TL;DR

AI is a useful drafting tool for press releases — but only if you prompt it properly and know what to fix afterwards. Here are 10 prompts I actually use, with honest notes on where the human editing starts.

Most people prompt badly and blame the tool

I use AI to draft press releases most weeks. It's good at structure, decent at tone, and fast. But "write me a press release about our new product" produces exactly the kind of generic, breathless copy that gives AI-assisted writing a bad name.

The difference between a useless AI draft and a useful one is almost entirely in the prompt. Good prompts give the model constraints, context, and a clear job to do. Bad prompts give it a blank canvas and hope for the best.

What follows are 10 prompts I've refined through actual use — drafting comms for tech companies, refining them, watching where the AI nails it and where it falls over. They work on Claude, ChatGPT, Gemini. The model matters less than the input.

One caveat before we start. These prompts get you to about 80% of a finished press release. The remaining 20% — voice, angle, newsworthiness judgment, the stuff that actually determines whether a journalist reads past the first line — is human work. If you're looking for a way to eliminate that human layer entirely, this isn't the post for you. Try working with someone who does this professionally instead.

From the field

AI was used to draft a product announcement for a robotics client recently. The first draft was structurally sound — had all the right sections, reasonable flow. But the quotes sounded like they had been written by a committee, the headline was painfully generic, and the opening paragraph buried the actual news under three lines of company background. Twenty-five minutes of editing turned it into something publishable. That is the real workflow: AI drafts, human edits. Not AI publishes.

The prompts

1. The Context-Setting Prompt

This isn't a press release prompt — it's the prompt before the prompt. Skip this step and everything downstream suffers. You're giving the AI the background it needs to write intelligently about your company.

You are a senior PR professional writing for a UK-based tech company. Here is the context you need before I ask you to draft anything: Company: [name, one-line description, stage, funding] Industry: [sector and sub-sector] Target media: [e.g. Sifted, TechCrunch, Financial Times — be specific] Tone of voice: [e.g. confident but understated, technical but accessible] Key differentiator: [what makes this company different from the obvious competitors] Recent context: [anything a journalist covering this space would already know] Confirm you understand this context before proceeding.

Why it works: it stops the AI from guessing. Most bad press release drafts fail because the model is filling in blanks with generic assumptions.

Watch out: AI will cheerfully confirm it "understands" even if the context is thin. Be specific. The more concrete detail you provide here, the less editing you do later.

2. The Newsworthiness Test

Before you write anything, find out if you should be writing it at all. This is the prompt most people skip — and it saves the most wasted effort.

Here's our announcement: [paste the core news] Rate its newsworthiness from 1-10 for each of these audiences: - A journalist at [specific publication, e.g. Sifted] - A potential enterprise customer in [sector] - An investor scanning deal flow in [space] For each rating, explain specifically why they would or wouldn't care. Be brutally honest. If this isn't newsworthy, tell me what would make it newsworthy.

Why it works: AI is surprisingly good at pattern-matching against what publications actually cover. Not perfect — but it catches the obvious "this isn't news" situations before you waste an afternoon.

Watch out: AI tends to be generous with ratings. If it gives you a 6, that probably means 4 in reality. Calibrate accordingly.

3. The Angle Finder

Here is our announcement: [paste details] Generate 5 different angles for this story, ranked by newsworthiness. For each angle: - A one-sentence pitch as you'd say it to a journalist - Which specific publication or beat it's best suited to - What makes it timely right now Do NOT include generic angles like "company announces growth." Every angle should pass the test: would a busy journalist stop scrolling for this?

Why it works: it forces the AI to think beyond the obvious framing. The constraint against generic angles actually produces better output — the model tries harder when you tell it not to be lazy.

Watch out: angle 5 is usually the best one. The AI front-loads safe options and gets more creative further down the list. Read from the bottom up.

4. The Headline Generator

Write 10 press release headlines for this announcement: [paste core news] Rules: - No headline should exceed 12 words - At least 3 should lead with the impact or outcome, not the company name - At least 2 should be specific enough to work as a news headline (think Reuters or Bloomberg style) - None should use the words "excited", "thrilled", "delighted", "revolutionise", or "game-changing" - Include one contrarian or surprising angle

Why it works: the constraints are doing all the work. Without them, you get ten variations of "[Company] Announces [Thing] to [Vague Benefit]."

Watch out: AI headlines tend to be technically correct but editorially flat. Pick the best structure, then rewrite the actual words yourself.

5. The Quote Writer

AI-generated quotes are almost always the weakest part of an AI press release. This prompt helps — but you'll still need to edit.

