If your PR strategy is "write a press release, put it on the wire, wait for AI models to cite it," the data says you're waiting on the smaller half of a very lopsided equation. A large-scale study of LLM citation sources found earned and news media accounted for well over a third of citations, while press releases barely registered. That gap is worth understanding before you spend another budget cycle on wire distribution.
By the iSuggest.ai Team · Updated for 2026
The numbers, without the spin
A Meltwater study covering more than 8 million citations across 8 major LLMs (May 2026) found that earned and news media accounted for 37.6% of citations, while press releases accounted for just 0.2%. Do the division and press releases are cited roughly 180 times less often than genuine news coverage. That is not a rounding error or a sampling quirk — it's a structural difference in how these two content types get treated by the systems doing the citing.
We covered the same dataset from a different angle in our post on brand mentions outweighing backlinks, where brand-mention correlation with AI visibility ran roughly 3x stronger than backlink correlation. The press-release number sits inside that same picture: it's a form of brand mention, just a weak one on its own.
Why the gap exists
It isn't that press releases are inherently untrustworthy. It's that a wire-distributed release is, by design, the same text syndicated to dozens of low-differentiation placements — often with little independent framing, quotes, or verification added. LLMs doing retrieval and citation tend to favor sources that read as independently reported: a journalist's write-up that adds context, cross-references other coverage, or quotes people outside the company saying something in their own words.
A press release is a company talking about itself. Earned coverage is someone else talking about the company. Models built to surface credible answers weight those very differently, and the citation data reflects that directly.
What this doesn't mean
It doesn't mean press releases are worthless, and we're not going to tell you to stop writing them. It means you shouldn't treat "we put out a release" as a GEO strategy on its own, and you definitely shouldn't promise anyone that a release will get cited by ChatGPT or Gemini. We wrote a fuller checklist on that exact temptation in what to check after you publish a press release — the short version is that the release itself is one input among many, not a citation guarantee.
The realistic play is to use the release as raw material: the facts, quotes, and figures in it are legitimate content, and your job is to get that content in front of people who might independently write about it — media, industry newsletters, partner blogs, communities where your customers actually spend time.
Turning an announcement into something worth citing
A few things move the needle more than the wire distribution itself:
- Publish the release on your own site first, as a real, crawlable page — not just a PDF or a wire-service mirror — with clear dates, named people, and specific numbers. That page is something you control and can improve later.
- Structure it clearly. A well-organized announcement with a clear headline, a factual first paragraph, and named entities (people, products, company) is easier for any system — human editor or AI crawler — to extract and reuse accurately. This is the same entity-clarity principle we cover in why AI engines need to know who and what you are.
- Pitch it as a story, not a document. Give a reporter or newsletter writer a reason to write their own take, with their own framing and quotes. That independent version is what actually shows up in the 37.6%, not the 0.2%.
- Keep the page live and accurate after the news cycle passes. A dead or edited-beyond-recognition announcement page is a worse long-term asset than a slightly less exciting one that stays put.
Where an AI directory snapshot fits
Once you have a real, well-structured announcement page on your own site, publishing a frozen snapshot of it to our AI directory gives it a second, independently crawlable presence that Gemini, ChatGPT, and Claude can read directly — see how the AI directory works across all three. That's not a substitute for earned coverage, and we won't claim otherwise. It's a way to make sure the facts in your announcement are at least structurally legible to the models doing the citing, on top of whatever earned coverage you're able to generate separately.
The honest sequence, in order of actual leverage: get real people to write about your news in their own words, make your own announcement page as clear and well-structured as possible, and use tools like an audit and an AI directory listing to make sure that page is technically readable by crawlers. Skipping straight to "distribute the release and hope" is the one step the data says does the least.
If you want to see where your own announcement or product pages stand today, run a free audit at how it works, or check your existing credits and published snapshots from your account.