What You Should Tell Candidates About AI in Your Hiring Process

What You Should Tell Candidates About AI in Your Hiring Process

A candidate should not discover after rejection that software scored an interview, ranked a resume or shaped the reason they did not progress.

“We use AI in recruitment” is not much better. AI can mean anything from drafting an email to automatically filtering applicants. The label does not tell a candidate what happened to their information or whether a machine changed their path.

Disclosure is helpful. It tells people where automation enters the process, what it examines, what it produces and who remains accountable.

If a hiring team cannot explain those hand-offs plainly, the problem is larger than candidate communication. The team may not understand its own system well enough to use it responsibly.

Start with the real workflow

The ERE practitioner article behind this post makes a sound opening recommendation: tell candidates upfront when AI will be part of recruitment.[1] But disclosure cannot stop at broad reassurance that automation is efficient or unbiased.

Map every tool that can affect a candidate, including tools hidden inside an applicant tracking system or assessment platform. Record:

• the stage where it is used;
• the information it receives;
• the task it performs;
• the output a recruiter or hiring manager sees;
• whether that output can progress, delay or exclude someone;
• the person accountable for reviewing it.

Include quiet uses as well as obvious ones. Resume parsing, job matching, application-question analysis, interview transcription, assessment scoring, ranking, rejection-message drafting and AI-generated summaries do not have the same consequences. Candidates should not have to infer which meaning of “AI-assisted” applies to them.

Answer seven questions before the candidate participates

1. Where and when is AI used?

Name the relevant stages. A short notice in the job advertisement can flag the overall use. Give more detail before a candidate meets a chatbot, completes an automated assessment, records a video response or submits information that will be scored.

2. What information does it examine?

List the actual inputs: resume text, application answers, skills, qualifications, assessment responses, audio, video, public-profile information or other data. Say when a third-party provider receives that information and link to the applicable privacy and retention details.

3. What does it evaluate, and what does it ignore?

Describe the job-related criteria in ordinary language. If a tool compares stated skills with role requirements, say so. If it analyses response content but not voice, expression or appearance, say that only after confirming the product behaves that way.

Do not promise that protected characteristics, proxies or inaccessible interaction patterns are ignored merely because a vendor brochure says the system is “bias-free”.

4. What output does it create?

Does the tool extract fields, summarise evidence, suggest questions, assign a score, place people in bands, rank candidates or trigger a pass/fail result? Explain how that output is used. A summary that helps a recruiter read an application is materially different from a result that determines who reaches interview.

5. What does a person actually decide?

“Human in the loop” is too vague. Identify the role of the reviewer, what underlying evidence they can inspect, whether they can disagree with the output, and who makes the progression or rejection decision.

A person who sees only a score and routinely accepts it is not providing meaningful oversight. Human review has to be designed into the work, supported by authority, time and useful evidence.

6. How can a candidate request an adjustment or alternative?

Tell candidates how to request an accessible format or another assessment route before the tool disadvantages them. US EEOC guidance warns that algorithmic tools may screen out people with disabilities even when they could perform the job with a reasonable accommodation.[3]

The route should be easy to find, confidential and usable without requiring the candidate to diagnose the system's technical failure.

7. How can someone correct, question or challenge the result?

Provide a real contact and explain what happens next. Can the candidate correct inaccurate source information? Ask which evidence influenced the outcome? Request human review? Learn how long the relevant record will be retained?

A challenge process that leads back to the same unexplained score is not a remedy.

The legal baseline depends on the jurisdiction

In the United States, there is no single sentence that satisfies every federal, state and local obligation. A joint statement from the FTC, EEOC, Department of Justice and Consumer Financial Protection Bureau says existing legal authorities apply to automated systems just as they do to other practices.[2] The FTC's consumer-protection remit and the EEOC's employment-discrimination remit are different, but buying a tool from a vendor does not remove the employer's responsibilities. Check the law where the role and candidate are located.

