Gains:
- Ability to create a governance framework and usage policy for the use of AI in HR
- Ability to embed bias control, transparency and human control principles into processes
- Establishing the discipline to verify AI output, cite sources, and hold responsibility to humans
Throughout this module, we used AI in dozens of HR tasks, from job posting to survey analysis. Now we are establishing the framework that holds all this together: governance. Governance is the system that determines who will use AI in the company, by whom, for what purpose, with what rules and with what control. The use of AI without governance is swift in the short term but dangerous in the long term: inconsistent decisions, privacy violations, implicit bias, and irresponsibility hiding behind the “AI said so” defense. In this final unit we will establish a governance framework that makes AI in HR safe, fair and defensible.
A few terms: Human oversight is the review and approval of AI output by a human. Transparency is the ability to explain where and how AI is used. A bias audit is examining whether a process systematically disadvantages certain groups. Accountability means that the responsibility for a decision belongs to a clear person.
The Golden Rule of HR
The most important sentence of this module is: AI is supportive; Decisions and responsibility that affect people always remain with people. Recruitment, promotion, dismissal, performance outcome, pay—all affect a person's life, livelihood and dignity. AI drafts, summarizes data, offers options, finds patterns in these decisions; but a qualified person bears the final judgment and responsibility. It's just like in engineering: in a safety-critical build, AI's calculation is no substitute for approval from a competent engineer. In HR, making a decision affecting an employee's career on the grounds that "AI said so" is both ethically unacceptable and a legal risk.
Decision type
The role of AI
final decision
Job posting writing
draft production
Human (low risk)
CV screening
Summary and ranking suggestion
Human (mandatory)
Interview evaluation
Note taking, rubric draft
Human (mandatory)
Performance result
Feedback draft
Human (mandatory)
Promotion/pay/termination
Analysis support
human (definitely)
Caution: Position AI as an assistant in HR, not a decision maker. Don't delegate automated rejection/acceptance decisions affecting a candidate or employee to AI. Human review should be a real review, not a formality; Pressing the "confirm" button without thinking is not control.
Establishing a Governance Framework
A good HR AI policy answers four questions: What can it be used for? What can't it be used for? Which vehicles are approved? How to inspect? You can have AI produce a draft of this framework — then adapt it for your own organization:
Your role: an HR governance advisor.Task: Prepare an “AI Use Policy” DRAFT for the HR team.Sections:1) Purpose and scope2) Permitted uses (e.g. drafting, summarization)3) Prohibited/restricted uses (e.g. automated rejection decision, raw personal data)4) Approved tools and data rules5) Human control and accountability6) Transparency: disclosure to candidate/employeeRule: Legal wording Mark with [VERIFY WITH LEGAL/KVKK EXPERT]. This is a draft; The final policy undergoes expert approval.
Bias Control and Transparency
AI can silently carry biases from the data on which it is trained. The heart of governance is to regularly challenge this bias:
Check the following screening criteria and process for bias: - Is there a criterion that indirectly favors/excludes a particular gender, age, origin, school? - Is there an element that is not actually relevant to the job but could create discrimination? - Could the process be putting a group at a systematic disadvantage? Present each risk and recommendations for correction in a table.
A simple principle for transparency: be in a position to give an honest answer when a candidate or employee asks about the use of AI in their process.
Write a simple and honest informational text explaining to candidates where and how AI is used in the hiring process. Clarify what is done by AI and what is done by humans. State the candidate's rights (objection, request for human evaluation).
Verification Discipline
AI can be convincingly wrong; This is called a hallucination (fabrication). The operational leg of governance is to validate each output:
Mark every factual claim, figure, legal reference, and corporate information in this printout with the [VERIFY] tag. Do not present anything as certain if you are not sure. If it's unclear, tell me.
Tip: Make the discipline of verification a habit: Before publishing each AI-generated text, ask “which claim in this sentence can I verify with an independent source?” ask. In particular, numbers, legal deadlines, and corporate claims such as "company policy is this" are the most frequently made and costly mistakes.
Three Mini Cases
Case 1 — Chaos without politics. In one company, every HR professional was using AI with different tools and in different ways; Some were pasting raw CVs, some were setting up automatic rejection. Once a governance policy was published (approved vehicle list, prohibited uses, human supervision rule), the risk was reduced and the process became consistent.
Case 2 — Catching implicit bias. One team noticed a hidden pattern in AI-powered screening that made a particular university stand out. Regular bias auditing revealed this; The criteria were rewritten around job commitment, and the candidate pool was diversified. Without supervision, prejudice would continue silently.
Case 3 — Prevention of hallucination. In one policy draft, AI fabricated a “legal obligation” that did not actually exist. Thanks to the discipline of verification (having every legal claim flagged), it was caught in expert approval. The team turned this into a case and spread the lesson that “AI is convincingly fallible” to the entire team.
