Restoring Hiring Signal Without Surveilling Candidates
AI can polish a resume, draft an application answer, solve an assessment and help someone rehearse an interview.
Hiring teams are right to ask whether the person in front of them can actually do the work. They are wrong to assume the only answer is to watch candidates more closely.
Lockdown browsers, copy-and-paste blocking, tab tracking, typing analysis and live proctoring promise to reveal who is “really” completing an assessment. They also turn a hiring process into an investigation before the employer has established much trust of its own.
The better response is to improve the signal being requested, then verify important claims in proportion to the decision.
The concern is real
An ERE article on generative AI and selection describes signal loss across resumes, assessments and asynchronous interviews. Its proposed safeguards range from honesty agreements and redesigned assessments to passive trace monitoring, lockdown browsers and live proctoring. It also warns that mouse movement, typing speed, tab switching and similar behavioural traces may disadvantage neurodivergent candidates or people with different levels of technical familiarity.[1]
That is a useful diagnosis with an important caution inside it.
Fluent writing is no longer reliable proof that someone writes fluently without assistance. An unsupervised answer may reveal less about unaided performance than it once did. A scripted interview response can sound convincing.
None of this proves that every candidate using AI is dishonest. Writing support may help a capable person communicate in a second language, work around disability, or remove irrelevant presentation barriers. In some roles, sensible use of AI is part of the capability being hired.
The hiring team first needs to decide what it genuinely needs to know.
Verification and surveillance are different
Verification tests a material claim or requirement. It asks for relevant evidence, explains the purpose, uses the least intrusive method that can resolve the uncertainty, and lets the candidate respond.
Surveillance collects behaviour around the assessment in case it reveals misconduct. It may record a screen, room, face, voice, keystrokes, browsing behaviour or device signals that are several steps removed from the work.
A video feed can show that a candidate stayed in frame. It cannot establish that they will make a sound decision during a difficult customer incident. Tab-switch data can show that another window gained focus. It cannot explain whether assistive technology, a connection problem, a notification or prohibited help caused it.
More observation does not automatically create better evidence.
It does create more personal information to collect, explain, secure, retain and interpret. Australian Privacy Principle guidance says personal information collection should be reasonably necessary, proportionate and limited to the minimum needed. It specifically notes that collecting information from every job applicant may be excessive when it is only needed for shortlisted candidates.[2] UK Information Commissioner's Office guidance applies the same discipline: personal data should be adequate, relevant and limited to what is necessary for the stated purpose.[3]
That is not a blanket ban on monitored assessment. A high-risk role may justify stronger controls for a specific, late-stage test. It is a reason to make surveillance the exceptional choice, not the default setting for everyone who applies.
Build a better evidence chain
Signal improves when each selection step has one clear job.
Start with the real work
Define the responsibilities, decisions, skills and practical conditions that need to be assessed. Separate day-one requirements from capabilities a strong candidate can learn.
If the role is vague, no monitoring tool can repair the assessment built on top of it.
Ask candidates for inspectable context
A skills label or polished claim is only a starting point. Ask where the capability was used, what the person contributed, the constraints they faced and what changed because of their work.
Useful support might include a portfolio item, public project, qualification, licence, work artefact that is safe to share, or a short explanation of a result. Evidence should remain candidate-controlled: never pressure someone to disclose confidential employer material merely to make a claim look stronger.
Corroborate in stages
Do not demand maximum proof from every applicant. Increase verification as the decision becomes more consequential:
- use the application to collect concise, role-relevant context;
- use structured interview questions and dynamic follow-ups to understand personal contribution and judgment;
- use a proportionate work sample for a capability that needs demonstration;
- confirm essential credentials, identity, work rights or references at the appropriate stage and with clear notice.
The work sample should match the real working conditions. If employees will normally use documentation, search, calculators or AI, banning every tool may test memory or rule compliance instead of job performance. If unaided ability is genuinely required, explain the restriction before the assessment and provide an adjustment path.
