Introduction

The CV is polished, keyword-rich and closely aligned to the job description. The candidate appears almost perfectly matched to the vacancy. Then the interview begins, and the picture changes. Examples are vague, the candidate struggles to explain decisions they supposedly made, and it becomes difficult to distinguish personal contribution from team output.

Generative AI has made it easier for candidates to improve grammar, structure experience and tailor applications. That is not automatically dishonest. The real recruitment risk arises when polished language is mistaken for proof of competence, or when AI is used to invent experience, exaggerate achievements or hide weak understanding.

Employers therefore need to ask a better question. Not, “Was this CV written by AI?” but, “Can the candidate substantiate the claims and demonstrate the capability required by the role?”

The objective is not to become an AI detective. It is to build a recruitment process in which presentation cannot substitute for evidence. Pasted markdown

Can Employers Reliably Identify an AI-Generated CV?

Not with certainty from writing style alone.

Highly polished language, repeated keywords, balanced bullet points or formal wording may indicate AI assistance, but they may also reflect professional editing, templates, coaching or the candidate’s own communication ability. Likewise, a poorly written CV does not prove that it is authentic or that the candidate lacks capability.

Employers should therefore avoid rejecting candidates simply because a CV “looks AI-generated” or because an automated detection tool assigns it a high AI probability. Such tools do not prove whether the candidate’s underlying experience is true.

A CV should be treated as a set of claims to examine, not a verified account of capability. Pasted markdown

AI Assistance Is Not the Same as Dishonesty

The key issue is not whether AI was used, but how it was used.

Using AI to improve grammar, organise genuine experience, identify transferable skills or draft a cover letter from truthful information may be legitimate. In contrast, inventing roles, qualifications, responsibilities, achievements or metrics is misrepresentation. Using AI during a controlled assessment where the employer explicitly prohibited it is also a breach of the stated rules.

This distinction matters because AI is becoming a normal workplace tool. Many employees will be expected to use technology productively. For some roles, the ability to use AI well may itself be valuable. The candidate must still remain accountable for the final output, verify accuracy and demonstrate independent judgement where the job requires it. Pasted markdown

What Has Changed About the Value of a CV?

The CV is still useful. It can show career chronology, employers, job titles, qualifications, claimed responsibilities and areas requiring clarification.

What it cannot safely establish on its own is whether the candidate:

• personally delivered the achievements described;
• understands the technical language used;
• can solve comparable problems in a new environment;
• exercises sound judgement under pressure;
• communicates effectively without extensive assistance;
• behaves consistently with the organisation’s values; or
• will perform successfully in the actual role.

This was true before generative AI. Employers were already vulnerable to copied job descriptions, professionally written CVs and exaggerated claims. AI has simply increased the scale and sophistication of the problem. Pasted markdown

Warning Signs That Require Verification, Not Automatic Rejection

No single feature proves that a CV is false or AI-generated. However, some signs justify deeper questioning:

• responsibilities sound impressive but lack scale, context or personal ownership;
• every achievement appears perfectly aligned with the vacancy;
• complex results are claimed without explaining the original problem or constraints;
• technical terms are used correctly on paper but poorly understood in conversation;
• percentages or financial results are presented without a credible basis;
• the candidate moves between “I” and “we” without clarifying individual contribution;
• dates, titles or reporting relationships differ across the CV, profile and interview;
• the candidate cannot explain a failure, trade-off or difficult decision; or
• the quality of the CV differs significantly from role-relevant writing completed under controlled conditions.

These are investigation points, not proof. Employers should ask for clarification and assess the response against consistent criteria. Pasted markdown

Move From Document-Led to Evidence-Led Recruitment

An evidence-led process does not discard the CV. It places the CV at the beginning of the assessment sequence rather than at the centre of the hiring decision.

A strong process should answer different questions at different stages:

1. CV: What experience and capability is the candidate claiming?
2. Screening: Can the candidate explain the scope, context and personal contribution?
3. Work sample: Can the candidate apply knowledge to a realistic task?
4. Structured interview: How does the candidate reason, decide and learn?
5. Verification: Can important claims be independently supported?
6. Final decision: Does the combined evidence meet the agreed success profile?

Confidence should increase as the employer compares what the candidate claims, explains, demonstrates and can have independently verified. Pasted markdown

Step 1: Define Capability Before Reviewing Applications

AI-generated applications become much harder to assess when the employer has not defined what success in the role actually requires.

