Hiring People Still Requires Talking to People

Companies should use AI aggressively in recruiting where it helps recruiters do better work. They should not use it as an excuse to stop talking to people. That is the line, and too many HR departments and executive teams are crossing it while pretending the decision was forced on them by technology.

What set me off was being asked to complete another AI interview after already having poor experiences with them. Not because I am anti-AI. I use AI. I build with technology. I understand why a recruiter would want help reviewing resumes, summarizing candidate information, drafting questions, organizing screening data, and identifying areas worth exploring. That is useful work. But sitting a candidate in front of a bot or a one-way camera before any real human conversation is something else entirely.

The problems are not theoretical. These systems can treat a normal pause as the end of an answer. They can stop recording before the answer is finished. They can give too little time to respond. They can struggle when the candidate asks for a question to be clarified or rephrased. In a real interview, that is ordinary conversation. With a bad automated system, it becomes a penalty box.

There is a technical reason for part of this. Voice AI has to decide whether someone is done speaking or simply thinking. That is called end-turn detection. Current research still describes distinguishing hesitation from turn completion as a difficult problem in spoken chatbot systems. When the system gets it wrong, it interrupts, cuts off answers, or sits there awkwardly while the candidate tries to figure out whether the machine is listening. That may be annoying in a customer service chatbot. In hiring, it can affect whether a person gets a job.

Speech recognition raises an even bigger fairness problem. Stanford FairSpeech research found average word error rates of 35% for Black speakers compared with 19% for white speakers across five major commercial speech-recognition systems. Those systems were tested in 2019, so I am not claiming those exact numbers describe every modern system in 2026. But even 19% is a hell of an error rate if speech technology is being used anywhere near an employment decision. And more recent research still finds performance differences involving dialect and accent, including Standard American English, African American Vernacular English, Chicano English, Spanglish, and regional accents.

Race is not the only concern. Accent, dialect, pronunciation, cadence, prosody, disability, speaking speed, and other communication differences can affect automated systems. The EEOC has specifically warned that AI used in employment is still subject to discrimination law, including the example of video-interview software scoring someone poorly because a disability affects their speech patterns. That should make every executive who approved these systems sit up straight. If your process disadvantages people because of how they speak, pause, process language, or communicate under pressure, calling it modernization does not make the problem disappear.

Then there is the simple question of whether candidates will tolerate it. A 2026 Monash field experiment involving 3,296 real job applicants found that asynchronous interviews reduced application continuation by around 53%, including among highly qualified applicants. Applicants also perceived asynchronous interviewing as less fair and more competitive than live interviewing. In plain English, a lot of people looked at the process and walked away. Some of them were good candidates.

That matters because hiring is not just a company extracting information from an applicant. It is a two-way evaluation. The interviewer is deciding whether the candidate fits the job. The candidate is deciding whether the manager, team, company, and opportunity are worth joining. Candidates need to ask follow-up questions. They need to hear how the organization describes the role. They need to understand how the work actually functions. They need some sense of the people they may be working with. A recorded prompt cannot give them that.

Research on real-time AI interviewers points to the same issue from another angle. A 2026 study documented information loss, premature termination, latency, interruption, and weak follow-up questioning. Participants' trust was affected not only by whether the bot worked well, but by what the organization's decision to delegate the interview to AI communicated about the organization itself. That is the part executives should stop dodging. The process sends a message.

And the message is not always flattering. It tells candidates, before they have met a single person, that the company may value recruiter time more than their time. It tells them the organization wants judgment, personality, communication, interest, and fit, but does not want to spend 15 or 20 minutes having a human conversation to find it. It tells them the company is comfortable demanding vulnerability from applicants while offering none in return.

This is where the corporate language gets slippery. Automation. Modernization. Scaling. Efficiency. Those words can describe real improvements. They can also describe cost cutting with better branding. Technology did not force companies to stop talking to candidates. Leadership made that decision. If HR designs the process and the C-suite approves it because it reduces recruiter time, scheduling effort, or cost per applicant, then they own the candidate experience and the risks that come with it.

There is a better way to use the technology. Let AI help recruiters prepare. Let it summarize resumes, flag inconsistencies, organize screening notes, generate structured questions, compare requirements, reduce administrative drag, and help identify areas that deserve a closer look. Some properly designed AI systems may even reduce certain forms of human bias. Good. Use the tools. But AI assisting humans is not the same thing as AI replacing human contact.

Saving 15 or 20 minutes may look good on a spreadsheet. But if it drives away qualified candidates, creates fairness concerns, weakens trust, and removes the candidate's ability to evaluate the company, that is a bad trade. Hiring is about people. Real people are making decisions about careers, income, families, and futures. They deserve more than a link, a camera, and an algorithm. At some point, if a company actually wants to hire people, somebody from the company should talk to them.