No to the Chatbot Interview: Why More Than Half of Candidates Reject Fully Automated AI Conversations

Marcus

Oct 16

6 min read

A candidate introduces himself to an AI avatar as “Chick Bongo” and reports seven years at “Sprucewood”, Richard included. Asked about a difficult situation, he answers with pallets, denim, and the insight that every pharmacist is a hallway. He is looking for a job with “raccoon protocols”. The avatar stays friendly, summarizes everything, asks with interest, and announces the next steps at the end.

The clip comes from satirist Graham Zip, who has gathered well over 90,000 followers on Instagram as @graham.zip. Futurism and t3n picked it up in September 2026, and it can also be watched on YouTube. It is funny, no question. For candidates who have to take such conversations seriously, it rarely is, and that is exactly why the fully automated chatbot interview deserves a second look.

What the clip shows – and what it doesn't

The finding is quickly told: the avatar does not check whether an answer makes sense; it simply keeps the conversation going. t3n explains this by noting that language models are tuned for conversation and agreement and keep going even when the substantive frame is abandoned. For screening, that is a problem: a system that cannot tell nonsense from experience gives hiring managers no reliable basis.

In fairness: a satirical clip is not a test. Which product was in use and what the evaluation would have looked like in the end remains open; Futurism itself notes that it is unclear how the application proceeded. The clip proves a failure mode, not its extent. I've already covered how fair and valid such procedures are in my post AI Interviews in Recruiting: Where Efficiency Ends – and Risk Begins. Here the focus is on the candidates' side.

What candidates think of chatbot interviews

For Germany, the Königsteiner Gruppe supplies the figure this post refers to in its title. In January 2024, the research institute bilendi surveyed 1,023 employees on its behalf: only about one in five people like job interviews with chatbots or avatars, while 52 percent reject them outright. If the AI even makes the hiring decision, 61 percent view that negatively and 19 percent positively. At the same time, 47 percent expect AI to influence personnel selection more strongly in the future. Rejecting and expecting are not mutually exclusive.

The survey is almost three years old and dates from before the current avatar wave. More recent figures come from the Ipsos AI Monitor 2026: of around 1,000 respondents in Germany, 64 percent reject AI-assisted pre-selection of applicants, and only 24 percent have no problem with it. That is a different question put to a different group, namely the general population rather than candidates, but the direction is the same.

And in everyday practice? The Greenhouse report 2026 surveyed 2,950 active job seekers, including 1,200 from the US as well as participants from the UK, Ireland, Germany and Australia. 63 percent have already been interviewed by an AI, 13 percentage points more than half a year earlier. 38 percent have dropped out of an application process because it included an AI interview, and another 12 percent say they would. The most common triggers: recorded video interviews scored by an AI without a human present (33 percent), lack of disclosure (27 percent), and AI monitoring during the process (26 percent). 70 percent of those affected were not clearly informed in advance, and one in five only learned during the conversation that an AI was involved.

Two caveats. First, Greenhouse itself sells HR software, so this is a vendor survey. Second, and more notable: according to Greenhouse, only 19 percent of respondents want less AI in recruiting, and perceived bias is almost equally high for AI and for human interviewers. Candidates therefore do not reject AI as such, but a process they can neither see through nor influence.

aktien.news describes how this feels, based on people affected: systems that repeat the same question, interrupt when someone takes a breath, or demand that candidates look straight ahead even though an eye condition makes that difficult. That is considerably less funny than the clip.

The counter-check: where candidates accept AI interviews

There is evidence pointing the other way, and it deserves space because it sharpens the argument. In a field experiment, Brian Jabarian (Chicago Booth) and Luca Henkel (Erasmus Rotterdam) had around 70,000 applicants for call-center jobs in the Philippines interviewed at random by humans, by an AI voice agent, or by their own choice. AI-led interviews raised the number of job offers by 12 percent, and among those who had a choice, 78 percent picked the AI. I covered the study in September 2025 under the title "AI beats Recruiters: Candidates prefer being interviewed by a robot".

The conditions are worth a look. The conversation followed a script; human recruiters evaluated it and made the decision, supplemented by a standardized test. It involved entry-level jobs with very high application volumes, in a single industry and a single country. According to the authors, applicants rated the conversation and the recruiter in a survey similarly to the human group. That is something different from a process in which candidates do not know what is being measured and an AI without a human counterpart co-decides who moves forward. The findings therefore contradict each other less than it seems. Above all, they show that design matters, not the “AI” label.

What TA teams can take away

The consequence is not rejecting AI from the conversation, but dividing the labor: AI may prepare, structure, and summarize. The conversation in which a person is decided on stays human. Anyone evaluating a tool can check five things:

Clarify the role: Does the AI support (scheduling, questions about the process, summaries for recruiters) or does it evaluate? Evaluation without a human counterpart is exactly the setting that most often leads to drop-outs in the Greenhouse survey.
Disclose upfront: that an AI is involved, what it measures, and how the result feeds into the decision.
Offer an alternative: on request, a conversation with a human, with no disadvantage in the process. In the Rotterdam experiment, 78 percent chose the AI when they had the choice.
Run the nonsense test: Before the pilot, play through absurd answers, contradictions, long pauses, dialect, and accent. Does the system probe further or simply agree? And do people on the team actually read the transcripts?
Measure drop-outs: Drop-out rate and candidate feedback per process step, separated by conversation with a human and with an AI. What is not measured only becomes visible when the good candidates are gone.

Legally, this is no side issue: transparency and human oversight are among the requirements of the EU AI Act for recruiting AI, even if its high-risk obligations only apply later, and data protection and anti-discrimination law apply regardless of whether a human or a model decides.

Funny only as long as nobody depends on it

That an avatar takes raccoon protocols for professional experience is a good joke. That the same technology co-decides who moves forward is not. The data speak less against AI in recruiting than against processes that evaluate people without information, alternatives, or a human counterpart. Anyone automating the conversation should therefore ask two questions: can a satirist with word salad outsmart the system? And does a candidate with a dialect, thinking pauses and real experience have a fair chance in it?

Sources

Transparency Notice on the Use of AI

Tools based on artificial intelligence were used in a supporting role in the creation of this post. They serve in particular for researching and structuring information, linguistic and grammatical editing, translation, and in part for creating or editing illustrations and image material.

Conception, substantive statements, assessments and conclusions come from the author and reflect his personal thoughts, professional experience and convictions. All content is editorially reviewed and the author takes responsibility for it before publication.

Posts on this website are not created or published automatically by AI. AI is used as a supporting tool; substantive and editorial responsibility rests entirely with the author.