How to Use Claude to Write Press Releases That Don’t Sound Like AI
By Gabriel Tan | March 2026
Most AI-written press releases are bad. Not bad in a subtle way. Bad in the way that makes an editor close the email, a journalist skip the pitch, and an investor question whether anyone at the company actually read the thing before it went out.
The problem is not that AI cannot write. The problem is how people use it. They open a chat window, type “write a press release about our Q4 earnings,” and hit enter. What comes back is a generic template stuffed with inflated adjectives, fake enthusiasm, and sentences that could belong to any company on any exchange in any country. It reads like AI because it was written like AI, with no context and no structure.
There is a better way. And it does not require you to become a prompt engineer or buy expensive tools. It requires preparation, a clear process, and a human at the end who makes the final call.
I have been using Claude to draft press releases for our clients. Not as a replacement for human judgment, but as a first-draft engine inside a governed workflow. The output is consistently better than what most junior consultants produce on a blank page, and it takes a fraction of the time. But only because the system around it is designed properly.
Here is how to set it up.
Step 1: Before You Write Anything: Two Files You Need to Prepare
Most people skip straight to the prompt. That is the mistake. The quality of what Claude produces depends almost entirely on what you feed it before you ask it to write. You need two reference documents loaded into your Claude project before you draft a single release.
File 1: An Anti-AI Writing Guide
AI has tells. If you have read enough AI-generated content, you know the feeling. Everything sounds polished but nothing sounds real. There are specific patterns that give it away, and you need to teach Claude to avoid them.
Your anti-AI writing guide is a markdown file that lives in your Claude project. Claude reads it before producing any draft. It contains the patterns you want Claude to watch for and correct in its own output.
What goes in this file: the specific language habits that make writing sound machine-generated. Trailing “-ing” phrases that gesture at importance without saying anything concrete (”…reflecting a broader shift in market sentiment”). The “not X, but Y” construction that creates fake drama. Em dash overuse.
Clusters of words that spiked in frequency after 2022, words like “pivotal,” “nuanced,” “multifaceted,” “cornerstone.” Vague attribution like “industry experts suggest” where no expert is named. Uniform sentence length that reads like a metronome. Groups of three adjectives used out of reflex rather than purpose.
The good news is you do not need to write this from scratch. Wikipedia’s AI Cleanup project maintains a list of common AI writing tells. Start there. Pull out the patterns that are most relevant to press release writing, write them up in plain language, and save the file as a markdown document in your Claude project.
The file does two things. First, it gives Claude a self-correction layer. It checks its own output against these patterns before showing you the draft. Second, it gives your human reviewer a checklist to run during QA. Same standards, applied twice.
File 2: Your Client’s Brand Voice Document
Every client sounds different. The way a Catalist-listed growth company communicates is not the way a blue-chip REIT communicates. A founder-led tech firm does not sound like a third-generation family conglomerate. If your press release could belong to any company, it belongs to no company.
Your brand voice document is a markdown file that captures how this specific client sounds. Claude reads it before every draft and matches the tone, vocabulary, and style.
What goes in this file: the client’s tone (institutional? confident? cautious?), their language level (do they say “accretive acquisition” or “value-adding deal”?), words and phrases they actually use in their existing communications, words they would never use, how their CEO speaks in quotes (short and direct, or measured and strategic?), and their formatting preferences.
The fastest way to build this is to pull three to five of the client’s recent press releases or their annual report chairman’s statement. Read them. Note the recurring phrases, the sentence rhythm, the level of formality. Write it up in a page or two. It does not need to be a 20-page brand bible. It needs to be specific enough that Claude can tell the difference between Client A and Client B.
Upload both files into a Claude project dedicated to that client. Claude will reference them automatically. You build them once, then refine as you go.
Step 2: Fill In the Details: The Intake Form
Now you are ready to brief the actual release. But instead of typing a loose paragraph into Claude and hoping for the best, your team fills in a structured intake form. This forces them to answer the questions that matter before any writing starts.
The form covers the basics that every press release needs:
Company name and ticker. Full legal name, exchange, stock code.
Release type. Earnings, contract win, expansion, M&A, crisis, product launch, leadership change. This tells Claude what architecture to follow.
Objective. What is this release trying to achieve? Are you trying to attract media coverage? Signal to investors? Both? The objective shapes the angle, and Claude needs to know what you are optimising for.
The headline fact. The single most important news point, in one sentence. If your team cannot write this sentence, they are not ready to brief a release.
Who. Key people involved. Full names and titles. Who is quoted. Who approved the release.
What. What happened. Specific numbers, values, percentages. No vague language. “Revenue increased” is not acceptable. “Revenue increased 18% to S$42.3 million” is.
Where. Geography. Market. Jurisdiction.
When. Date of event. Effective date. Announcement date. Embargo if applicable.
Why it matters. Why this is news. What changes because of this. What was the situation before. This is the field most people fill in poorly, and it is the most important one. If you cannot explain why this matters, the release has no angle.
How. The mechanism. How the deal was structured. How the target was achieved. What drove the result.
Quote source and tone. Who should be quoted, and what tone they should strike. Confident? Cautious? Forward-looking?
Sensitive information. Anything that must not appear. MNPI restrictions. Regulatory constraints. Embargo details.
If a field is not relevant, your team writes “N/A.” Blank fields are not allowed. The discipline of filling in every field, even to say it does not apply, forces the kind of thinking that produces a better brief.