The European Union's AI Act treats many recruitment and selection systems as high-risk. Article 26 says deployers of relevant Annex III systems that make or assist decisions about people must inform those people that they are subject to the system.[4]

In the United Kingdom, the ICO's 2026 recruitment work says candidates should be told when automated decisions are used, how the tools work and how they may affect the application; it also points to contesting a decision and requesting human review.[5] In Australia, the Australian Human Rights Commission's recruitment checklist groups transparency with privacy, redress and keeping humans in control.[6] The checklist is useful guidance, not a substitute for advice on the law applying to a particular employer.

Canada's federal public-service guidance offers a practical model even outside that setting: explain the AI tool's role, the criteria or data it used, the assessment it produced, and how the hiring manager interpreted the output.[7]

Publish the boundary, then make the product match it

RoleSage makes its own boundary public. Its AI Transparency page says AI structures and organises candidate information, surfaces relevant skills and evidence, and supports hiring workflows. It also says AI does not make hiring decisions or automatically accept or reject candidates.[8]

Candidates can review and edit the profile information produced from their resume, then inspect role-specific evidence and match analysis. Before submitting a RoleSage application, they can see the candidate-facing alignment narrative, the values-alignment contribution within Match Analysis, and how their current match compares with the opening's Minimum Match. They can return to the relevant source information, recalculate the analysis, and the final application action asks them to confirm that the details are accurate and reflect their real experience.

That visibility does not make an AI interpretation infallible. It does mean the candidate can see how they are being represented, and the practical effect of that representation, before their application reaches the hirer. Hirers see match bands, supporting skills, activities, answers and gaps rather than receiving an AI-selected winner.

AI helps organise and explain the available material; the hirer owns the decision.

That boundary is useful only while the workflow continues to honour it. A disclosure should therefore be maintained with the tool inventory, assessment design and candidate communication, not written once and forgotten in a privacy policy.

Use a notice people can act on

A practical notice can be short if it is specific:

At [stage], we use [tool or function] to examine [information] against [job-related criteria]. It produces [output], which [explains how it influences the process]. It does not evaluate [verified exclusions]. [Named role] reviews [underlying evidence] and makes [decision]. To request an adjustment or alternative, contact [route]. To correct information or question an outcome, contact [route]. We share data with [providers] and retain it for [period or linked policy]. Our rules for candidate use of AI are [link].

Use the job advertisement for the first signal, the application or assessment instructions for stage-specific detail, and the outcome communication for a meaningful explanation and review route. Update all three when the tool or its influence changes.

Disclosure does not prove that an AI system is accurate, fair or lawful. It does something more basic and necessary: it forces the hiring team to understand the system and gives the candidate a fair chance to participate, seek an adjustment and question an error.

If you cannot fill in the notice without vague claims or missing answers, pause the tool before asking candidates to trust it.

References and further reading

1. ERE: What to Tell Candidates About AI in the Hiring Process - the US practitioner article prompting the recommendation to disclose AI use upfront; some of its broader assurances should be tested against current product evidence and regulator guidance.
2. FTC, EEOC, DOJ and CFPB: Joint Statement on Enforcement Efforts Against Discrimination and Bias in Automated Systems - the US agencies' statement that existing legal authorities apply to automated systems and AI.
3. US EEOC: Artificial Intelligence and the ADA - official resources on disability discrimination, reasonable accommodation and algorithmic assessment tools.
4. European Commission AI Act Service Desk: Article 26 - EU obligations for deployers of high-risk AI systems, including informing people subject to relevant AI-assisted decisions.
5. UK Information Commissioner's Office: What jobseekers need to know about automated recruitment decisions - current UK regulator guidance on notice, impact, contesting decisions and human review.
6. Australian Human Rights Commission: AI and recruitment compliance checklist - Australian guidance on privacy, transparency, candidate redress and accountable human control.
7. Public Service Commission of Canada: Artificial intelligence in the hiring process - Canadian federal public-service guidance on notice, explanation, assessment criteria, accountability and recourse.
8. RoleSage: AI Transparency - RoleSage's public explanation of its assistive AI boundary, candidate control, human oversight and limitations.
9. RoleSage: What AI Hiring Tools Should Explain to Candidates and Hirers - a companion article on evidence, uncertainty and explainable match signals.