Here's the uncomfortable truth about AI in communications: most people are using it wrong. Not because they're bad at their jobs - but because no one told them that the quality of your output is almost entirely determined by the quality of your input.
The people getting transformative results from AI aren't doing anything magical. They've simply learned to write prompts the way a senior editor briefs a junior writer. Specific. Contextual. With clear expectations for format, tone, and outcome.
This guide covers everything you need to do that - from the basic anatomy of a PR prompt to advanced chaining techniques used by agency teams handling high-volume work.
"The best prompt is a clear brief. If you can't brief a human, you can't brief an AI. The difference is AI won't ask follow-up questions - it'll just make things up."
Why Your Prompts Are Failing
Before learning what works, it's worth understanding what doesn't. Most bad AI outputs in PR work trace back to one of three root problems:
Problem 1: Under-specification
The AI doesn't know your brand, your audience, your preferred style, or the purpose of what you're creating. So it defaults to generic. That's not the AI being lazy - it's the AI doing exactly what you asked, with the information you provided.
Problem 2: Prompt as headline, not brief
A prompt like "write a press release about our new product launch" is not a brief. It's a headline. A good brief tells the AI: who the client is, who the audience is, what the key message is, what the tone should be, what NOT to say, and what format you need. Length matters. Facts matter. Stakes matter.
Problem 3: One-shot thinking
Communications is an iterative discipline. We draft, revise, cut, and layer. But most people treat AI like a vending machine - one prompt, one output, take it or leave it. The biggest productivity gains come from treating AI as a working session, not a request window.
The Anatomy of a Great PR Prompt
Every high-quality PR prompt contains some combination of these five components. You don't always need all five - but the more you include, the better the output.
3 Prompt Frameworks for Comms
Frameworks give your team a repeatable structure so prompts stay consistent across projects and people. Here are three that work specifically for PR and communications work.
Framework 1: RACE (Role, Action, Context, Examples)
Best for: drafting tasks where you need a specific voice or style.
You are a senior PR strategist at a B2B technology agency with 15 years of experience writing for enterprise software clients. Your action: Write a 350-word media pitch for [JOURNALIST NAME] at [PUBLICATION]. Context: Our client, [COMPANY], just launched [PRODUCT/ANNOUNCEMENT]. The key news hook is [HOOK]. The journalist covers [BEAT]. They recently wrote about [RELEVANT STORY - shows you've done research]. Use this example pitch structure as your model: [paste a sample pitch you like]. Tone: Direct, no fluff, journalist-friendly. No buzzwords. No "game-changing" or "revolutionary."
Tip: Replace bracketed placeholders with real specifics every time. Vague placeholders produce vague outputs.
Framework 2: CSI (Context, Scope, Iteration)
Best for: iterative creative work like messaging development or key message frameworks.
Context: I'm building a messaging framework for [COMPANY], a [BRIEF DESCRIPTION]. Their target audience is [AUDIENCE]. Their primary differentiator is [DIFFERENTIATOR]. Key competitors are [COMPETITOR A] and [COMPETITOR B]. Scope: Create a draft key messages document with: one master narrative (2–3 sentences), three supporting pillars, and two proof points per pillar. I'll give you feedback after each section. Start with just the master narrative - wait for my response before proceeding.
Tip: "Wait for my response before proceeding" is one of the most underused phrases in AI prompting. It keeps the AI in working session mode rather than dumping a wall of text.
Framework 3: PREP (Problem, Result, Example, Probe)
Best for: getting the AI to help you think, not just produce - useful for crisis prep, strategy, or briefing documents.
Problem: [COMPANY] is facing a potential media inquiry about [SENSITIVE TOPIC]. We haven't done anything wrong, but the optics are complicated because [NUANCE]. Result I need: A list of 10 tough questions a journalist might ask, followed by recommended holding statements for each. Example of our previous crisis voice: [paste a past statement or approved language]. Probe: After you complete the list, ask me two questions that would help you sharpen the holding statements further.
