Internal announcements and change comms · AI for HR & Recruiters

There is a quiet asymmetry in internal communication. The announcements that matter least to the company — the holiday calendar, the new sambandhit-vibhag (related department) liaison, the change in lunch vendor — are also the ones the company is most likely to communicate well, because the writer is relaxed. The announcements that matter most — a restructure, a redundancy round, a leadership change, a return-to-office mandate — are the ones the company is most likely to communicate badly, because the writer is anxious, late, and reaching for a template. AI offers a tempting shortcut here: it produces fluent, calm-sounding text in any tone you ask for, in seconds. The temptation is exactly the risk. The very speed that helps you on the lunch-vendor email is what tempts you to skip the human work the redundancy email actually requires.

The standard announcement workflow

For the routine majority of internal comms — policy updates, process changes, new hires, system roll-outs, office logistics — AI is straightforwardly useful. The shape that works is the one you have seen elsewhere in this course: structured fact block in, structured draft out, light human edit, send.

A working pattern for an announcement prompt:

Draft an internal announcement email to staff. Use only the facts below. Do not invent benefits, dates, or commitments not provided.

What is changing: the office hours for the Lalitpur branch are moving from 9:30 to 6:00 to 10:00 to 6:30, effective Shrawan 1, 2083. When: announced Asar 15, effective Shrawan 1, 2083. Why: the morning rush to Patan Dhoka now reliably adds 25-30 minutes to commute time; the change shifts staff out of the worst window. Who is affected: all 42 Lalitpur-branch staff. Kathmandu and Pokhara offices unchanged. Next steps: line managers will confirm receipt with each team member by Asar 25; any genuine constraint on the new timing is to be raised with HR by Asar 22. Contact: Sangita (HR Lead), sangita@firm.com.np, 98XXXXXXXX. Tone: clear, warm, no corporate language. Bilingual — English first, Nepali below, both saying the same thing.

The output is something a manager spends three minutes reviewing rather than thirty minutes writing. The “what / when / why / who / next steps / contact” structure is what every routine announcement actually needs; ask the model for it and it delivers.

Change comms — where AI must stop being the voice

The category called “change comms” — restructures, redundancies, leadership transitions, return-to-office mandates, mergers, acquisitions, branch closures, salary freezes — is a different category, not just a more sensitive version of the same thing. In change comms the people receiving the message are absorbing something that affects their livelihood, their identity at work, or both. They will read the message many times. They will forward it. They will quote it back to you, sometimes years later. And they will know, with surprising accuracy, whether it came from a human leader who had thought hard about what they were going to say, or from a model.

The rule that holds is this: AI may produce the structural draft. The empathic dimensions — the why, the acknowledgement, the commitment, the line that signals “the leadership has actually thought about you” — must come from a human leader in their own voice. This is not a stylistic preference. It is a trust contract. A redundancy announcement that reads like generic management prose, even if it is technically accurate and procedurally correct, tells the affected staff that the company did not value them enough to write to them as individuals. The lasting damage is not in the redundancy; it is in the message.

The second rule, and the easier one to break: never let the model invent commitments the firm has not actually made. When you ask a chatbot to write “a compassionate redundancy announcement”, it will helpfully add lines like “we are committed to supporting affected staff through their transition with dedicated outplacement services and extended healthcare coverage”. If the firm has agreed those things, fine. If it has not, the model has just made commitments on the firm’s behalf that staff will hold the firm to — and reasonably so, because the message went out under the firm’s name. Strip every line that promises something the firm has not actually decided.

The no-surprises rule

There is one rule in change comms that predates AI by decades and that AI makes both more important and easier to break: the people most affected should hear about something material before the all-hands email lands in their inbox. A staff member learning they have been made redundant from the same email that announces the redundancy to two hundred colleagues will not forgive the firm. A team learning their manager has resigned from the company-wide mail rather than from the manager themselves will tell that story for years.

AI’s quiet danger here is that it removes the friction that used to enforce this rule. When writing a hard email took two hours, the friction itself made the leader stop and ask “have I had the one-on-one conversations first?” When AI can produce a polished all-hands email in eight minutes, the temptation to just send it — and to deal with the individual conversations afterwards, or in parallel, or via the email itself — becomes much stronger. Resist it.

