For a hundred years, press releases targeted one species: the working journalist.
They were stapled to fax machines, wired through PR Newswire, and embargoed until 6 a.m. Eastern. The format optimized for one thing: a tired reporter scanning for a quote.
That reader is gone. In its place: a probabilistic model, an answer engine, an autonomous agent. They don't skim. They don't follow links. They don't care about your boilerplate.
The press release of 2026 is still being written for an audience that was extinct when it was designed.
Eight principles for the
AI-native press release.
A working draft. We expect to be wrong about half of these in 18 months. We expect to be right about the rest for a decade.
- 01
Structured over stylish.
Schema.org markup, JSON-LD, atomic facts. The lede is metadata, not metaphor.
- 02
Citeable, not clever.
Every claim is a standalone sentence with a source. LLMs cite sentences, not paragraphs.
- 03
Quotes as facts.
Attributed quotes need full context inline. Models will quote you without your headline.
- 04
Machine-first metadata.
Date, entity, role, location, and relation are first-class fields — not buried in prose.
- 05
One canonical version.
No reformatted PDFs, no stylized image releases. One URL, one machine-readable truth.
- 06
Atomic, not narrative.
Break the announcement into discrete claims. Agents recompose; they don't paraphrase.
- 07
Linkable evidence.
Filings, datasets, demos. If it can't be retrieved, it won't be trusted.
- 08
Built to be ingested.
Robots.txt, sitemaps, RSS, and feed endpoints. If the crawler can't reach it, it doesn't exist.