AI Checks Its Own Homework Before the Interview Even Starts

Steve Swan gets the same phone call often. A hiring manager comes back from an interview and says, “these people can’t talk about what’s on their resumes.” Swan, CEO of executive recruiting firm The Swan Group, told BioSpace that when he traces the problem backward, it never starts in the interview room. It starts three steps earlier: artificial intelligence writes the job description, another AI rewrites the candidate’s resume to match it, and a third AI screens that resume to decide who gets an interview. “So, AI’s kind of checking its own homework, right?” Swan said.

That line names something most talent acquisition teams have not priced in. Three AI tools touch a candidate before a human ever does: one writes the job posting, one reshapes the resume to fit it, one screens the resume against that same posting. We think that sequence amounts to a self-referential loop, not a screening process, since nothing outside it tests whether the resume actually describes what the candidate can do. The interview is what is left, and most teams still run it like a formality, a courtesy before an offer, rather than the one point in the process where a claim gets tested against a real person.

The Loop With No Outside Check

Swan also noted that AI may interview candidates too, compounding the problem. If that step is also pattern matching against the resume rather than probing it, the process closes completely before a human ever gets involved. Three tools end up grading each other’s output, and nothing in the chain checks it against what the candidate can actually do.

The Symptom Shows Up Too Late

The hiring manager’s complaint, that a candidate cannot talk about what is on their own resume, sounds like a hiring manager problem. It is not. Swan called it what it actually is: the identification process was “completely muddied” long before that interview happened, and the company does not even know if the candidate has the right skill set. The pool the manager is choosing from was compromised at the source. Even when a hire does get made, Swan said, the company may have found “a” person rather than “the” person.

In our work with hiring teams, the moment a candidate cannot defend their own resume rarely marks where the problem began. It marks where the problem finally became visible, several stages after it was created. If you treat interviews as a rubber stamp rather than a test, ask yourself this. How many of your last ten conversations ended with a hiring manager quietly wondering whether the resume in front of them matched the person sitting across the table?

The Networking Workaround Only Helps the Well-Connected

Swan’s own recommended workaround shows how seriously he takes this. His advice to candidates: skip the identification process entirely. “Find somebody that knows somebody that knows somebody that knows somebody inside that company,” he said, and have that contact vouch for you directly: “He’s a good worker. He shows up on time. He’s a real person.” That is not a hiring process. It is a shortcut built for candidates who happen to have the right network, and it leaves everyone else running the process Swan just described.

A workaround that depends on who you know cannot scale to a full requisition, and it does nothing to fix the process for candidates who apply cold. If networking is the only reliable route around a hiring pipeline that grades its own work, the system itself needs a check, not candidates with better contacts.

Salary Opacity Runs the Same Play

The AI process is not the only place trust breaks down before the offer stage. Laura Helmick, founder and managing partner of life sciences recruiting firm LHB Clinical, told BioSpace that small biotechs sometimes withhold information about company stability, hoping candidates will not notice, only for the candidate to leave within a year once the gap becomes obvious.

Swan described a parallel pattern on pay. A company might post a salary range of $175,000 to $300,000 while quietly capping the role tens of thousands of dollars lower. If a candidate wants more because the posted range goes higher, Swan will not present them. “I don’t want to waste my time, their time or the candidate’s time,” he said.

Different mechanism, same failure. Helmick and Swan are both describing companies that let a distance between what is claimed and what is real survive. That distance becomes someone’s problem later, whether it surfaces in a resume, a stability story, or a salary band. Whoever owns the interview owns the job of narrowing that distance before the offer goes out, not after.

The Interview Was Never Supposed to Carry This Alone

None of this argues for slower hiring or more interview rounds. It argues for treating the interview differently, as the one stage where a claim about a skill, a project, or a piece of ownership either survives contact with a specific question or it does not. Swan’s phone calls from frustrated hiring managers are evidence this is already happening, just too late and too passively, as a discovery rather than a design.

The pipeline that produced the candidate breaks down at up to three separate points before anyone sits down across a table. The interview does not have to be the fourth.

Original reporting: BioSpace.