AI for Communications Teams

03 — Resources

Comms teams are under unusual pressure: more channels, faster turnarounds, smaller budgets, and a public that's increasingly skeptical of polished prose. AI can buy back real hours — but every shortcut in communications carries reputational risk. This guide is about using AI for the work where speed matters, while keeping the human judgment on the work where trust matters.

"AI drafts. Humans personalize, contextualize, and decide."

A — Where AI helps and where it doesn't

Sort the work into two buckets.

Not all communications work belongs in AI. The teams that get this right draw the line clearly — and put the line in writing before anyone needs it.

High AI leverage

— First drafts of internal memos and announcements
— Social media variants (writing six posts from one story)
— SEO content and metadata
— Meeting recaps and stakeholder summaries
— Audience persona development for messaging
— Tone variations of the same message (formal, casual, executive)
— Editorial calendar brainstorming

Human-first work

— Crisis communications and public statements
— Executive remarks and personal letters from leadership
— Sensitive employee communications (layoffs, restructuring, deaths)
— Regulatory or legal disclosures
— Public-facing data and statistics (without verification)
— Anything quoted in your name

B — The four pillars

The standards a comms team has to hold.

1. Brand voice protection

AI defaults to a neutral, polished, slightly-too-formal voice. If your brand voice is anything else (warm, irreverent, technical, plainspoken), every AI draft drifts toward the mean. Counter it by uploading examples of your actual voice and writing custom GPT instructions that include "do not" rules: don't use corporate jargon, don't add throat-clearing intros, don't soften strong language.

2. Disclosure and transparency

Decide your default position on AI disclosure before you need it. Some channels (regulatory, academic, legal) require disclosure. Some audiences (journalists, funders, board members) will ask. Most won't — but if they do, having a written policy beats improvising on the call. When in doubt, the path that holds up is "AI-assisted, human-reviewed and approved."

3. Hallucination defense

AI fabricates statistics, quotes, citations, and historical claims confidently. In communications, a single fake stat in a press release is a reputational hit you can't take back. Every external-facing piece needs a named verifier. See the verifying AI output guide for the protocol.

4. Crisis discipline

The temptation in a crisis is to use AI to draft fast. The risk is using AI to draft without thinking. Use AI for tone variants and length compression once the message and judgment are settled — never for the message itself. The first draft of a crisis statement is always human.

C — The prompts that actually pay off

Worth keeping in a comms team library.

Brand voice training

You are our communications assistant. Our brand voice is [describe in 3-5 adjectives, e.g., warm, direct, plainspoken, occasionally wry]. Avoid: corporate jargon, opening with "In today's fast-paced world," throat-clearing phrases, exclamation points except in social posts. Match the tone of the three example pieces I've uploaded. When in doubt, write shorter and more concrete.

Tone variants from one source

Take the attached announcement and produce three versions: (1) a 60-word internal Slack post, (2) a 150-word company blog post, (3) a 40-word LinkedIn post. Maintain our brand voice and the same core message. Do not invent facts or numbers — use only what's in the source.

Social variants from one story

Take this 800-word feature story and produce six social posts: three for LinkedIn (professional audience), two for Instagram (broader audience), one for X (concise, no hashtags). Each should pull a different angle from the story — no repetition. Include a clear call to action.

Audience persona development

Develop three audience personas for our [product/service/cause], based on [industry/region/segment]. Each persona should include: role, daily pain points, content preferences (channels, length, tone), what would make them act, and what would make them disengage. Use these to inform our editorial calendar.

Editorial calendar brainstorm

Generate 20 content ideas for a [industry/role] audience for the next quarter. Mix evergreen topics, timely angles tied to [upcoming events/seasons], and customer story formats. Group them by audience persona and assign a primary channel for each.

SEO content rewrite

Rewrite this webpage with SEO best practices while keeping our brand voice. Use clear H2/H3 structure, concise paragraphs, and natural keyword integration around [target term]. Do not stuff keywords. Do not change any facts or claims. Flag any sentences where you weren't sure of the meaning.

Press release first draft (NOT crisis)

Draft a press release announcing [event/launch/partnership] using only the facts in the attached briefing. Lead with the most newsworthy element. Include a quote placeholder from [executive name] — do not invent the quote. Keep it under 400 words. Output should be ready for editorial review, not final.

Stakeholder summary

Summarize the attached meeting transcript or document into a one-page brief for [audience: board, executive team, frontline managers]. Lead with three takeaways, three decisions required, and three risks raised. Match the audience's vocabulary — more strategic for board, more operational for managers.

D — A crisis playbook

When the situation moves faster than usual.

Crisis communications is the highest-risk place AI can help — and the highest-risk place to use it wrong. A working sequence:

Step 1: Human-only draft. The core message, what you're acknowledging, what you're committing to, what you're not saying yet. AI does not write this.

Step 2: AI for tone variants. Once the message is settled, use AI to produce variants for different channels (internal, customer, media, social) and lengths. The message is locked; only the wrapper changes.

Step 3: AI for stakeholder briefings. Draft a one-pager for the executive team, a Q&A document for spokespeople, and a holding statement for any channel you can't update in real time. AI is fine here.

Step 4: Human review on every external piece. No AI-assisted statement goes out without legal or executive sign-off. The speed AI buys is meaningless if the statement makes the crisis worse.

Step 5: After-action review. Within 72 hours, document what AI helped with, what it didn't, and what you'd do differently. Most teams skip this and lose the learning.

E — Common failure modes

Where comms teams get burned.

Voice drift. Six months in, every piece sounds slightly the same — generic, polished, mid-Atlantic corporate. Counter: re-upload voice examples quarterly and audit recent output against original tone.
Hallucinated quotes. AI invents quotes that sound like an executive said them. They didn't. Counter: a hard rule that AI never generates a named quote — only placeholders.
Crisis speed trap. AI-drafted crisis statements that were "good enough" until a journalist parsed them word by word. Counter: human-first crisis drafting, every time.
Disclosure ambush. A reporter asks "did you use AI to write this?" and the team improvises. Counter: a written disclosure default approved by leadership before anyone asks.
Statistic laundering. AI generates a stat, an intern includes it in a deck, the deck becomes a press release, the press release lands in a board update — and no one ever verified the original number. Counter: every stat traceable to a primary source, named in the doc.

F — Pair with

Resources that strengthen this one.

Spot the Fake: Verifying AI Output — the verification protocol every external-facing communication needs.

AI Policy Starter Kit — defines the rules of the road that comms decisions sit inside.

AI Prompt Framework — the nine-step prompt structure underlying every prompt on this page.

← Back to Resources