How to use AI to build a simple staff communication script for delivering bad news — a redundancy, a pay freeze, or a policy change

Only 23% of employees strongly agree their manager communicates effectively, according to Gallup's 2023 workplace survey{:target="_blank"} — which means most small business owners are already starting from a communication deficit before they ever deliver a piece of hard news. If you need to know how to tell an employee they are being let go, a script built with AI is the fastest way to get it right. This post walks you through how to use an AI tool to build a working script for three of the most common difficult conversations: redundancy, pay freeze, and policy change. Done properly, a 20-minute prompting session produces a script, a set of employee objections to rehearse against, a follow-up email, and a team talking points sheet — the full communication package, not just a rough draft.

What You Need Before You Start

ChatGPT{:target="_blank"}, Claude{:target="_blank"}, or Gemini{:target="_blank"} — any of the three main AI assistants will handle this task. ChatGPT-4o, Claude 3.7 Sonnet, and Gemini 2.0 Flash are all capable of producing role-specific, situation-specific scripts in under two minutes when given structured input. The free tier covers this use case on all three platforms. If you're doing this regularly, ChatGPT Plus costs $20/month (as of March 2026); Claude Pro costs $20/month; Gemini Advanced runs $19.99/month as part of Google One AI Premium{:target="_blank"}. You do not need a paid plan to follow this guide.

Time required: 20–30 minutes for a full communication package including script, objection prep, follow-up email, and team talking points. Basic script only: 5–10 minutes.

Skill level: No technical background required. You need to be able to type a paragraph of context into a chat window. That's the entire prerequisite.

The Three Scenarios That Need Different Handling

Before you prompt anything, you need to understand what kind of conversation you're preparing for. The emotional framing, the legal disclosures, and the tone vary significantly across the three most common scenarios.

Redundancy or layoff is the highest legal-risk scenario. UK employment law via ACAS{:target="_blank"} requires a fair process including individual consultation, a genuine redundancy reason, and notice pay. In the US, at-will employment reduces the procedural floor, but careless language still creates wrongful termination exposure — particularly if the conversation could be read as pretextual or discriminatory. The critical phrase here is: the role is being made redundant, not the person. Conflating redundancy with performance dismissal is a liability. Your script must keep that distinction explicit.

Pay freeze or pay cut carries lower legal risk but high retention risk. Research from Harvard Business Review{:target="_blank"} shows that vague or indirect language during difficult conversations increases perceived unfairness and employee anxiety. A pay freeze script needs to acknowledge disappointment directly, explain the business reason plainly, and commit to a specific review timeline — not "we'll revisit this when things improve," but a named quarter or date. Vague promises increase resentment and turnover, which costs more than the pay increase you were trying to avoid.

Policy changes — return-to-office mandates, benefit reductions, shift pattern changes — are the lowest legal risk of the three but carry the highest culture risk in a small team. When everyone knows each other, a tone that reads as corporate or dismissive lands worse than it would in a large company. The rationale matters more here than in the other two scenarios.

What to Give the AI Before You Ask It to Write Anything

The quality of the output is directly proportional to the specificity of the input. These five data points are the difference between a generic script you'll need to rewrite entirely and one you can use with light editing.

1. The employee's role and tenure. "Senior account manager, four years with the company" produces a different script than "part-time warehouse operative, six months in." Tenure affects tone and, in the UK, statutory redundancy pay calculations.
2. The reason for the decision — without assigning blame. "The role is no longer required due to restructuring of the client services function" is usable. "We've decided to go a different direction" is too vague and sounds evasive. Be specific about the business driver, not the individual.
3. The jurisdiction. UK, US state, or elsewhere — this determines what legal checkpoints the script needs to hit. Even if you're not asking the AI for legal advice, specifying jurisdiction prompts it to flag the right procedural elements (notice period, appeal rights, consultation requirements).
4. What happens next. Severance terms, notice period, final pay date, return of company property, access to an employee assistance program if you have one. If you don't know these details before the conversation, you're not ready to have the conversation.
5. The desired tone. "Direct but compassionate" is a workable instruction. "Professional" is too vague to be useful. If you know this person well and the team is small, tell the AI that — it will adjust the register accordingly.