The discipline is operational. Before any change-comms email goes out, write down the names of the people most directly affected, and confirm each of them has heard about the change in a one-on-one or small-group conversation with the right person — usually their line manager, sometimes a founder, occasionally an HR partner. Only then does the announcement go. The list is sometimes uncomfortably long; do not skip it. AI does not change this rule; it makes the rule harder to honour by making the shortcut easier to take.

The bilingual question

For any internal announcement that materially affects staff, the practical default at firms with mixed-language staff is to send the message in both English and Nepali, with the same content in both, in the same email. The reasons are unromantic: an English-only mail to a 60-person firm where 30 staff are most comfortable in Nepali means 30 staff are reading a translated, partial, anxious version that they have asked the colleague next to them to interpret. A Nepali-only mail to a firm with international partners or a non-Nepali CFO means the international staff are excluded from material information.

The model is genuinely useful here. Drafting the message once in English, then asking the model to produce a Nepali version that says exactly the same thing — same commitments, same dates, same names, same numbers — is faster and more consistent than two parallel drafts. The discipline is to have a fluent Nepali reader on the team check the translated version word by word for two specific failures: numbers and names mis-converted (Nepali digits and English digits drifting), and commitments quietly softened or hardened by translation choice. For routine announcements a fifteen-minute review is enough; for change comms, the bilingual review is part of the no-surprises checklist, not separate from it.

Tone register by audience — never leave it unspecified

Models default to a register that sounds vaguely like an American HR business partner — earnest, warm-ish, slightly corporate, full of phrases like “please reach out” and “we appreciate your patience”. This register is wrong for almost every Nepali context. It reads as imported, distant, and translated even when the message is in English. It reads as actively strange when rendered into Nepali, where the equivalent phrasing has a corporate-pamphlet feel that no one in the office actually uses.

The discipline is one line in the prompt and one moment of attention in review. The prompt line: “Tone: direct, warm, conversational. The way a respected line manager would talk to their team in this office. Avoid corporate phrases like ‘please reach out’, ‘we appreciate’, ‘circle back’, ‘going forward’. Use the contractions and informal connectives a real person uses.” The review moment: read the draft out loud as if you were the manager delivering it in person. If you would not say it that way in a meeting, change it. The model produces what it produces; the firm is the one whose voice is on the page.

A second register lever: who is the announcement from? An email from “the HR Team” sets a very different expectation than an email from a named founder or CEO. For routine logistics, “from HR” is correct. For change comms, the message is from a specific human, signed by name, and the tone must match that human’s actual voice. The model can imitate any voice you tell it to; it cannot guess. Tell it.

Check your understanding

Quick check

—

Your company is restructuring three teams into two and making four staff positions redundant. Leadership has asked you to draft the all-hands announcement using AI to save time. Which workflow best preserves trust in the company?

Ask the chatbot to “write a compassionate redundancy announcement”, review the draft for tone, and send it to all staff once leadership signs off — the four affected staff will receive the message at the same time as everyone else.
Use AI to produce the structural draft (what is changing, when, the procedural facts, who to contact), have a human leader rewrite the empathic and acknowledgement sections in their own voice, strip any commitments the model invented that the firm has not actually agreed to, confirm each of the four affected staff has had a one-on-one conversation with their line manager before the email goes out, and have a fluent Nepali reader check the bilingual version line by line.
Skip the AI entirely and have a senior leader write the whole message from scratch the night before sending it.
Use AI to draft both the all-hands email and the individual notifications to the four affected staff, sent simultaneously so the messaging is consistent.

What comes next

A clean policy and a well-handled announcement are the visible surface of HR communication. The less visible — and arguably more leveraged — surface is what happens in the first weeks of an employee’s life with the company: the training materials they read, the onboarding checklist they tick through, the FAQ they consult at 4pm on day three when they cannot remember which approval workflow to use. The next section is about training and onboarding materials — where AI is a real multiplier, where it absolutely cannot replace the human trainer, and how to build materials that actually answer the questions a new hire has on day one.