Weak Prompt / Strong Prompt
Weak approach: Directly implement whatever the AI says; verifying output, questioning process.
The result: implicit bias, fabricated information, and delegated responsibility; A risk that may explode one day.
Powerful approach:[certified tool + clear rules of use + human moderation + bias control + transparency + discipline to verify every factual claim]
The result: a fast but safe, fair and defensible use of AI.
Common mistakes
- Transferring responsibility to AI. "AI said so" is neither an ethical nor a legal justification.
- Reducing control to a formality. Hitting “approve” without thinking is not true control.
- Never questioning prejudice. AI is not without bias; Regular inspection is essential.
- Avoiding transparency. Hiding the use of AI undermines trust and reputation.
- Skipping verification. The hallucination is convincing; verify each factual claim.
- Writing the policy and not implementing it. A policy that sits on a shelf is a policy that does not exist.
In summary
- Governance is the system that determines who will use AI, for what purpose, and with what rules and control; It is indispensable in HR.
- Golden rule: AI is supportive; Final decisions and responsibilities that affect people, such as recruitment, promotion, and termination, always remain with people.
- Establish an AI usage policy: allowed/prohibited uses, approved tools, human moderation, transparency.
- Check for bias regularly and keep AI use transparent to the candidate/employee.
- Verify every factual claim, figure and legal reference; The hallucination may be convincing, the responsibility is yours.
Application task
Draft a one-page “AI Use Policy” for your own HR team (or a fictional team). Have the AI produce a draft that includes: permitted uses, prohibited uses (such as automatic rejection, raw personal data), approved tools, human review rule, and transparency clause. Then have your screening process (real or fictional) inspect the AI for bias and correct any risks that emerge. Finally, mark each legal statement in the policy as “to be verified by an expert.”
checklist
- [ ] Is every final decision affecting a human being made by a human being?
- [ ] Are permitted and prohibited uses written down?
- [ ] Are only approved, safe tools used?
- [ ] Is the process regularly audited for bias?
- [ ] Is the use of AI transparent to the candidate/employee?
- [ ] Is every factual claim and legal reference verified?
- [ ] Is the policy actually implemented, not just written?
Module Exam
1. What is the best guideline for inclusivity when writing a job posting with AI?
- A) Asking to eliminate discriminatory statements that imply gender, age or a certain group ✔
- B) Liven up the ad by adding phrases such as 'A young and dynamic team'
- C) Using language that appeals only to a certain gender
- D) Leaving attributes as vague as possible
Explanation: Purifying the advertisement language from expressions that imply a certain gender, age or group (such as 'young and dynamic', 'male element') is necessary both legally and ethically and expands the candidate pool. Explicitly telling the AI to eliminate these statements is an effective method.
2. Which of the following is true to reduce the risk of bias when screening CVs with AI?
- A) Giving the AI full power to automatically reject candidates
- B) Limiting the evaluation to objective criteria related to the job and leaving the final decision to people ✔
- C) Putting name, age and gender at the center of the evaluation
- D) not doing any checks because the AI is unbiased
Disclosure: AI can reflect biases from the data on which it is trained. The way to reduce the risk; limiting the evaluation to objective criteria that are truly tied to the job, excluding irrelevant information such as name/gender/age, and leaving the final decision to a human.
3. What is most useful to ask the AI for a structured interview?
- A) Randomly, different questions are asked to each candidate.
- B) Questions that only ask about the candidate's hobbies
- C) Competency-based common questions and scoring rubric to be asked to all candidates ✔
- D) Free chat without evaluation criteria
Explanation: For a fair and comparable interview, common competency-based questions to be asked to all candidates and an evaluation guide (rubric) showing how to score each question are ideal. This de-subjectifies the decision.
4. What is the best approach when preparing a rejection (negative result) message to a candidate with AI?
- A) Send a long text with detailed, personal criticism
- B) Not responding at all and leaving the candidate in limbo
- C) Using expressions that attribute the reason for rejection to the age or status of the candidate
- D) Producing a respectful, short, thanking message that maintains the brand tone ✔
Description: The rejection message is one of the most critical moments of the candidate experience. Being respectful, brief, thanking the candidate, and leaving the door open if possible maintains a ton of employer branding. When writing the reason for rejection, statements that would create a legal risk, imply discrimination or are unfounded should be avoided.
5. What is the most efficient way to use AI when creating onboarding content for a new employee?
- A) Specifying the role and duration, producing a structured plan and checklist, then filling it with company-specific information ✔
- B) Using a single document that is general, not role-specific
- C) Leaving onboarding entirely to AI and not personalizing it at all
- D) Giving no plan to the new employee and waiting for him to learn on his own
Description: Giving role, department and duration (first day/week/month) information to AI and having it produce a structured plan, checklist and 30-60-90 day targets; then populating it with company-specific information is the most efficient approach.
