How to Use AI for Internal Communications (Without Sounding Like a Corporate Memo Generator)

Why Most Comms Teams Get AI Backwards

A comms lead I worked with last spring had four browser tabs open, each one a slightly different version of the same reorg announcement: one for engineering, one for sales, one for the leadership talking points, one for the all-hands slide. She’d written the core message once, in about twenty minutes, and then spent three hours manually re-tailoring it for each audience. By 6pm she was still stuck rewriting the sales version because the tone kept coming out wrong.

That’s the actual bottleneck in internal communications, and it’s almost never the first draft. Most people’s instinct when they hear “use AI for comms” is to have it write the announcement from scratch. That’s backwards. The core message, what’s actually changing and why, needs a human who understands the politics, the history, and the specific anxieties of the room. What eats the afternoon is everything downstream of that: reformatting the same message for email, Slack, a manager cheat-sheet, and an FAQ, four times, in four slightly different voices.

I train HR and comms professionals on this distinction constantly, because it’s the difference between AI saving you real hours and AI producing four flatter, more generic versions of something that was fine to begin with.

The Readiness Gap: What the 2026 Data Actually Shows

Gallagher’s State of the Sector 2026 report, drawn from over 1,300 comms and HR professionals across 40 countries, names this directly: a “Readiness Gap” between what organizations need from internal communications and what their teams are actually resourced to deliver^[1]. On AI specifically, the number is stark: 75% of internal-communications functions are still stuck in early-stage discussions or ad-hoc experimentation, not a real strategy^[1].

The report’s own framing is worth sitting with: AI doesn’t create maturity, it amplifies whatever maturity already exists^[1]. A comms function with a clear voice, a real audience segmentation, and a defined approval process will get faster and better with AI. A comms function that’s already inconsistent and reactive will just produce inconsistency faster.

Meanwhile the cost of not fixing this is measurable. More than a quarter of employees, and 38% of managers, report feeling overwhelmed by excessive communication, and employees reporting high information overload are 52% less likely to intend to stay at the company^[1]. AI used well reduces that overload by making messages sharper and better targeted. AI used badly, as a volume generator, makes it dramatically worse.

The One Workflow Worth Building: One Message, Many Formats

Here’s the practical fix for the four-tabs problem. Write the core announcement yourself, in your own voice, the way you always have. Then hand that single draft to AI with a specific instruction covering every format you actually need, in one prompt, not four separate ones.

A prompt that actually saves the afternoon

Something like: ‘Here is our internal announcement about [the change]. Rewrite it into: (1) a 3-sentence Slack version for the #company-updates channel, casual tone, (2) a one-page FAQ anticipating the five questions employees will actually ask, (3) a five-bullet talking-points sheet for people managers to use in team meetings. Keep every fact identical to the source. Do not add new claims or soften anything I said.’

That last instruction matters more than it looks. Left unconstrained, AI models will often smooth over genuinely uncomfortable details in a reorg or layoff announcement, because “softer” language is what most of their training data rewards. For internal comms specifically, that’s a real risk: employees can tell when a message has been sanded down, and it reads as evasive rather than kind.

Check every version against the source before it goes out

This is the step people skip when they’re in a hurry, and it’s the one that matters most. Read the Slack version and the FAQ version side by side with the original. If a number, a date, or a scope detail drifted between versions, that’s the AI filling a gap with something plausible rather than something true. Fix it before it’s in four places instead of one.

Using AI for the Hard Ones: Change and Policy Announcements

Reorgs, layoffs, benefits changes, and policy updates are where the stakes are highest and where AI is most tempting to lean on, because these are exactly the messages people dread writing. They’re also where AI’s tendency to default to generic, reassuring corporate language does the most damage.

The workaround isn’t to avoid AI for these messages. It’s to use it only for the mechanical parts: formatting a timeline, structuring an FAQ around questions HR has actually fielded before, or translating a finished, human-approved message into a second or third language for a global team. Never let AI originate the framing or the reasoning behind a change that affects people’s jobs or pay. That part has to come from someone accountable for it.

• Use AI for: reformatting an approved message across channels, drafting an FAQ from a list of real anticipated questions, translating a finished announcement, generating a manager talking-points version.
• Don’t use AI for: deciding what to say about a layoff or reorg, softening language that was intentionally direct, inventing reassurances that weren’t part of the original decision.

