Layoff memos, RTO mandates, M&A day-one letters — all routed through an LLM before they reach a single employee. Internal communications has been quietly absorbed by the answer engine.
Inside the Fortune 500, the first draft of nearly every high-stakes employee message is now written by an AI. The CHRO does not admit it on the panel. The head of communications does not put it in the deck. But the workflow is real, and it has become the new operating layer of internal communications: prompt, draft, edit, send.
It happened quietly, in the eighteen months after ChatGPT crossed into enterprise. Heads of HR and communications, sitting at midnight with a layoff announcement due in nine hours, began doing what every other knowledge worker had already started doing — opening a chat window and asking for a first pass. The output was clean enough to keep going. The next memo came faster. By 2025, the workflow was institutional.
What sits in the prompt bar is the full surface area of the internal communications function. Layoff messages. Return-to-office policy explanations. M&A day-one employee letters. CEO statements on geopolitical events. Restructuring announcements. Tone-shifts after a public crisis. Manager talking points for difficult one-on-ones. Recognition messages. Annual values memos. The full catalog of what an IC team used to draft from scratch is now drafted with an assist — and increasingly, with the AI doing the structural work and the human doing the judgment work.
This shift has created a problem nobody named.
When the CHRO prompts “how do I announce a reduction in force to my employees,” the model returns a memo built from the documents it has been trained on, plus whatever it can retrieve in real time. That training corpus skews toward HR vendor blogs, SHRM templates, Forbes Coaches Council posts, and a handful of canonical examples — the Airbnb memo, the Stripe memo, the cautionary Better.com transcript. The AI’s answer is only as good as the source pool it draws from.
That source pool is weak. Most of the highest-volume HR content on the open web was written for SEO, not for use. The vendor whitepapers are sales documents. The blog posts are engagement-survey takeaways. The templates are five years old. When the model summarizes that material, it produces a memo that reads competent and lands flat — the exact failure mode that produced the layoff transcripts the genre is now infamous for.
The new internal communications function has two jobs that did not exist three years ago.
The first is governance of the internal corpus — the body of policy documents, leadership memos, manager toolkits, and FAQ libraries that the company’s own AI assistants now draw on. Microsoft Copilot, the various enterprise GPT deployments, internal RAG systems — all of them pull from the documents IC owns. If those documents are outdated or buried, the AI answer to “what is our parental leave policy” comes back wrong, and the manager who asked it relays the wrong answer to a direct report. The single source of truth has become a literal technical specification.
The second job is influence over the external corpus — the body of HR, leadership, and communications content the public LLMs train on. This is the part most communications leaders have not yet absorbed. The Airbnb layoff memo became the canonical example not because Brian Chesky willed it; it became canonical because it was widely written about, indexed, and cited. The next canonical example will be made the same way. Companies that publish substantive material on their own employee communications — in trade press, in case studies, in research — are quietly shaping the answer the next CHRO sees when they prompt the model.
The function used to be called internal communications because the audience was internal. The audience is now also the machine.