How to Prompt the AI to Write Your Let-Go Script

Use this prompt structure as your starting point. Paste it directly into ChatGPT, Claude, or Gemini, filling in the bracketed sections.

You are helping a small business owner prepare for a difficult staff conversation. Write a spoken script — not a formal letter — for the following situation.

Scenario: [redundancy / pay freeze / policy change] Employee role: [job title] Tenure: [X years / months] Reason for decision: [specific business reason, 1–2 sentences] Jurisdiction: [UK / US — state if relevant] What happens next: [notice period, severance if applicable, next steps] Tone: Direct but compassionate. This is a small team where the manager and employee know each other well.

The script should:

• Open by getting to the point within the first two sentences — do not bury the news
• Clearly state [the role is redundant / there will be no pay increase this year / the policy is changing] in plain language
• Explain the business reason briefly without over-justifying or apologising excessively
• State what happens next in concrete terms
• Leave space for the employee to respond
• [For redundancy in UK: include a reference to the right of appeal and the consultation process]
• [For redundancy in US: avoid any language that could be read as performance-related or discriminatory]

End with a note on timing: when in the week and day is best to have this conversation, and why.

The output you should expect: a 300–500 word spoken script structured in clear paragraphs, with natural pauses indicated, plus a one-paragraph timing recommendation. Verify that the script names the decision explicitly in the first 60 words — if it doesn't, the AI has buried the lead and you should ask it to revise the opening.

The reason the timing note matters: SHRM guidance{:target="_blank"} recommends early in the week and early in the day — not Friday afternoon — so the employee has access to support. An AI that buries this recommendation is giving you incomplete output. Ask for it explicitly.

How to Stress-Test Your Script by Asking the AI to Play the Employee

This is the step most owners skip, and it's the most valuable part of the workflow. Once you have your script, send this follow-up prompt:

Now play the role of the employee receiving this news. Respond to the script as that employee — include at least five likely questions or emotional reactions they might have, ranging from practical ("What about my pension contributions?") to emotional ("Why me and not someone else?"). For each reaction, suggest how the manager should respond in 2–3 sentences.

What you get is essentially a rehearsal document — a list of the objections and questions you need to have answers for before the real conversation. The common ones that surface: questions about selection criteria for redundancy, questions about whether the decision is final, requests to negotiate, and expressions of anger or disbelief. If any of those questions expose a gap in your preparation — you don't actually know the answer to "how was I selected?" — that's critical information to have before you're sitting across from the employee.

What to Leave Out — Phrases That Create Liability

The AI will generally avoid these if you've given it jurisdiction context, but verify the draft against this list before you use it.

Never say "we've decided to go in a different direction" as the sole explanation for a redundancy. It sounds like a performance dismissal in disguise and creates grounds for an unfair dismissal claim in the UK.

Never mention protected characteristics — age, gender, disability, pregnancy, race, religion — even in passing or with good intent. "We needed to keep the more junior staff" is age-adjacent language. Remove it.

Never make unconfirmed promises. "We might be able to bring you back in six months" is a contractual risk if circumstances change. If you're not certain, don't say it.

Never read word-for-word without rehearsing. A script read from a screen sounds corporate and cold. The research is clear: how the message is delivered matters as much as the message itself. Use the script to memorise the structure and key phrases, then speak naturally.

Turning One Prompt into a Full Communication Package

Once the script and objection prep are done, two more prompts complete the package.

Follow-up email:

Based on the script above, write a brief follow-up email (150–200 words) that the manager can send to the employee within two hours of the conversation. It should summarise the key points discussed, confirm next steps and dates, and provide a contact for any further questions. Tone: warm but factual.

Team talking points:

Write a short set of talking points (bullet format, under 150 words) for the manager to use when informing the rest of the team about this change. The points should explain what has happened without disclosing private details, reassure the team about their own positions, and invite questions. Tone: transparent and calm.

The full package — script, objection prep, follow-up email, team talking points — takes 20–30 minutes to produce and covers every communication touchpoint created by a single difficult event. The alternative is improvising each one under stress, which is where legal exposure and team trust damage actually happen.

---

When Something Goes Wrong

Symptom: The script sounds corporate and generic, nothing like how you actually talk. Root cause: You didn't give the AI enough context about the relationship and team size. Fix: Add this line to your original prompt — "This is a 6-person team. The manager has worked alongside this employee directly for [X] years. The tone should reflect that personal context, not a corporate HR department." Then regenerate.