Tip: Asking the AI to probe you back with follow-up questions turns a static output into a working dialogue - significantly improving final quality.
Real-World PR Use Cases
Here's how to apply structured prompting to the most common communications tasks.
Writing a Press Release
You are a seasoned PR writer with deep experience in [INDUSTRY] media. Write an AP-style press release announcing [ANNOUNCEMENT]. Key facts to include: - [FACT 1: what happened] - [FACT 2: why it matters] - [FACT 3: quote from executive] - [FACT 4: company boilerplate - paste here] Audience: B2B technology journalists and trade press. Length: 400–500 words. Format: Standard press release format with headline, dateline, body, boilerplate, contact info. Tone: Professional, factual. No superlatives. No "excited to announce." Do not include phrases like "cutting-edge," "world-class," or "disruptive."
Preparing Spokesperson Talking Points
Create concise talking points for [EXECUTIVE NAME], [TITLE], for an upcoming media interview about [TOPIC/ANNOUNCEMENT]. Interview context: [PUBLICATION], estimated 20-minute conversation, journalist is [JOURNALIST NAME] who covers [BEAT]. Format: 5 key messages, each as a single sentence the executive can say naturally, followed by a 2–3 sentence expansion and one supporting data point or example. Voice: [EXECUTIVE NAME] speaks in plain language, avoids jargon, often uses analogies. Here's a sample quote from a previous interview: "[PASTE SAMPLE QUOTE]." Flag any messages where we don't have strong proof points so I can address those gaps.
Researching a Journalist Before Outreach
I'm about to pitch [JOURNALIST NAME] at [PUBLICATION] about [STORY ANGLE]. Based on the following recent articles they've written [paste headlines or summaries], help me: 1. Identify recurring themes or angles they gravitate toward 2. Note any coverage gaps our pitch could fill 3. Draft a one-sentence pitch hook that speaks directly to their demonstrated interests 4. Suggest what NOT to include based on their coverage patterns Be specific. This is a real pitch and I want to be respectful of their time.
Crisis Scenario Planning
You are a crisis communications strategist. I want to run a scenario stress test. Scenario: [COMPANY] is about to be mentioned in a [TYPE] story about [TOPIC]. The story is factually accurate but lacks context and may damage reputation. Step 1: List the 5 most likely ways this story could escalate in the first 48 hours. Step 2: For each escalation path, recommend one proactive action. Step 3: Draft a 150-word holding statement we can issue before the story publishes. Keep the tone calibrated to a brand that values transparency. Do not recommend "no comment."
Locking in Brand Voice
Generic AI output is a symptom of missing context, and nowhere is that more painful than brand voice. Here's how to make AI sound like your client - not like everyone else's client.
The Voice Snapshot Method
Before any drafting session, paste this block at the top of your prompt (or in a Claude Project / ChatGPT custom instruction):
BRAND VOICE REFERENCE - [COMPANY NAME] Tone: [3–5 adjectives. Be specific: "direct and credible, never hype-driven" beats "professional."] Audience: [Who they are and what they care about] We always: [List 3–4 things the brand always does. E.g., "lead with data," "acknowledge complexity," "use plain language"] We never: [List 3–4 prohibitions. E.g., "use passive voice," "make claims we can't substantiate," "use competitor names"] Vocabulary: [Brand-specific terms, preferred phrases, words to avoid] Example approved copy: [Paste 1–2 paragraphs from approved content] Apply this voice to everything you produce in this conversation.
Tip: Save this as a template and fill it in for each client. The 10 minutes you spend building it saves hours of editing.
Using Claude Projects for Persistent Voice
If you use Claude, the Projects feature lets you set instructions that persist across every conversation - no copy-pasting required. Add your brand voice snapshot to the project instructions, upload approved assets, and every session starts with full context already loaded.
The 7 Most Common Prompt Mistakes in PR
"Write a media pitch for our new app."
Specify the journalist, publication, hook, company context, and format. Treat it like a real brief.
Every AI defaults to "professional." This tells it nothing about your brand's specific voice.
