Gains:
- Ability to prepare first drafts of policy and procedure texts with structure and clarity with AI
- Ability to produce announcements, emails and sensitive internal communication texts with the right tone and empathy
- Ability to establish a process that requires expert/legal approval on legal and sensitive policy texts
HR is the voice of the company as well as the rule maker. HR constantly produces texts, from leave policies to remote working rules, from disciplinary processes to holiday announcements, from raise letters to restructuring statements. Some of these texts have legal consequences, while others directly shape company culture and employee morale. Artificial intelligence (AI) produces fast and consistent first drafts of both types of text. But the role of AI in policy texts and its role in announcement texts are very different; In this unit, we will clarify this distinction and learn how to work safely in both areas.
A few terms: Policy is a written rule, often with a legal basis, that binds employees. A procedure is a document that explains step by step how to implement a policy. Internal communication is all of the company's announcements, e-mails and information to its employees. Tone is the emotional color and formality of the text; the same information is perceived very differently with different tones.
Internal Communications: AI's Comfort Zone
AI can be used freely in texts such as announcements, emails, congratulatory and thank you messages; The risk is low, the benefit is high. The real skill here is getting the tone right.
Your role: a corporate communications specialist.Task: Write the following announcement as an email to employees.Subject: Transition to new hybrid work arrangement (office 2 days a week)Tone: clear, positive, employee-focused. Explain the rationale and benefit of the change. Length: short, scannable with bullet points. Specify at the end where the questions should be forwarded. Rule: Leave the exact date/person/location information [SQUARE BRACKETS] and I will fill it in.
Tip: Always specify the tone in internal communication: “clear and positive,” “empathetic,” “formal but warm,” etc. The same information should be delivered in a completely different tone for bad news (e.g., benefit cut) and good news (e.g., new benefits). AI sets the tone surprisingly well — but only if you say it.
When it comes to sensitive news (restriction, change, bad news), striking a balance of empathy and transparency is critical:
Write a text announcing the following sensitive news to employees. News: This year's performance bonus budget is lower than last year. Tone: honest, empathetic, not defensive. Don't hide the truth by softening it, but give hope and future perspective. Don't be accusatory or cold. Give a respectful answer to the employee's "why" question.
Policy Texts: AI's Mindful Space
Policies such as leave, discipline, overtime, and data privacy are legally binding. Here, AI produces the draft, but the text is never directly put into effect. Safe step by step process:
- Emphasize the draft. Tell the AI “this is a draft, not the final text.”
- Give the structure. Purpose, scope, definitions, rules, process, exceptions.
- Tick the legal phrases. “Verify” label instead of made-up.
- Leave company-specific information blank. You fill in the numbers and times.
- Get expert/legal approval. This step cannot be skipped.
Your role: an HR policy writer.Task: Create a DRAFT for the annual leave policy.Structure: purpose, scope, definitions, leave entitlement, request process,approval flow, exceptions.Rule: Mark legal statements and durations with the [VERIFY ACCORDING TO REGULATION] tag. FITTING a specific article of law or period; Make it clear that you are not sure. This is a draft, not the final text.
The model lays out a good framework here (“Annual paid leave is determined by the employee's seniority [VERIFY REGULATION]”), but is clearly restricted from making up legal details. You can also question the shortcomings of an existing policy:
List the points that are missing or unclear in this draft leave policy:- Which situations remain undefined? (e.g. half day leave, excused leave) - Which statement is open to more than one interpretation? - What are the points that may confuse the employee? Offer a suggestion, but do not make up the legal period/article; If it's unclear, say "verify".
Three Mini Cases
Case 1 — The power of tone. A company announced its decision to return to the office in a cold, imperative manner, and employees reacted. HR rewrote the same information with AI in an empathetic and reasoned tone: the reason for the decision, its benefit, and the question channel were added. After the second announcement, the number of negative comments in the internal channel decreased significantly.
Case 2 — Fake substance hazard. One HR expert directly told AI to “write leave policy”; The text contained a convincing but incorrect annual leave period. Fortunately, he was caught during the legal approval step. After this incident, the team standardized the "[VERIFY ACCORDING TO REGULATION]" and "fabricate" constraint to all policy prompts.
Case 3 — Simplification. No one was reading a 6-page remote working policy written in heavy legal language. HR had AI simplify the text with the constraint “don't change the meaning, just improve clarity”; The document was reduced to 2 pages, the reading rate increased, and the number of repetitive questions received by HR decreased by approximately one third.
Weak Prompt / Strong Prompt
Weak prompt:Write us a permissions policy.
The result: a text that looks real but may contain incorrect legal periods and clauses, and is risky if used directly.
Strong prompt: [outline emphasis + clear structure + “mark legal statement, don’t make it up” constraint + leave company-specific information blank with square brackets]
The result: a secure draft prepared for review, risks flagged, and ready for expert approval.
The Role of AI by Text Type
text type
The role of AI
Mandatory step
Announcement / email
Full outline, tone adjustment
Information accuracy check
Celebration / inner message
free production
Light review
Sensitive news (outage, etc.)
Draft + tone
Empathy and fact checking
Leave/overtime policy
draft skeleton
Legal/expert approval
Disciplinary/termination process
just skeleton
Must have legal approval
Caution: Do not publish any policy text produced by AI without approval by an expert or legal unit in accordance with applicable labor legislation and company conditions. The model can convincingly provide information that is outdated or pertaining to the law of another country. An incorrect leave period or termination procedure puts the company at serious legal risk. AI speeds up the first draft; Legal correctness is always the responsibility of man.
Common mistakes
- Publishing the policy directly. The legal text should not come into force without expert approval.
- Making AI make up legal details. Have the duration and items marked and verified.
- Not specifying the tone. If the tone is not stated, the text will come out either cold or inappropriate.
- Breaking sensitive news coldly. Lack of empathy creates reactivity and loss of trust.
- Having AI fill in company-specific information. Leave the date/person/number blank with square brackets.
- Shifting the meaning while simplifying. Be sure to add the "do not change meaning" constraint.
In summary
- AI can be used freely in internal communication texts (announcement, email, celebration); Be sure to specify the tone and check the accuracy of the information.
- Balance empathy and transparency on sensitive news; Use respectful and hopeful language without hiding the truth.
- Policy texts (leave, discipline, overtime) have legal consequences; AI only produces drafts, the text does not come into effect directly.
- Don't make AI adapt legal wording and deadlines; have it marked with the "verify" tag and confirmed by the expert; You fill in the company-specific information.
- Do not publish any policy text without legal/expert approval; AI speeds up drafting, legal accuracy is human responsibility.
Application task
Produce two texts. The first is a low-risk internal communication: Have the AI print out a change announcement, specifying the tone (“empathetic,” “clear and positive”) and leave the exact information blank with square brackets. The second is a high-risk draft policy: ask for a leave or remote work policy; Mark legal expressions with the "[VERIFY]" tag and impose a "fabricated" constraint. Then make a list of each ticked statement in the policy draft that needs to be confirmed with actual legislation/expert opinion.
checklist
- [ ] Is the risk of the text type (low internal communication or high politics) classified correctly?
- [ ] Is the tone clearly stated in the internal communication text?
- [ ] Has the balance of empathy and transparency been struck in sensitive news?
- [ ] Have legal statements been marked in the policy draft and a "fabricated" restriction placed?
- [ ] Is the company-specific date/person/number left blank with square brackets?
- [ ] Is legal/expert approval planned for the policy text?
- [ ] Was the "meaning change" constraint used in simplification?