Press Releases with Descriptive Section Headings Receive 61% More AI Citations, New Notified Study Finds

Key Facts:

  • First-of-its-kind research: Notified's new report, "Written for People, Read by Machines," is the first to compare AI citation behavior for newswire-distributed press releases over time across leading AI systems. The Q4 2026 report includes a dataset of more than 25,000 GlobeNewswire releases and nearly 1 million AI citations.
  • Top findings: Press releases built around topical subheadings receive 61% more AI citations. One of the most effective ways to boost visibility is to link to a company's own website; releases that do so receive 57% more AI citations.
  • Why it matters: The research uncovers that a press release's reach extends well beyond distribution day, as more than 56% of AI citations from ChatGPT and Claude arrive after week one.

NEW YORK, Oct. 29, 2026 (GLOBE NEWSWIRE) -- Press releases structured with clear, topical subheadings receive 61% more AI citations, according to a new study from Notified, the communications technology company that operates GlobeNewswire. The research also finds that releases continue to appear in AI answers weeks after publication.

The full findings are available in the report, "Written for People, Read by Machines," which analyzed more than 25,000 GlobeNewswire releases and nearly 1 million AI citations to pinpoint the content characteristics behind stronger citation performance and examine how leading AI systems discover, index and cite content.

"Our groundbreaking research earlier this year showed the way AI models surface press releases. Now, the story has changed," said Erik Carlson, Chief Executive Officer at Notified. “What surprised us in this study was the staying power releases have. AI systems don't just cite a release once and move on. They keep citing it, again and again, for weeks across multiple models. A press release isn’t just an announcement, it’s a source for AI answers."

Key Findings by SOAR Pillar

The report is organized around the pillars of Notified's proprietary SOAR Content Framework™ (Structure, Originality, Authority, Recency), the data-backed model for creating content AI can easily read, trust and cite.

  • Structure: Press releases with descriptive subheadings such as "Key Findings" or “Financial Highlights” received 61% more AI citations, and those with high fact density earned 29% to 33% more.
  • Originality: Releases built on proprietary research earned no citation lift unless they stated their novelty plainly. Announcements that used "first-of-its-kind" language garnered 60% more AI citations.
  • Authority: Press releases that linked to a company's own website received 57% more AI citations. Including a generic corporate media email address, such as media@company.com, was associated with 38% more citations, while naming only an individual contact was associated with 12% fewer.
  • Recency: More than 56% of citations from ChatGPT and Claude arrived after week one, and both models mostly cited the version they crawled first.

What This Means for Communicators

  • Treat the release as a long-lived source: The majority of ChatGPT and Claude citations arrived after day 7, and both models mostly cited the version they crawled first, highlighting the importance of getting the story right before distribution.
  • Link to your website: Linking to a company’s website was the most consistent signal associated with higher AI citations.
  • Make structure and novelty explicit: Use clear subheadings and state plainly what is new.
  • Use a corporate media contact: Generic corporate media contact email addresses were associated with more citations than individual contacts.

Download the report to explore the full findings and get a ready-to-use, seven-step checklist for improving AI visibility.

Methodology: Findings are based on 26,670 English-language GlobeNewswire releases (June to July 2026) and comparative cohorts from April and July 2026 totaling more than 850,000 timestamped AI citation events. Lifts are observed associations, not proof of causation. Model-behavior findings are inferred from citation and crawl logs and do not represent vendor market share.