaluates fairly. Employers that can't describe their own screening tool rarely have a strong hiring process behind it.

Are HireVue AI interview tips different from voice AI interview tips?

Yes, meaningfully. HireVue's one-way video format rewards eye contact with the camera, tight 90-second answers, and rehearsed structure, because there's no interviewer to redirect you. Conversational voice AI like PSG's Anna rewards the opposite: interactive back-and-forth, wider vocabulary, more topics covered, fewer filler words. Prepping for the wrong format is one of the more common reasons strong candidates underperform on the screen.

Does walking out of an AI interview get me blacklisted?

Usually not from the company, but often from the specific requisition. Most ATS platforms flag "candidate withdrew" as a neutral status, and you can re-apply to future roles. What you do lose is the current pipeline slot, and in tight markets that slot doesn't reopen. If you're going to walk, do it in writing with a specific, professional reason so the recruiter has something to file.

What if my resume got me the interview but doesn't match what the bot will ask?

That's the highest-risk version of an AI screen, because voice AI is very good at surfacing mismatches between claimed and actual experience. Rewrite the resume against the specific posting before the call so the story you tell the bot matches the story that got you in the door. Pasting the posting into Refolk and getting a tailored resume and cover letter back is the fastest way to close that gap.