The Prompt
Once both reference files are in the Claude project and the intake form is filled in, you paste the completed form into Claude with the following prompt. This is the actual prompt. Copy it, adapt the bracketed sections, and use it.
I need you to draft a press release based on the intake details below.
Before writing, do the following:
1. Read the anti-AI writing guide in this project. Apply every rule in that document to your output. Do not mention that you have done this. Just apply the rules silently.
2. Read the client brand voice document in this project. Match the tone, vocabulary, sentence rhythm, and formatting preferences described in that document throughout the entire draft.
3. Review the intake details below. If anything is unclear, contradictory, or missing critical context, ask me clarifying questions before you begin drafting. Do not guess. Do not fill gaps with generic language. Ask.
Once you have the clarity you need, draft the press release following this structure:
- HEADER: [FOR IMMEDIATE RELEASE] or [EMBARGOED UNTIL date/time/zone]. Include ticker and exchange if the company is listed.
- HEADLINE: Active voice. Company name included. Strongest news point leads. No adjectives like 'exciting,' 'thrilled,' 'delighted,' 'pleased,' 'unprecedented,' or 'proud.'
- SUB-HEADLINE: One sentence of supporting context.
- DATELINE: [CITY] — [Date]
- LEAD PARAGRAPH: Three sentences maximum. Sentence 1: the quantitative news (the what). Sentence 2: the strategic driver (the why). Sentence 3: the impact on the business.
- BODY: Evidence, data, context. Use data tables for financial comparisons where relevant. Every claim supported by a number or a named source.
- EXECUTIVE QUOTE: Match the quote style from the brand voice document. The quote must add something the data does not: strategic direction, confidence, or intent. It must sound like something the person would actually say in a meeting.
- BOILERPLATE: Use the client's standard company description exactly. Do not rewrite it.
- CONTACT DETAILS: As provided in the intake form.
After completing the draft, suggest:
MEDIA ANGLES: 2–3 angles a journalist would find interesting. For each, explain what makes it newsworthy (tension, novelty, trend, counter-narrative, or human element), which type of outlet would pick it up, and a suggested headline rewrite for that angle.
INVESTOR ANGLES: 2–3 angles an analyst or fund manager would care about. For each, explain what valuation signal it sends, which data point in the release supports it, and a suggested headline rewrite for that angle.
Do not rewrite the full release for each angle. Suggest the angles and alternative headlines only. My team will decide the direction and revise.
Forward-looking statements must include appropriate disclaimers.
[PASTE COMPLETED INTAKE FORM HERE]
Two things to note about this prompt. First, it tells Claude to ask clarifying questions before writing. This is deliberate. A human editor would push back on a vague brief. Claude should do the same. If the “why it matters” field says something weak like “this is good news for shareholders,” Claude should ask what specifically makes it good news and for which shareholders.
Second, the media and investor angles come after the draft, not before. The draft establishes the facts. The angles reframe them. Your team reviews both, picks a direction, and edits the draft accordingly. Claude suggests. Humans decide.
Step 3: After the Draft: The Human QA Process
Claude’s job ends at the draft and the angle suggestions. From here, a human takes over. This is not optional. No AI-generated press release should reach a client, a journalist, or a wire service without a human review.
The reviewer works through a simple QA pass. This is what they check:
Facts. Every number verified against the source document. P&L, board minutes, signed contract. No number goes out on trust.
Names and titles. Every person named in the release checked against their current title. Not from memory. From LinkedIn or the latest annual report.
Regulatory compliance. Forward-looking statement disclaimers present. MNPI restrictions respected. Embargo details correct.
Tone. Read the draft against the brand voice document. Does it sound like this client? Would the CEO read this and recognise it as their company’s voice?
Anti-AI second pass. Read the draft out loud. If any sentence sounds like it was written by a machine, flag it and rewrite it in plain language. Claude runs the anti-AI check internally, but the human runs it again. Two passes catch what one misses.
Headline test. Show the headline to someone who has not seen the brief. Can they tell you what the news is? If not, the headline is not working.
Quote test. Read the executive quote out loud. Would this person actually say this in a board meeting? If the quote sounds like a template, it needs to be rewritten.
Links and references. Every hyperlink works. Every source cited exists.
Format. Correct template. File naming convention followed. Distribution list confirmed.
Once the reviewer is satisfied, the draft goes to the client for review. Not before. The rule is straightforward: the drafter confirms the content, the reviewer confirms the quality. Two people sign off before it leaves the building.
What This System Actually Changes
The press release workflow in most agencies starts with a blank page and ends with a senior consultant rewriting a junior’s draft from scratch. The senior spends two hours on production work that should have taken 30 minutes. The junior never learns what went wrong. The client gets a release that is adequate but not sharp.
This system changes the starting point. Instead of a blank page, your team starts with a client voice filter, an anti-AI writing guide, and a structured brief. The result is a first draft with controlled tone and suggested angles they would not have thought of on their own. The senior’s time shifts from rewriting to reviewing. The junior’s time shifts from guessing to filling in a structured form and learning from Claude’s output.
The quality of the first draft goes up. The time to produce it goes down. The governance is built in. And the final decision, the angle, the tone, the word that matters, stays with a human.
That is where it should be. AI produces the draft. Your team makes it right.
Gabriel Tan is the founder of Mekong Bridge Advisory. He builds structured execution systems for PR and communications firms.
