“We Sent The Release. Now We Wait And See If Anyone Cares”
“Coverage was about the same as last month.” That line closes the PR slide in every quarterly review, and nobody in the room asks a follow-up.
Reporters don’t reply. The pickup count barely moves. After a while the sending itself becomes the point, which is roughly the emotional register of writing holiday cards.
I’ve felt that too. Press releases are the part of the job where the feedback loop is basically broken.
But the thing on the other end of that loop stopped being only reporters at some point, and nobody sent us a memo about it.
Somebody actually went and looked at what happens after a release goes out — more than 8,000 of them, tracked for a week. So let’s look at what they found.
How Long Does It Take An AI To Cite A Press Release?
About eight hours, on average, from the moment it hits the wire — and most of the citations happen in the first 24 hours.
The numbers come from Notified, published July 9, 2026. Notified runs GlobeNewswire, a — the paid channel that pushes your announcement out to newsrooms and databases in one shot. Which, for you, means the numbers below come from a vendor measuring its own pipe. Worth keeping in mind as you read them.
The sample is over 8,000 releases distributed on GlobeNewswire during one week in May 2026 — local, national and global circuits, product announcements and earnings alike. Profound, a generative-AI analytics firm, did the citation counting.
And here’s what came out:
- average time to first AI citation: 8 hours
- most citations fall inside the first 24 hours
Compare that to SEO (search engine optimization), where a new page takes weeks or months to start pulling anything. The AI side answers the same day, so the gap between doing a thing and seeing whether it worked shrinks to hours. For a PR team, that’s a bigger deal than it sounds.
If Almost Everything Gets Cited, The Question Becomes How Much
The same study carries a louder number: 99.3% of the analyzed releases were cited by ChatGPT or Claude.
I wouldn’t take that one home as-is, and here’s why.
- Notified sells the distribution service, so “releases get cited” is a conclusion it profits from
- the measurement only covers what went through its own wire, with no control group of undistributed content
- what counts as a “citation” isn’t spelled out anywhere
So the causal read — put it on a wire and you’re 99.3% cited — isn’t supported by this study alone.
But flip it around and the number does useful work. If nearly everything gets cited at least once, then “will we be cited” stops being the interesting question.
How much you get cited is what’s left. And that’s a different game with a different lever.
Structured Content Pulled 2.6x The Citations
The lever the study points at is structure. Content that was well structured earned 2.6x the average citations.
Structure here means something plain: are the facts, names and context sitting where a model can grab them fast, or are they buried three paragraphs into a mood piece?
There’s a second number in the same direction. Owned content backed by a press release was twice as likely to get cited in its first 30 days.
They give an example, too. An earnings release from American Battery Technology Company picked up over 1,600 AI citations in 30 days — beating its industry peers, and beating national chains several times its size.
So maybe the thing that separated them wasn’t company size at all, but the shape of the announcement. One example doesn’t settle it, but it points the same way the 2.6x does.
Three Of SOAR’s Four Signals Show Up In Other People’s Data
Notified also published a joint report with PRWeek covering 250,000+ releases and roughly 250 million AI citations. Out of that came SOAR, their four-signal framework for what makes an AI pick one source over another.
- Structure — can a model locate the facts and proper nouns quickly
- Originality — is there first-party data, original research, real human expertise
- Authority — is the publisher credible as a source
- Recency — is it fresh, updated, re-distributed
Coming from the wire company, that’s a framework I’d normally file under marketing. But three of the four line up with data from people who have no stake in selling distribution.
On originality: original research is the most-cited content type at 82%. On structure: adding sources and numbers moves citations by around 40%. On recency: 65% of what AI cites is from the past year.
Independent parties, different entry points, same direction. That’s the kind of agreement I’ll take as corroboration even when I don’t trust the original messenger.
Write The Next Release For A Reader That Extracts, Not Skims
The 99.3% is an upper-bound number from an interested party, and it doesn’t need to be the thing you carry out of here. What’s worth keeping is the eight hours and the 2.6x: one says you’ll see results by tomorrow instead of next quarter, the other says two identical announcements land very differently depending on how they’re written. Both are in a PR team’s own hands, which is more than you can say for most of this stuff.
So here’s the version I’d actually run, and it’s two moves.
Rewrite the next release for extraction: at least one original number in it, and the facts, proper nouns and dates pushed toward the top of each paragraph. Trade a paragraph of atmosphere for a paragraph of specifics; that single edit is the whole 2.6x argument, applied.
Then go count. Ask the models 24 hours after it goes out, and again at 30 days, with the same questions both times. Nobody can tell you whether your announcement got picked up unless somebody actually looks.
The eight-hour figure comes out of one week in May, so I wouldn’t bet a strategy on it holding to the minute. But a press release is one of the few surfaces where you control 100% of the content, and you’re already publishing them every month — the experiment costs you one announcement.
Sources
- Notified, “More Than 99% of GlobeNewswire Press Releases Are Cited by ChatGPT or Claude, New Study Finds”, July 9, 2026, globenewswire.com (Over 8,000 press releases distributed on GlobeNewswire during one week in May 2026, with citations measured jointly with generative-AI marketing analytics firm Profound. Covers local, national and global distribution circuits and includes product announcements, earnings and non-earnings news. 99.3% of the analyzed releases were cited by ChatGPT or Claude; average time to first citation was 8 hours, with most citations occurring within the first 24 hours. Well-structured content earned 2.6x the average citations, and owned content backed by a press release was twice as likely to be cited in the first 30 days. As an example, an earnings release from American Battery Technology Company earned over 1,600 AI citations in 30 days. Caveats: publisher Notified is a press release distribution provider with a commercial interest in this conclusion; measurement is limited to material sent through its own wire, with no control group of undistributed content; the procedure for counting a “citation” is not disclosed, so the figures should be treated as upper-bound.)
- Notified / PRWeek, “AI Is Your Audience: Making Your Content Soar”, July 22, 2026, globenewswire.com (Analysis of over 250,000 press releases and roughly 250 million AI citations, distilled into the SOAR framework of four signals AI uses when choosing what to cite: Structure, Originality, Authority and Recency.)