A long job description is not the same as a success profile. Before screening, the employer should identify the most important outcomes in the first 6–12 months, essential technical knowledge, decisions the employee must make independently, problems the role must solve, stakeholders the person must influence, behaviours required in the work environment and any non-negotiable qualifications, licences or experience.

This helps recruiters distinguish relevant evidence from language that merely sounds impressive. It also reduces the risk of selecting a polished candidate simply because the interview panel was impressed by presentation rather than measuring agreed requirements. Pasted markdown

Step 2: Turn CV Claims Into Verification Questions

Every major CV claim can be translated into questions about context, action and result.

If a candidate says they “led a successful transformation,” ask what was changing, what authority they had, who resisted, what they personally decided and what changed as a result.

If they claim they “reduced costs by 30%,” ask which costs, over what period, what the baseline was, how the figure was calculated and what trade-offs were involved.

If they say they “implemented an ERP or HRIS,” ask which modules, what their personal role was, what went wrong, what decisions they made and how adoption was measured.

Candidates who genuinely performed the work may not remember every number immediately. The aim is not to create a memory test. It is to determine whether the person can reconstruct the work with credible detail, ownership and judgement. Pasted markdown

Step 3: Use Structured Pre-Screening and Interviews

A short pre-screening conversation can verify essential information before a full interview.

Ask each candidate the same core questions about why they are interested in the role, the most relevant problem they have solved, the size and complexity of their responsibilities, their personal contribution to a claimed result, tools or systems they have actually used, remuneration expectations, notice period and any important requirement they do not yet meet.

During the main interview, avoid generic questions such as “Are you a good leader?” Instead, ask the candidate to unpack a specific experience.

Useful follow-up questions include:

• What was happening when you took responsibility?
• What did you diagnose as the main problem?
• Which actions did you take personally?
• What resistance or trade-off did you face?
• What result followed, and how was it measured?
• What did not work?
• What would you do differently now?

For technical roles, move beyond definitions. Ask the candidate to interpret information, compare options, identify risks or explain how they would diagnose a problem.

Genuine understanding is usually visible in how a person reasons, not only in the terminology they use. Pasted markdown Pasted markdown

Step 4: Use Job-Relevant Work Samples

A work sample is one of the strongest ways to move beyond presentation because it asks the candidate to perform a task similar to the actual work.

Examples could include an accountant reviewing a short financial scenario and identifying control weaknesses, an HR manager responding to an employee-relations case, a salesperson preparing an account-entry plan, a data analyst interpreting a controlled dataset or a manager presenting a 90-day response to a defined business problem.

The exercise should be proportionate. Employers should not ask candidates to perform substantial unpaid consulting work or solve live commercial problems.

A fair work sample should test job-relevant capability, give candidates comparable instructions and time, state clearly whether AI or internet research is allowed, use an agreed scoring guide and include follow-up questions about the candidate’s reasoning.

The follow-up discussion can be as valuable as the written output. Ask what assumptions were made, what information was missing, what the candidate would verify and what they would change with more time. Pasted markdown

Step 5: Verify Qualifications, Experience and Achievements Proportionately

Verification should match the risk and seniority of the role.

Depending on the position, employers may confirm academic and professional qualifications, licences or professional registration, employment dates and job titles, responsibility for major projects, relevant work history or portfolios and structured references.

Reference questions should test the same claims examined during selection.

For example:

What responsibilities did the candidate personally own? How did they perform under pressure? What support did they require? Would the referee reappoint them to a comparable role?

Checks should remain relevant and transparent. They should not become an uncontrolled search into the candidate’s private life. Pasted markdown

Step 6: Integrate the Evidence

The final decision should combine evidence from several stages.

A highly polished CV should not outweigh a weak work sample. Equally, nervous interview performance should not automatically outweigh strong job-relevant evidence if polished speaking is not central to the role.

A practical decision framework can assess relevant experience, technical capability, judgement, delivery, communication, values and conduct, and learning capacity.

Panel members should record evidence before debating the overall candidate. This reduces the risk that confidence, personal similarity or polished language becomes a substitute for capability. Pasted markdown

Should Employers Ask Candidates Whether They Used AI?

They may, where the answer is relevant, but the question should be framed carefully.