"Speaks like a confident practitioner - credible, direct, never breathless. Uses short sentences. No superlatives."
The first draft is the AI's best guess. It's the starting point, not the deliverable.
"The second paragraph is too vague. Tighten it and add a specific data point. Preserve everything else."
AI will fill the space it thinks is appropriate - often too long, sometimes too short.
"300–350 words, no more." Word count guidance is one of the most reliable format signals.
This is undirected. AI doesn't know your definition of better.
"The opening is too slow. Rewrite it to lead with the news hook in the first sentence."
AI will speculate, estimate, and generalize if you don't set boundaries.
"Only use facts I've provided. If you don't have enough information, ask me rather than making assumptions."
You lose accumulated context - the AI forgets everything you established.
For client work, keep context in a Project. For campaign work, keep everything in one thread.
Advanced: Chaining Prompts
Prompt chaining is the practice of breaking complex deliverables into a sequence of focused prompts, where the output of each step becomes the input for the next. It's how experienced AI users build high-quality, complex documents without losing coherence or control.
Example: Building a Full Campaign Brief
A 4-Step Chain for a Campaign Messaging Brief
Step 1: Audience Analysis. "Based on [CLIENT DESCRIPTION] and their target audience [DESCRIPTION], what are the top 3 audience insights that should drive our messaging strategy? Provide evidence-based reasoning."
Step 2: Message Platform. "Using those audience insights, draft a message platform with one master narrative and three supporting pillars. Format as a brief document with headers."
Step 3: Message Testing. "For each pillar, identify the strongest potential objection a skeptical audience member might have, and a one-sentence response to that objection."
Step 4: Executive Summary. "Synthesize everything from our conversation into a one-page executive summary that a CMO can present to their leadership team. 250 words max."
Each step builds on the last. You're not asking AI to do everything at once - you're guiding it through a professional workflow.
System Prompts & Projects
If you're doing sustained client work, the highest-leverage thing you can do is set up a system prompt or project that holds your context permanently. This means you never lose time re-briefing the AI every session.
What to include in a standing system prompt:
- Brand voice snapshot (tone, audience, dos and don'ts)
- Client background (who they are, what they do, key differentiators)
- Ongoing campaign context (what's live, what's upcoming)
- Approved terminology and brand vocabulary
- Formatting preferences (how you like outputs structured)
- A note on the working relationship ("Treat me as a senior comms strategist. Push back if a request conflicts with communications best practices.")
In Claude, use the Projects feature to store this context. In ChatGPT, use custom instructions or memory. In Gemini, use Gems. Each platform has a way to hold persistent context - use it.
Ready to audit how your brand shows up in AI answers?
Take the AI Visibility Assessment →Frequently Asked Questions
Prompt engineering for PR is the practice of crafting precise, structured inputs to AI tools like ChatGPT and Claude so they produce outputs that are actually useful for communications work - on-brand, accurate, and ready to use with minimal editing.
Context is the most important element. AI models can't read your mind. The more relevant context you give - your audience, your brand voice, the desired output format, what NOT to do - the better the result. Most bad outputs are caused by under-specified prompts.
Both are excellent. Claude tends to produce longer, more nuanced drafts with stronger editorial judgment - ideal for strategic memos, media pitches, and crisis drafts. ChatGPT is highly versatile and excels at structured formatting and rapid iteration. Many PR pros use both depending on the task.
The solution is specificity and voice modeling. Give the AI examples of your writing, tell it who the audience is, describe your brand's tone in concrete terms (e.g., "direct and credible, never hype-driven"), and always ask it to avoid vague superlatives. Using a system prompt or project instructions to hold brand voice permanently is a game-changer.
Related Guides
How to Cowork with Claude
Build your fractional AI team with Projects and persistent context
The PR Pro's Guide to Claude Code
Set up and use Claude Code for AI visibility workflows
Vibecoding for Comms
From using AI tools to building your own - what you can ship in a weekend
AI and Communications
What's actually working in 2026 - the full practitioner's guide