If your certifying body or company has a defined change-communication playbook, that’s your accuracy anchor, the same way a professional body’s own materials anchor exam prep. Feed AI your own approved templates and past examples rather than asking it to generate change-comms language from its general training, which tends to default to vague, corporate-safe phrasing that erodes trust exactly when trust matters most.

A Worked Example: Turning a Benefits Change Into Three Formats

Here’s what the one-message-many-formats workflow actually looks like in practice, using a real category of announcement: a change to health insurance premiums that takes effect at open enrollment. The comms lead writes the core message herself: three short paragraphs explaining what’s changing, why (a genuine cost increase from the carrier, not a euphemism), and what employees need to do before the enrollment deadline.

That draft goes into AI with the multi-format prompt from the section above. The Slack version comes back as three sentences: what’s changing, the deadline, and a link to the full FAQ, casual enough to fit the channel without losing the actual facts. The FAQ version expands into eight questions, pulled from what HR actually fielded during last year’s enrollment period (how much will my specific plan go up, does this affect my HSA contribution limit, what happens if I miss the deadline), each answered in two or three sentences grounded in the source draft.

The manager talking-points version is the one that matters most and gets skipped most often. It’s a five-bullet cheat sheet a people manager can glance at thirty seconds before a team meeting: the headline fact, the one likely pushback question, and an honest, non-scripted way to say “I don’t have the answer to that, let me find out” for anything outside the FAQ. Managers who get caught flat-footed on a benefits question in front of their team lose credibility fast, and a decent talking-points sheet is cheap insurance against that.

Total time for all three formats, once the core message is written: about ten minutes of AI drafting plus fifteen minutes of a human checking every fact against the source. Compare that to the comms lead in the opening example, still stuck on her second manual rewrite at 6pm, and the actual value of this workflow becomes obvious. It’s not that AI writes better change communications. It’s that it removes the multiplication tax on a message a human already got right the first time.

What the Internal Comms Tools Actually Do Differently

General AI models are the right tool if your team is small and your real bottleneck is drafting time. Purpose-built platforms earn their cost once you need to know whether people actually opened and read what you sent, which a general AI tool simply can’t tell you. Staffbase’s own 2026 write-up on AI in internal comms describes the current wave of tools as increasingly combining targeting, analytics, and AI generation into one workflow rather than treating them as separate steps^[3], which tracks with where the market is clearly headed.

If you’re a lean team without budget for a dedicated platform, the ChatGPT-plus-manual-workflow approach above gets you most of the drafting benefit for free. You just lose the analytics, so you won’t know if the reformatting actually helped until you ask people directly.

Where AI Backfires on Internal Comms

Here’s my honest skepticism. Over-personalization is the fastest way to make an internal message feel creepy instead of thoughtful. An AI-generated email that name-drops an employee’s tenure, team, and a recent project in the first line doesn’t read as “they really see me,” it reads as “this was generated,” especially once employees clock the pattern across a few messages. Personalize the content (which FAQ questions apply to their team, which policy section actually affects them), not the performance of warmth.

The second failure mode is losing institutional memory. A comms team that leans on AI to draft every announcement from a generic prompt, instead of from their own past messages and established voice, will drift toward sounding like every other company’s internal comms: competent, forgettable, and slightly hollow. The fix is boring but effective: build a running library of your best past announcements and always ground new AI drafts in that library, not in the model’s general training.

None of this is an argument against using AI here. It’s an argument for using it on the eighty percent of the work that’s genuinely mechanical, reformatting, translating, structuring, and keeping a human firmly in charge of the twenty percent that actually requires judgment about what to say and how direct to be.

One more practical safeguard worth building in: have a second person, not the one who drafted the AI prompt, read every reformatted version against the source message before it goes out, specifically checking for drift in tone or fact. It takes five minutes and catches the kind of quiet inconsistency that’s easy to miss when you’ve been staring at the same message across four different formats for an hour.

Sana Mian, Co-Founder of Future Factors AI

Sana is an AI educator and learning designer specialising in making complex ideas stick for non-technical professionals. She has trained 2,000+ learners across corporate teams, bootcamps, and keynote stages. Future Factors offers AI Bootcamps, Corporate Workshops, and Speaking & Consulting for businesses ready to adopt AI without the overwhelm.

More about Sana →