Symptom: The script buries the main news three or four paragraphs in. Root cause: The AI defaulted to a softening structure that prioritises emotional cushioning over directness. Fix: Add this explicit instruction — "The decision must be stated clearly in the first two sentences. Do not use a preamble longer than one sentence before delivering the news." HBR's research is unambiguous that direct language reduces emotional fallout, not increases it.

Symptom: The objection prep produces generic questions that don't match your actual situation. Root cause: The "play the employee" prompt didn't have enough role-specific context. Fix: In the follow-up prompt, add the employee's role, tenure, and any specific circumstances — "This employee has been here eight years and recently turned down an external offer to stay." The more specific the character brief, the more realistic the objections.

---

What to Do Next

Save your script template inputs — the five data points — as a reusable document so that next time you face a difficult conversation, you're filling in known fields rather than starting from scratch. If your business has more than three employees, it's worth building one template per scenario type (redundancy, pay freeze, policy change) now, before you need them.

For related workflows on building HR-adjacent processes without a full HR team, see how to use AI to write your employee handbook and how to use AI to handle performance review conversations.

---

FAQ

How do I tell an employee they are being let go without saying the wrong thing? Use the prompt structure in this post to build a spoken script before the meeting. The key rules: state the decision in the first two sentences, name the business reason specifically, confirm what happens next in concrete terms, and avoid any language that could be read as performance-related if the reason is redundancy. Rehearse the script against the objection list the AI generates — that's where most managers get caught out.

Can I use AI to write a redundancy script if I'm in the UK and have to follow the ACAS code? Yes, but you must tell the AI explicitly that you're in the UK and that the script needs to align with the ACAS code of practice on disciplinary and grievance procedures{:target="_blank"}. The AI will not automatically apply UK-specific requirements unless you specify jurisdiction. At minimum, a UK redundancy script needs to reference the right of appeal, the consultation process, and statutory notice — verify all three are present before you use the output.

Is the free version of ChatGPT, Claude, or Gemini good enough for this, or do I need a paid plan? The free tier is sufficient for this use case as of March 2026. ChatGPT-4o mini (free), Claude 3.5 Haiku (free tier), and Gemini 2.0 Flash (free) all produce usable scripts from a well-structured prompt. The paid plans — $20/month across all three — give you access to the more capable models (ChatGPT-4o, Claude 3.7 Sonnet) which tend to handle nuance and jurisdiction-specific language more reliably. If you're dealing with a legally sensitive situation, the $20 is worth it for that one session.

How do I make sure the AI isn't giving me legally incorrect advice? Treat AI output as a drafting tool, not a legal opinion. The script it produces is a starting point — a structure that ensures you hit the key communication checkpoints. For any situation involving potential wrongful termination claims, a redundancy affecting more than one employee, or anything where the employee has raised a prior grievance, spend £150–£300 on an hour with an employment solicitor (UK) or an HR attorney (US) to review the script and process before the conversation happens. That cost is negligible compared to the cost of an employment tribunal or wrongful termination lawsuit.

What's the cost of getting this wrong — is it really a legal risk or just an uncomfortable conversation? Both, and the financial exposure is real. In the UK, an employee with two or more years of service can bring an unfair dismissal claim — the average tribunal award in 2023–24 was approximately £14,000, with some awards significantly higher. In the US, wrongful termination settlements vary widely by state and circumstance, but even a straightforward case generates legal fees in the thousands before it resolves. Beyond legal cost, SHRM research{:target="_blank"} documents the trust damage to remaining staff when a departure is handled poorly — and replacement hiring costs for a lost employee typically run 50–200% of that employee's annual salary. A 20-minute prompting session is a low-cost hedge against all of that.

Should I share the AI-generated script with the employee or HR advisor before the meeting? Share it with an HR advisor or solicitor if the situation has legal complexity — that's the right use of the document. Do not share it with the employee in advance. The script is a preparation and rehearsal tool. Reading it aloud in the meeting, or worse, handing it to the employee to read, undermines the personal nature of the conversation and signals that you're not actually present in the room. Use it to prepare, then put it away.