What happened
Pangram Labs' analysis, distributed through Stacker's data journalism syndicate on September 16, examined four datasets totaling more than 190,000 press releases from PR Newswire, PRWeb, and Newswire.com between August 2025 and July 2026. The share of releases containing AI-generated text doubled in roughly a year, climbing from 22% to 48%. Of the flagged releases, a quarter were fully AI-written and three-quarters were mixed human-AI content, with a median AI contribution of 52% in flagged documents overall. The breakdown by industry was stark: advertising hit 64% overall AI use and 41% fully AI-generated, while finance, energy, and biotechnology all stayed under 5% fully AI-generated, suggesting regulated or high-stakes sectors are still writing releases by hand, likely due to legal review requirements. One content marketing agency, HelloNation, saw 97% of its 2,265 tracked releases flagged as fully AI-generated, an outlier that shows how far the practice can go at scale.
Why AI-written press releases matter now
MagnetizeX builds founder visibility systems for B2B firms.
Journalists and editors are getting better at spotting AI-generated pitches, and multiple recent surveys show newsrooms actively filtering them out before they're even read past the subject line. A press release landscape where nearly half of everything distributed is AI-flagged means the releases that clearly aren't, specific, sourced, written in an actual human voice, stand out by sheer contrast with the flood around them. For a founder sending news to press, this is less a warning about using AI tools at all and more a warning about using them lazily, since the industry-level data on advertising's 64% AI-use rate suggests that's exactly where reporters' skepticism is currently concentrated hardest, ahead of nearly every other sector Pangram tracked.
- Read your next release for detectable AI patterns first.Pangram's own detection tooling flags generic transitions, unsupported superlatives, and structure-first writing. Run your draft through a detector before sending, or at minimum read it aloud slowly and listen for the tells yourself.
- Lead with one fact a generator couldn't invent on its own.A specific customer name, a real dollar figure, an internal metric you're choosing to disclose publicly. Pangram's data shows fully AI-generated releases skew toward generic superlatives; concrete specifics are the fastest way to read as genuinely human-written.
- Avoid the release template your whole category is using.If you're in a high-AI-use category like advertising or marketing tech, assume your pitch starts more, not less, suspect in a journalist's inbox, and over-invest in a distinct structure and voice relative to competitors pitching the same story.
What to do this week
Before your next press release or pitch goes out, have one person on your team read it cold and flag any line that could have been written about literally any company in your category. Replace those lines with a specific number, name, or detail unique to your business, since that specificity is exactly what's getting harder to fake convincingly and easier for editors to notice its absence in a crowded, increasingly AI-flagged inbox.