A press release format is the fixed structure a news announcement follows: headline, dateline, lead paragraph, body, boilerplate, contact information, end marker. Wire services, journalists, and AI systems all expect the same order. A release that follows the format moves through every distribution channel without friction. A release that skips a part forces an editor — or a model — to guess at what matters.
I've written press releases since the 1990s. The format itself has barely moved. What changed is who reads the release first — and what that reader is optimizing for.
The Press Release Retrieval Study 2026 quantifies the stakes: 81% of wire-distributed press releases produce zero AI citations. The 19% that do get cited share structural characteristics — numbers-first leads, entity density, schema-marked boilerplates — that are format decisions, not budget decisions. Everything below is written with that data in mind.
The Classic Format, Briefly
Every release still runs through the same seven parts.
- Headline. States the news in one line, present tense, under 12 words.
- Dateline. City, state, and date.
- Lead paragraph. Answers who, what, when, where, and why in the first sentence.
- Body. Supporting detail and quotes, ranked by importance — most important first.
- Boilerplate. A fixed paragraph describing the company. Identical release to release.
- Contact information. Name, title, phone, email.
- End marker. ### or -30-. Signals nothing follows.
That order has held for decades because it serves an editor scanning fifty releases before lunch. It still serves that editor. Now it also serves a model reading the same release in a fraction of a second.
The skeleton I hand clients, filled in as a template:
FOR IMMEDIATE RELEASE
[Headline: States the News, Present Tense, Under 12 Words]
[CITY, STATE], [Month Day, Year] — [Company Name] today announced [specific, numeric claim]. [One sentence of context, stated as fact, not enthusiasm.]
"[Direct quote from a named executive. Full name and exact title, spelled the same way every time the company publishes them.]" said [Full Name], [Exact Title], [Company Name].
[Paragraph of supporting facts, ranked by importance. Each sentence should stand on its own outside the paragraph and remain true.]
[Second quote or data point, if the release supports one.]
About [Company Name] [Fixed boilerplate paragraph. Copy it from the last release. Do not rewrite it.]
Media Contact: [Full Name], [Title] [Email] [Phone]
###
What Changed Structurally in the AI Era
Most guides to press release format cover the seven parts above and stop there. That advice holds up. It was written for one reader: a human editor deciding whether to pass the release along. It ignores the reader that now sees the release first — more often than the editor does.
Language models ingest press releases directly — through wire feeds, news APIs, and crawled coverage — well before most releases reach a human inbox. That reader does not skim for tone. It extracts facts, attaches them to entities, and cites the source later when someone asks a question in ChatGPT, Claude, Gemini, or Perplexity. Three parts of the format carry more weight than they used to.
Numbers-first leads. A lead that opens with scene-setting delays the fact a model needs to extract. "In an era of rapid change, Acme is proud to unveil a new chapter" buries the number. "Acme Corp raised $40 million in Series C funding led by X" puts it in the first clause. The second sentence gets pulled into an answer. The first gets skipped. The retrieval study found a ~3x citation multiplier for releases with a citable number in the first 100 words.
Standalone factual sentences. Write each sentence so it carries its meaning without the sentence before it. Models tend to extract and cite single sentences or short spans — not full paragraphs. A sentence that leans on "this" or "it" to refer back two sentences loses its subject the moment it gets lifted out. Every sentence that carries a fact should name its own subject.
Consistent entity naming. Spell the company name, product name, and executive name identically every time. "5W AI Communications" in one release, "5W" in the next, "5W Public Relations" on the website — that splits the citation across three strings instead of consolidating it behind one. A model builds a profile around a name. A name that drifts dilutes how much the model can say about the entity with confidence. Same for titles: "Chief Executive Officer" in one release and "CEO" in the next reads as identical to a person and as two different strings to a system matching text at scale.
The boilerplate matters more for this reason, not less. It is the one paragraph guaranteed to repeat, word for word, across every release a company issues. It is the most reliable string a model has for compressing what an entity is into something it can cite. The retrieval study found a ~2x multiplier for schema-marked boilerplates.
None of this replaces the classic format. It sits on top of it. Releases that perform well in both worlds still open with a dateline and close with a boilerplate. They have simply stopped spending the first sentence on adjectives.
The Traps
Adjective-first leads. "Proud," "thrilled," "excited to announce" push the fact back a sentence and leave a model nothing to extract from the first one. State the fact — add color after.
Inconsistent naming across releases. A different spelling, abbreviation, or title for the same entity each time splits the citation and slows how quickly a model attaches authority to the name.
A boilerplate that gets rewritten every time. Freshening the "About" paragraph for variety removes the one guaranteed repeating string that ties every release back to the same entity.
Quotes that restate the headline. A quote should add a fact, a number, or a reason. A quote that repeats the news in a different register wastes the most citable slot in the release.
A missing dateline. Skipping the dateline strips the release of the timestamp a model needs to resolve "recently" or "this week" against an actual date.
FAQ
What is the correct press release format? Headline, dateline, lead paragraph, body, boilerplate, contact information, and an end marker (### or -30-). Standard since wire services set the convention. Still applies.
Does press release format matter for AI visibility? Yes. AI systems parse structured, factual text more reliably than narrative text. A release with a numbers-first lead and consistent entity naming gets cited more often than one that buries the fact under adjectives. The Press Release Retrieval Study 2026 measured a ~3x citation multiplier for releases with a citable number in the first 100 words.
Should I write one press release format for journalists and a different one for AI systems? One format serves both. The seven-part structure still works for a human editor. Numbers-first leads, standalone factual sentences, and consistent entity naming make the same release work for a model reading it first.