A practical application statement could say:

You may use appropriate tools to prepare your application, but all information submitted must be accurate and must reflect experience and capability you can explain and substantiate. Assessment instructions will state where tool use is permitted or restricted.

This sets an honesty standard without assuming that all AI assistance is misconduct.

For roles that will use AI, employers may intentionally include an AI-enabled exercise. The candidate can be asked to produce an output, identify weaknesses, verify claims and explain where human judgement was required.

Where independent performance is essential, the employer may prohibit external assistance for a specific assessment. The rule should be stated in advance and applied consistently. Pasted markdown

Why AI-Detection Scores Should Not Make the Decision

AI-detection tools may appear attractive, but they create important risks.

A score assesses characteristics of the text, not whether the candidate’s experience is true. False positives may disadvantage honest candidates. Employers may also not know how the tool was trained, how it performs across language backgrounds or what happens to uploaded CV data.

If detection technology is used at all, it should be treated only as a limited signal and never as proof. Human review, structured assessment and verification remain necessary.

In practical terms, the employer remains accountable for the recruitment decision even where technology supports the process. Pasted markdown

Fairness, Data Protection and Automated Recruitment in Kenya

Technology does not remove an employer’s existing responsibilities.

Section 5 of Kenya’s Employment Act, 2007 addresses equal opportunity and non-discrimination in employment, including recruitment. Candidate CVs, assessments, interview notes and screening results are also personal data, so employers should manage them in line with Kenya’s Data Protection Act, 2019. Pasted markdown

Employers using AI-enabled recruitment tools should examine:

• the lawful and transparent basis for processing candidate data;
• what information the tool collects or infers;
• whether third-party providers retain or reuse candidate information;
• who can access candidate records;
• how long information is retained;
• whether the tool may produce discriminatory outcomes;
• where meaningful human review occurs; and
• how candidates can correct inaccurate information or raise concerns.

Higher-risk automated recruitment tools should receive appropriate legal, privacy and governance review before deployment. Pasted markdown

A Practical Employer Policy for AI Use in Recruitment

Employers do not need a lengthy policy to establish minimum standards. A practical position can cover:

1. Accuracy: Candidates remain responsible for all information submitted.
2. Permitted assistance: AI may be used for preparation unless a specific assessment says otherwise.
3. Disclosure: Disclosure is required only where relevant and clearly requested.
4. Assessment rules: Each exercise states whether tools are permitted, restricted or required.
5. Verification: Material qualifications, experience and claims may be checked.
6. Human review: No adverse decision is made solely from an unexamined screening or detection score.
7. Fairness: Selection criteria are job-related and monitored for inconsistent impact.
8. Privacy: Candidate information is processed only through approved systems and providers.
9. Documentation: Reasons for progression and rejection are recorded consistently.
10. Communication: Applicants receive sufficient information about the process and any material automated use.

The policy should align with the organisation’s wider recruitment, data-protection, information-security and acceptable-use frameworks. Pasted markdown

How a Recruitment Partner Can Help

Professional recruitment support should offer more than CV collection.

A capable recruitment partner can help employers define a realistic success profile, verify career scope during screening, design structured interview questions, introduce proportionate work samples, assess candidate evidence consistently, manage references and qualification checks, communicate assessment rules clearly and maintain fair, secure candidate records.

Technology can support this work, but accountability must remain visible.

Recruiters should be able to explain why a candidate was shortlisted, what evidence was obtained and where uncertainty remains. Pasted markdown

Final Thought: Recruit the Best Evidence, Not the Best Application

AI has not made the CV irrelevant. It has made overreliance on the CV more dangerous.

Employers should assume that candidates may receive help when preparing applications, just as organisations use technology to write job advertisements and manage recruitment.

The fair response is not suspicion of every polished applicant. It is a transparent process that gives candidates a reasonable opportunity to demonstrate genuine capability.

When CV claims are tested through structured questions, realistic work samples, consistent interviews and proportionate verification, artificial polish loses its power to distort the decision.

The organisation does not need to prove who used AI. It needs to prove that its preferred candidate can perform.

ACCUREX supports employers through targeted recruitment, structured assessment, executive search and high-volume hiring. We help clients move beyond application quality and evaluate whether candidates can deliver in the actual role.

Contact ACCUREX to discuss your vacancy or request a review of your candidate-assessment process. 

www.accurex.co.ke or email info@accurex.co.ke