HBR: AI Quietly Broke Hiring. Now 6,380 Screening Sessions Show Companies Are Selecting for Who’s Best at Interviewing, Not Best at the Job

Harvard Business Review just put a number on something a lot of hiring managers have only whispered about. After interviewing 120 talent-acquisition leaders and analyzing 6,380 recorded first-round screening sessions, the authors concluded that companies now risk selecting for candidates who are best at navigating the hiring process, not best equipped to do the job (Harvard Business Review, June 2026).

Here’s our read. For decades, resumes and interviews were never the point. They were proxies, cheap stand-ins for the thing employers actually wanted to measure: can you do the work. Generative AI just made those proxies free and infinitely fakeable, so the whole selection machine is now optimizing for the wrong variable. If you want to understand why companies are slowing hiring and tightening their screens at the same time, this is the engine underneath it.

What the 6,380 sessions actually revealed

The HBR team didn’t just survey opinions. They spent five months talking to 120 talent-acquisition leaders across 87 companies, from staffing firms to Series A startups to enterprise tech, then layered in analysis of 6,380 recorded screening calls.

The pattern they kept hitting: the screen rewards the performance of competence more than competence itself. When the test is a polished resume, a smooth recorded answer, or a take-home you can run through a model, you’re measuring presentation, not capability.

• The sample was broad. 87 unique companies spanning staffing, startups, and enterprise tech, not one narrow slice of the market.
• The method mixed both layers. Qualitative interviews plus thousands of real recorded sessions, not just self-reported vibes.
• One caveat worth naming. The authors are technologists building a startup in this space, not independent academics, so the framing leans toward the problem their product solves.

Interview Guys Take: The most uncomfortable line in the whole study isn’t about fraud. It’s that the system can be working exactly as designed and still pick the wrong person, because the thing it measures stopped predicting the thing that matters. That’s not a glitch you patch. That’s a proxy that died.

The cheating curve went vertical

If you think of AI gaming as a few bold candidates with a chatbot open in another tab, the assessment data says you’re way behind.

AI-assisted cheating on technical hiring assessments more than doubled in 2025. Proctored fraud rates jumped from 16% to 35%, and according to the same report, 88% of online exams now face active AI cheating risk (Aiseptor / Talview AI Threat Index).

• Fraud rate on proctored tests: up from 16% to 35% in a single year.
• Online exams at active risk: 88%, which means the remote take-home is barely a filter anymore.
• Looking ahead: Gartner projects that by 2028, one in four candidate profiles worldwide will be partially or entirely fake.

Why your resume stopped being a signal

When everyone has the same writing tool, everyone’s paper looks the same. SHRM estimates that between 40% and 80% of applicants now use AI to write resumes, craft cover letters, and prep for interviews, producing a flood of near-identical applications.

That flood breaks the AI matching on the other side too. The screening tools and the applicants are reading from the same playbook, so the keyword game cancels out. If you’ve ever wondered how many companies are using AI to review resumes, the answer is enough that both sides of the table are now running models at each other.

We’ve tested this dynamic ourselves in the ChatGPT vs Gemini cover letter showdown, and the honest takeaway is that the output is good enough to be indistinguishable, which is exactly the problem. When a signal is free, it stops being a signal.

Interview Guys Take: Everyone keeps asking which AI writes the best application. Wrong question. The moment a tool is available to everyone for free, the advantage it gave you evaporates. You’re not standing out, you’re blending into a pile of documents that all sound like the same competent stranger.

Hiring managers know, and they’re rattled

This isn’t a quiet suspicion anymore. Checkr surveyed 3,000 American managers and found that 59% have suspected a candidate of using AI to misrepresent themselves, and 62% agree that job seekers are now better at faking their identities with AI than hiring teams are at detecting them (Checkr, 2025).

On the platform side it gets starker. Between September and November 2025, Huntress flagged 23.2% of job applicants as a fraud risk based on signals like email, phone, LinkedIn, and identity traces.

• 59% of managers have suspected AI misrepresentation in a candidate.
• 62% admit candidates are now outpacing their detection.
• 23.2% of applicants got flagged as a fraud risk in one company’s three-month window.

The expensive part nobody budgets for

When the screen selects the wrong person, the cost shows up later and it’s not small. SHRM’s 2025 Benchmarking Report puts the average cost-per-hire at $5,475 for non-executive roles and nearly $36,000 for executive roles.

Now multiply that by every hire who interviewed beautifully and couldn’t do the job. The math is why employers are about to change the rules, and why the change is going to be sharper than the slow drift you might expect.

The pivot is already underway: work samples and live problem-solving

Here’s the part the alarm bells miss. Interviews aren’t the villain. The Sackett meta-analysis discussed by SIOP found structured interviews still carry the highest mean operational validity among selection tools when applied properly.

The crisis isn’t the interview. It’s the remote, asynchronous, low-accountability formats that AI gaming exploits. So employers are pivoting to the formats AI can’t fake for you: live problem-solving, real work samples, and watching you think in real time. That’s also why video interviews keep climbing, the camera is the accountability layer.

This is the same logic behind the broader move toward skills-based hiring. Show the work, not the words about the work.

• Live exercises replace take-homes. If you can’t paste the prompt into a model unobserved, the test starts measuring you again.
• Work samples beat credentials. A realistic task done in front of someone is far harder to fake than a line on a resume.
• Structured beats casual. Same questions, same rubric, scored consistently, which is where the validity actually lives.

What this means if you’re the one applying

The contrarian takeaway is good news if you can actually do the job. As the proxies collapse, raw demonstrated ability is becoming the scarce, valuable signal again.

Use AI to get through the door, sure. But understand that it only gets you to the room where you have to perform live. The candidates who win the next two years aren’t the best prompters. They’re the ones who can do the thing on demand, with someone watching.

That’s why building real, demonstrable capability matters more than ever, whether that’s a credential you can defend like the Google Business Intelligence certificate or the practiced ability to walk through your actual work under pressure.

Interview Guys Take: The cruel irony for anyone who leaned hard on AI to fake their way in: you’re about to face a hiring process specifically redesigned to catch exactly that. The skills gap between your application and your ability used to stay hidden until month two. Live formats move that reveal to minute ten.

Strip away the fraud panic and the deepfake forecasts and you’re left with a simple shift. The cheap signals are dead, and the expensive ones, real work done in real time, are coming back into fashion because they’re the only ones left that mean anything.

Build the thing you’d otherwise be tempted to fake. Practice doing it out loud, on demand, the way you’d practice for a tough business analyst interview. When the whole system pivots to watching you actually perform, the people who can perform stop competing with the people who can only prompt.