Write a quote for [name, title] to include in a press release about [topic]. The quote should: - Sound like something a real person would actually say out loud - Make ONE specific point, not three vague ones - Be no longer than 3 sentences - Include at least one concrete detail (a number, a name, a specific outcome) - NOT use any of these phrases: "we're excited to", "this partnership will", "we look forward to", "we believe that" For reference, this person's communication style is: [e.g. direct and technical / warm and visionary / understated and precise]

Why it works: the banned-phrases list is essential. Left to its own devices, AI writes quotes that could come from any CEO at any company about any announcement. The style reference helps it approximate a real voice.

Watch out: even with this prompt, the quote will need rewriting. Read it aloud. If it sounds like a press release, it's not a good quote. The best test: would this person actually say this in a conversation?

6. The "So What" Rewriter

Rewrite this press release so that the first paragraph answers "so what?" for a [journalist covering enterprise tech / Series B investor / potential customer in healthcare]. The reader should understand why this matters to them within the first two sentences. Do not start with the company name or founding date. Start with the problem or the impact. Here's the current draft: [paste draft]

Why it works: most press releases bury the news. This prompt forces the AI to lead with relevance. Running it three times — once per audience — often reveals which angle is strongest.

Watch out: the AI sometimes overcorrects, making the opening so audience-specific that it loses the actual news. You need both: relevance and information.

7. The Boilerplate Refresher

Here's our current company boilerplate: [paste] Rewrite it in 3 versions: 1. Factual and neutral (for broadsheet press releases) 2. Confident and forward-looking (for industry/trade media) 3. Short and punchy (50 words max, for email pitches) Each version should include: what the company does, who it serves, one proof point (funding, customers, traction), and where it's headquartered. No superlatives. No "leading provider of."

I use this about once a quarter. Boilerplates go stale and nobody notices until a journalist quotes one that's two funding rounds out of date.

Watch out: AI boilerplates are competent but forgettable. The best ones have a point of view. Add something that only your company would say.

8. The Competitor Differentiator

Here are our three main competitors and their positioning: 1. [Competitor A]: [their one-line positioning] 2. [Competitor B]: [their one-line positioning] 3. [Competitor C]: [their one-line positioning] Here's our press release draft: [paste] Rewrite it so that a reader who knows our competitors would immediately understand what makes us different. Do not mention competitors by name. The differentiation should be implicit — in the framing, the claims, the proof points. Show, don't tell.

Why it works: PR people think about competitive positioning all the time. Most founders don't — or at least, they don't translate it into their external comms. This prompt bridges that gap.

Watch out: AI sometimes differentiates by making exaggerated claims rather than finding genuine distinctiveness. Check that everything in the rewrite is actually true of your company.

9. The Follow-Up Email Prompt

Write a pitch email to accompany this press release. The email is going to [journalist name] at [publication], who covers [beat]. Rules: - Subject line under 8 words - Email body under 150 words - First sentence explains why this is relevant to their specific beat - Include one line that a lazy journalist could copy directly as a story lead - End with a clear, low-friction call to action (not "let me know if you'd like to discuss") - Tone: helpful, not salesy. You're offering a story, not asking for a favour. Press release: [paste or summarise]

Why it works: the word limit is the key constraint. Journalist pitch emails need to be short. AI defaults to writing 300-word emails. Force it to be concise and the quality goes up dramatically.

Watch out: AI can't know the journalist personally. The "relevant to their beat" line will be generic unless you add specific context about what they've recently covered. Do your homework first.

10. The Multi-Format Prompt

Take this press release and create all of the following. Each should be a standalone piece — not just a shorter version of the press release. 1. A 150-word LinkedIn post from the CEO's perspective (first person, conversational, one clear takeaway) 2. A tweet thread (4-5 tweets, each self-contained, thread opener must hook without context) 3. A 50-word email pitch to a journalist (subject line + body, total under 50 words) 4. An internal Slack message announcing this to the team (warm, specific, credits people by role not name) Press release: [paste]

This is the workhorse prompt. One announcement, four formats, ten minutes. The alternative is writing each from scratch, which takes an hour.

Watch out: the LinkedIn post will probably be too polished. Real LinkedIn posts have rougher edges. Add a personal observation or an admission — something AI wouldn't include.

What AI can't do

These prompts are tools. Good ones, I think. But there's a category of PR work that no prompt can touch, and it's worth being explicit about what that is.

Newsworthiness judgment. AI can pattern-match, but it can't tell you whether your announcement will actually land. That requires understanding the current news cycle, the journalist's recent coverage, what stories are oversaturated and what gaps exist. Human work.

Relationship context. A pitch email generated by AI is structurally fine. But it can't know that this journalist just covered a competitor's round and might be receptive to a counter-narrative. Or that they're personally interested in a specific technology angle. That's where real PR professionals earn their fee.

Timing. When to release, when to embargo, when to hold. These are judgment calls informed by market context, competitive dynamics, and editorial calendars that no model has access to.

Voice. AI gets close. It doesn't get there. The last pass on any press release should be a human making it sound like your company — not like a company.

If you want someone to handle the 20% that AI can't — the strategy, the judgment, the relationships — that's what we do.