6. What is the most important element that increases quality when preparing an educational material with AI?
- A) Just saying 'prepare a training' and not giving details
- B) Keeping the content as theoretical and example-free as possible
- C) Clearly stating the target audience level, measurable learning target and duration ✔
- D) Producing the longest content without considering the target audience
Description: Clearly stating the level of the target audience, measurable learning goal and duration; Adding scenarios, examples and exercises to the content makes the training effective. If the level and goal are unclear, the content will be either too simple or too complex.
7. What is the best approach to drafting performance feedback with AI?
- A) Allowing AI to fabricate feedback without knowing actual performance
- B) Input real observations and produce constructive drafts, keeping the final decision with the manager ✔
- C) Send the draft to the employee without any editing
- D) Keeping feedback either all negative or all positive
Description: AI is helpful in putting concrete behavioral examples into constructive language; However, the content of the feedback should be based on actual observations and the final evaluation should be the responsibility of the manager. AI produces draft, human decides.
8. What does the SBI (Situation-Behavior-Impact) framework provide in performance feedback?
- A) Focus feedback on personality traits
- B) Emphasizing only the positive aspects and hiding the areas of improvement
- C) Keeping the feedback general and abstract
- D) Giving concrete and observation-based feedback by separating the situation, the observed behavior and its impact ✔
Description: SBI; It structures the feedback as situation, observed behavior and impact. This makes the feedback concrete and observational, away from personality attacks; It clarifies what the employee needs to change and why.
9. What is the most critical step when drafting an HR policy text (e.g. leave or disciplinary policy) with AI?
- A) Publishing the AI-generated text directly as official policy
- B) Review and approve the draft by an expert/lawyer in accordance with the current legislation ✔
- C) Accepting the legal articles as given by AI as absolute truth
- D) Not letting anyone review the policy
Explanation: HR policies have legal consequences. Although AI produces a good first draft, the text must be reviewed and approved by an expert/legal according to applicable labor legislation and company conditions. Legal statements made by AI should not be used without verification.
10. What is the most useful approach when analyzing open-ended employee survey responses with AI?
- A) Grouping the responses according to main themes and emotional state and revealing patterns ✔
- B) Choosing a few answers at random without reading each answer one by one
- C) Trying to match anonymous answers with people
- D) Only receive positive comments and ignore the negative ones
Description: Hundreds of free text responses can be grouped by main themes and mood (positive/negative/neutral) with AI; this reveals overlooked patterns. Findings must still be interpreted by human beings and anonymity must be maintained.
11. Which of the following is true in terms of KVKK when using AI with candidate and employee data?
- A) Send and store all candidate data to any free tool
- B) Using personal data for purposes other than purpose, in an unlimited manner
- C) Processing data limited to the purpose, at a minimum level and with appropriate security measures ✔
- D) Processing employee data without consent and basis
Description: Personal data; It should be processed limited to the purpose, as much as necessary (data minimization) and with appropriate security measures. Sending personal data, such as candidate CVs, to tools that do not have institutional guarantees and analyzing them without anonymization is a KVKK risk.
12. What is the best course of action in terms of personal data before having a CV summarized by an AI tool?
- A) Masking direct identification information and processing only job-related information ✔
- B) Send the CV as is without any changes
- C) Including all the candidate's contact and identity information in the analysis
- D) Store the data in the vehicle indefinitely and use it for other purposes
Explanation: In accordance with the data minimization principle; If a CV is sent to a tool without corporate security, masking direct identification information such as name-surname, telephone, address, date of birth and processing only job-related information (experience, competence) reduces the risk. The healthiest thing is to use a corporate/safe vehicle.
13. How should an HR professional use AI when they see low 'application → interview → offer' conversion rates in the recruitment funnel?
- A) Ignore the metrics and continue the process as it is
- B) Having conversion rates interpreted and possible causes and improvement suggestions generated, then validating them with data ✔
- C) Giving AI full authority to automatically screen all candidates
- D) Making decisions only by intuition without measuring metrics at all
Description: AI is a good helper in interpreting metrics and generating possible causes and improvement suggestions: it can suggest reviewing posting language, candidate communication, process speed. However, suggestions should be validated with data and field knowledge, and the decision should remain with the person.
14. Which of the following is most accurate as a general principle of using AI in HR processes?
- A) AI should make hiring and firing decisions alone
- B) Human supervision should be removed because it reduces efficiency
- C) Since AI is unbiased, its decisions should be implemented without question
- D) AI is a supporting tool; Final decisions affecting people and responsibility remain with people ✔
Description: Decisions made in HR directly affect people's careers and lives. AI is a powerful drafting and analysis tool, but final judgment and responsibility for decisions such as hiring, promotion, firing, etc. should always remain with the human; AI is a supporter, not a decision maker.