Make the AI rules explicit
“Do not cheat” is not an adequate AI policy.
Tell candidates what assistance is allowed at each stage. Editing for clarity may be acceptable in an application while live generation may be prohibited in a particular assessment. Another role may require candidates to use AI and explain how they checked the result.
Transparent rules give honest candidates a fair chance to comply. Hidden traps mainly test whether people guessed the employer's expectations.
The US Equal Employment Opportunity Commission says selection procedures should be job-related, should measure the skill or aptitude they claim to test rather than an applicant's impairment, and should provide reasonable accommodations where required.[4] CIPD guidance similarly recommends clear, objective, structured and transparent selection based on the candidate's ability to perform the role.[5]
How RoleSage supports evidence without policing candidates
RoleSage is designed to make relevant context inspectable before a hiring team reaches for invasive controls.
Candidates build an editable profile that connects skills to activities, personal contribution, outcomes and optional evidence links. They decide what to include and share. A public link, work example or detailed activity can support a claim, but RoleSage does not pretend that every candidate-provided statement has been independently authenticated.
Hirers can define structured role requirements and review matched skills, related skills, activities, application answers, alignment and gaps together. Match bands organise attention while the supporting view shows why a signal is direct, related, incomplete or still unclear. AI helps organise and explain the material; it does not decide who should be hired.
RoleSage also keeps trust signals distinct from role fit. LinkedIn verification confirms the account-to-profile linkage, not the truth of every career claim. Optional independent professional relationship support is another visible signal, not a ranking boost. Neither replaces a structured interview, a suitable work sample, a credential check or a reference where one is justified.
This creates a more honest chain:
Role requirement → candidate-controlled context → visible support or uncertainty → proportionate human verification
The aim is not to prove that a candidate never used AI. It is to gather enough job-relevant information to understand what they can do, what remains uncertain and which next step can resolve it fairly.
Set a higher standard for the hiring process
Before adding monitoring to an assessment, ask:
- What specific risk are we trying to control?
- Is the information we would collect genuinely necessary?
- Could a better question, work sample or follow-up produce stronger evidence?
- Would the method disadvantage candidates using assistive technology or unfamiliar devices?
- Have we explained the rules, data collection, retention and adjustment process?
- Are we applying the control only at the stage where it is justified?
- Will a person review the result and give the candidate a way to challenge an error?
Candidates should be accountable for truthful applications. Hiring teams should be accountable for fair, proportionate selection methods.
Trust will not be restored by treating every applicant as a suspect. It will be restored by asking for better evidence, explaining how it will be used, and verifying the few claims that carry the decision.
Request relevant evidence. Verify proportionately. Keep people in control.
References and further reading
- ERE: Why the Entire Selection Process Is Losing Its Signal (and How to Fix It) - US recruiting-practitioner analysis arguing for layered safeguards, including monitoring, alongside its warning about bias risks from behavioural trace data.
- Office of the Australian Information Commissioner: APP 3 - Collection of solicited personal information - Australian guidance on necessity, proportionality, data minimisation, fair collection and job-applicant information.
- UK Information Commissioner's Office: Data minimisation - UK guidance on collecting and retaining only personal data that is adequate, relevant and necessary for a stated purpose.
- US Equal Employment Opportunity Commission: Employment tests and selection procedures - US guidance on job-related assessment, disparate impact, accurate measurement and reasonable accommodation.
- CIPD: Selection methods - UK professional guidance on clear, structured and transparent interviews, work samples, tests and other job-relevant assessment methods.
- RoleSage: The Application Avalanche Is a Signal Problem, Not a Volume Problem - why stronger candidate-controlled context is more useful than processing weak proxies faster.
- RoleSage: Why Candidate Fit Should Be Explained, Not Just Scored - how evidence, gaps, limits and next questions should remain visible behind a fit signal.