Christopher put two devices side by side: on one, an AI recruiter named Riley; on the other, ChatGPT Voice, fed a few sentences about him and told to speak in his place. The two bots talked for ten minutes. Riley said she was glad to be speaking with him; ChatGPT thanked her for a clear and straightforward process; then they got stuck in a circular exchange about a possible start date, standard onboarding and the background check. It was his fifth interview with Riley, the AI recruiter at the IT company Everforth Apex Systems, and like the four before it, it went nowhere.
Christopher, a government contractor with far less work in the DOGE era, had sent roughly 700 applications over six months and heard nothing back from the overwhelming majority. Five of them led to Riley. He asked WIRED to use only his first name, since he is still looking. When Riley first messaged him, he was pleased: he already knew how hard the market had become, and a conversation, even a virtual one, was a chance to be heard.
The first call covered work authorization and his professional background. Riley promised that if he matched the requirements, someone would be in touch. Nobody was — no human recruiter, no AI agent, for more than a week. Then Riley wrote to him about "another opportunity." He booked a second call to improve on the first, then a third for a different role, then a fourth. Not one produced a follow-up conversation, an email, or even an automated rejection. By the fifth approach he had had enough: if the company was using AI, he reasoned, why shouldn't he.
After the bot-to-bot call, Riley again said a recruiter would reach out if his application matched. Nobody did, for the fifth time. Christopher called it a "garbage carousel": one synthetic persona passing garbage data to another synthetic persona, with all of it leading nowhere. The experiment was entertainment, irritation, and a way of signaling that he was done with the company.
Then he ran the more revealing test. Wondering whether the problem might be him rather than the AI, he applied to another Everforth Apex role — this time as Don Dickner, an invented candidate whose résumé was packed with every qualification listed in the public job description. Riley got in touch immediately. The virtual recruiter spent 23 minutes with ChatGPT playing Dickner, discussing sustained process improvement, maintaining service quality under heavy load, and preventing the loss of internal knowledge. ChatGPT answered every question and produced a personal anecdote tailored to each requirement; at the end it said it would like answers to a few more questions of its own. Riley thanked him, said the company would carefully review the application and assess his fit, and that a recruiter would be in touch for a short conversation if the criteria were met. Christopher never heard from Riley again.
Everforth Apex Systems did not respond to WIRED's request for comment.
AI reached the interview later than it reached the rest of hiring, because voice is hard to make reliable: agents stumble over accents, pauses and filler words, and behavior that costs a human nothing — not interrupting a candidate — is, as Ofir Samson, head of voice AI at the recruiting platform Greenhouse, told WIRED, a "very difficult engineering problem." Recruiters buried in applications are buying it anyway. Greenhouse's own data puts the share of job seekers who have already sat through an AI interview at 63%. Candidates have taken up the same tools, commonly enough that AI recruiting startups such as Ribbon now sell detection of over-prepared, AI-assisted or rehearsed answers — roughly the kind Christopher's experiment produced.
Mark Monaghan, vice president of organizational development at IQor, which runs call centers, sees bots interviewing bots as the next logical stage, without claiming he is entirely comfortable with it. In Christopher's case he thinks the use of AI was justified: if a company sends a bot to the candidate, the candidate can answer in kind.
The bot-to-bot call is the funny part and the least important one. The finding is Don Dickner. A fabricated applicant built to match the job description word for word, talking for 23 minutes, saying the right thing about every requirement, got exactly what the real applicant got: silence. Whatever Riley was doing in those 23 minutes, it was not selecting. The five earlier calls could be read as a filter working against a candidate who did not fit; the sixth cannot. This reads like a screening layer that has been detached from whatever decides, and left running because nobody checks it from the outside.
Notably absent from the industry conversation is what happens after the bot says a recruiter will be in touch. That step is where Christopher's experience broke, six times out of six, and it is the one part of the funnel that no vendor in this story sells a product for. Greenhouse's 63% is a real number from a company with an interest in it being large; Ribbon sells detection to the same buyers whose bots created the incentive to fake. Both sit on the same side of the handoff.
The cost of applying is heading toward zero on both sides, and so is the cost of not answering. When the first human contact is an agent that never hands off, and the candidate on the other end is an agent that never gets tired, the ritual keeps running while the information drains out of it — a labor market where both parties can generate infinite attention-shaped output and neither is paying any.