Employers Are Quietly Calling People You Didn’t List as References. AI Made Them Stop Trusting the Ones You Did.
A reference used to be a simple ritual. The candidate handed you three names, you called them, they said kind things, you moved on. In 2026 that ritual quietly broke. As AI-written résumés, AI interview coaching, and even AI-drafted reference letters made application materials harder to trust, employers went hunting for candor somewhere the candidate couldn’t stage-manage. Reporting through July 2026, in Bloomberg, Becker’s, and elsewhere, described the return of the “backdoor reference”: the hiring manager who skips your list and calls a former colleague you never named, usually someone one LinkedIn hop away. Once reserved for executive hires, it is now reaching ordinary roles. For a small team, it feels like diligence. It is mostly the wrong fix.
The instinct behind the backdoor call is easy to understand. If a model can write a flawless résumé, coach a candidate through the interview, and produce a glowing reference on request, then every polished input starts to look suspect, and a manager reaches for the one source that feels un-fakeable: a quiet word with someone who actually worked with the person. The problem is not the impulse. It is that the backdoor call answers a real question with a method that is biased, invasive, and, for a small team, more expensive than it looks.
Why the Backdoor Came Back
The reference a candidate hands you was never strong evidence. The candidate picks the names, chooses people who like them, and often briefs them on what to expect. You are, in effect, interviewing witnesses the defense selected and prepped. HR practitioners have said for years that the on-list check has become an obligatory step with little real signal in it. It survived anyway, because it was cheap and because the rest of the file still carried some weight.
That last bit is what changed. In a Gartner survey of 3,290 job seekers, 39% said they used generative AI during the application process. Polish is now free, tailoring is free, and a share of candidates have started submitting AI-assisted or outright fabricated references. So the front door stopped reassuring anyone. The backdoor reference is the reaction: if the paper the candidate controls can no longer be trusted, go around the candidate entirely. What used to be an executive-search habit has spread down to ordinary roles, precisely because the trust problem is now everywhere, not just at the top.
The Secret Call Has Three Problems
The first is that it is hearsay wearing the costume of data. An off-list contact was not chosen for relevance to your role. You get one person’s mood about a candidate from three years and two jobs ago, unweighted, unaccountable, and impossible for the candidate to answer. A single sour memory can sink someone, and you will never know whether it was fair.
The second is quieter and worse. The backdoor check punishes the candidate who does not share your network. Take two applicants of equal ability. The one who happens to have a mutual connection at your company gets vouched for or torched; the outsider, often the career-changer, the person from a different industry, the one without the right alumni network, gets judged on nothing or skipped. You have quietly turned “who do we both know” into a selection criterion, which is a good way to keep hiring people who look like the people you already have.
The third is consent, and it carries real cost. Contacting people about a candidate without the candidate’s knowledge means processing personal data with no clear basis, which sits badly under GDPR and against the direction European hiring rules are moving. Even where it is legal, candidates often find out, and what they learn is that you investigated them behind their back before you were willing to trust them to their face. For a ten-person team whose entire edge is being the human, above-board option in a market full of bots, that is an own goal.
You Can’t Verify Your Way Out of a Broken Signal
Step back and the backdoor call reveals what it really is: an attempt to answer “can this person do the work” with secondhand paper, just sourced more secretly. A former colleague’s general impression is not evidence that a candidate can do your job, on your team, with your constraints. You are adding investigation, not adding signal.
Verification and evaluation are different problems, and the industry keeps confusing them. Verification asks whether the past is real. Evaluation asks whether the person can do the thing in front of you now. Backdoor references chase the first and quietly hope it stands in for the second. It does not. The only reliable way to find out whether someone can do the work is to watch them do something close to it, which is exactly the place where AI polish and network gossip both fall away.
What a Small Team Should Do Instead
Move the signal into the room, with consent. A short, structured exercise or conversation built around the actual demands of the role tells you more in twenty minutes than any off-list call, because it asks the candidate to do the work now rather than describe it later, and because they know it is happening. That last part matters. Signal you gather openly is signal you can act on without wondering whether you crossed a line to get it.
If you still want references, run them in the open and make them earn their place. Ask for people in specific relationships: a former manager, a peer, someone the candidate managed. Ask behavioral questions tied to your role instead of “what were they like.” Tell the candidate you may ask for an extra name. Consent plus structure beats secrecy plus gossip every time, and it does not cost you the trust of the person you are hoping to hire.
The point is not to reject AI or to romanticize the old reference call. It is to aim automation at the work it is good at, the sourcing, sorting, scheduling, and scoring against what the role truly requires, so a human on your team reaches the real evaluation faster instead of reaching for the phone behind the candidate’s back. That division of labor, the machine on the logistics and a person on the judgment, out in the open, is what we built Kynto around. In a market where everyone else is quietly snooping to rebuild trust they lost, being the team that earns it in the open is not just cleaner. It is faster, and it is the reputation a small company cannot afford to spend.
Key Takeaways
- Backdoor references came back in 2026 because AI made application materials hard to trust. With 39% of candidates now applying with AI, the paper they control stopped reassuring anyone, so employers started calling people the candidate never listed, often one LinkedIn hop away.
- The secret call carries three costs a small team can’t absorb: it is unaccountable hearsay, it punishes candidates who don’t share your network, and contacting people about someone without their knowledge is a consent and legal problem, sharpest in Europe under GDPR.
- Don’t verify harder behind the candidate’s back. Move the signal into a consented, structured evaluation early, and run any reference check in the open with role-specific questions. Watch the person do the work, and keep the final judgment with a human.
The teams that come through this well will not be the ones with the best contacts in the industry or the sharpest instinct for who to call on the side. They will be the ones who noticed that the reference ritual quietly stopped meaning anything, who refused to answer a trust problem by trading away their own trustworthiness, and who spent their scarce human attention where it still changes the outcome: in front of a real person, early, watching them do the actual work. You can learn a lot from a quiet phone call. You just can’t learn the one thing you needed to know.
When the paper can’t be trusted, the answer isn’t to snoop harder behind the candidate’s back. Kynto handles the sourcing, scoring, and scheduling so a human on your team reaches a real, in-the-open evaluation faster, and still makes the call.
See how Kynto keeps a real evaluation at the center