Do press releases get cited by ChatGPT? What the data says

Ask whether press releases help AI search and the data gives a split answer. The release as it sits on a newswire's own website is almost never what ChatGPT, Perplexity or Gemini cites. The news and editorial pages that carry or cover the same announcement show up constantly.

That distinction matters more than any single statistic. Below is what the independent studies found, what the wire vendors claim, why the two sets of numbers look so different, and what to do about it.

What independent studies found

Several research groups have counted where AI answers get their citations. Their methods differ, but on press releases they land in the same place.

Meltwater's conclusion is worth quoting: the effect of distribution is "usually indirect." A release helps when journalists, trade outlets and local news pick it up, and those pages get cited. Meltwater counted earned and news media at 37.6% of citations, which is lower than Muck Rack's 84% because the two firms classify sources differently. Both put press releases near zero.

Muck Rack's earned-media share has ranged from 82% to 89% across editions of its report. The Institute for Public Relations summarized the finding as sustained journalism coverage beating "paid content or press releases alone."

Loganix's study is the most useful for anyone buying distribution, because it tested specific URL paths. The wire domains themselves never appeared. Releases that landed on Yahoo Finance's /news/ path got cited by ChatGPT and Gemini. Loganix sells a product built around that path, so treat it as vendor research. Other analysts have read the same data as showing that secondary pickup matters more than wire copy.

The vendor numbers that say the opposite

Wire services and distribution companies have published their own studies, and some of the headline figures look like a flat contradiction.

Notified: 99.3% of releases cited

Notified, which owns GlobeNewswire, reported in July 2026 that 99.3% of GlobeNewswire releases in a study of 8,000 were cited by ChatGPT or Claude. It also reported an average of 8 hours to first citation, said 17% of citations arrived in the first 24 hours, and said well-structured releases got 2.6 times more citations. The materials we reviewed did not show the methodology, including which prompts were run and what counted as a citation.

Stacker: a 239% median lift

Stacker, which places brand-funded stories across publisher sites, ran a study with its measurement partner Scrunch. It found a median 239% lift in AI citations for distributed stories compared with brand-owned content. 97% of distributed stories earned at least one citation, and 64% of those citations pointed to third-party publisher URLs. An earlier pilot of 8 stories moved from an 8% citation rate on owned sites to 34% with distribution. Stacker's data is more detailed than most in the category and names its measurement partner. Still, it was run on Stacker's own customers, and one critical review called it "a measured citation-rate change inside distribution cohorts, not a universal lift."

ACCESS Newswire has also published vendor research saying newsroom-published press releases account for 18% of ChatGPT's citations.

All three of these studies come from companies that sell distribution. That doesn't make them wrong, but it means the methodology deserves more scrutiny than the headline.

Why the numbers conflict

They mostly measure different things. Some studies divide by all AI citations. Others divide by the releases a vendor distributed. Others measure change rather than share.

A release can be cited once and still make up a vanishing share of all citations. Prompts matter too. If a test asks about a specific company or announcement, a model may surface the release because few other sources exist. Category questions like "best field service software" are a harder test, and that is where Loganix found zero wire domains. Notified's materials don't say which kind of prompts it used, so we can't tell which explains its result.

Read closely, Stacker's data supports the independent studies: most of its citations came from publisher URLs, not the brand's own pages.

Which domains do AI assistants cite most?

The biggest cited domains are not news sites. Each assistant has its own habits.

• Ahrefs (78.6M searches): YouTube is about 16% of Perplexity citations and about 9.5% of Google AI Overviews citations, with Reddit also near the top.
• Profound: Wikipedia is 7.8% of ChatGPT citations, and Reddit holds 21% of AI Overviews' top-source share. Perplexity favors review and reference sites such as G2, Gartner, NerdWallet, PCMag, TripAdvisor and Yelp.
• Peec AI (30M sources): Reddit, YouTube, LinkedIn, Wikipedia and Forbes are the top five across engines.

The more useful finding for a company with news to share is how little the assistants agree. Everything-PR and 5WPR both report that only about 11% of domains are cited by both ChatGPT and Perplexity. No single placement covers every assistant. A press release won't put you on Reddit or Wikipedia, but it can put you on the news domains that sit in the long middle of every assistant's source list.

How long does it take for AI to cite a press release?

Faster than search rankings, in the few tests that exist, but not on a fixed schedule.

• Otterly.ai (vendor test): a wire-distributed release on markets.businessinsider.com was cited by an AI search engine 2 days after publication. The first ChatGPT citation came about a week after.
• Notified (vendor study): an average of 8 hours to first citation for its own releases, methodology not public.
• Everything-PR: used a 90-day window and found 19% of wire releases cited within it.

The practical point: a single check on publication day tells you little. Check at intervals over the following weeks.

What this means if you publish press releases

The placement domain matters more than the wire

The Loganix and Otterly results point the same way. The citations went to pages on Yahoo Finance's /news/ path and Business Insider's markets section, not to the wire's own site. When you compare distribution options, ask which domains the release will appear on, and whether those are domains AI assistants already cite.

Breadth across trusted domains

With so little overlap between assistants, several credible news domains carrying the same facts give each model more chances to find you. The count of pages matters less than whether they sit on domains assistants already cite.

Be clear about what a placement is

A syndicated release on a news domain is not a journalist-written story. Muck Rack's 27% journalism figure counts reporters' work, and a press release placement doesn't become journalism by appearing on a news site. What it does give you is an indexed, dated page on a trusted domain that names your company and states your facts. Many of those links are nofollow or sponsored, so the value is the mention and the citable page, not link equity. We cover the SEO side in Do press releases help SEO in 2026? and the differences between formats in press release vs sponsored article vs earned media.

Write so the facts can be lifted

Notified's 2.6x figure for well-structured releases is vendor data, but the logic is sound: a model can only quote what is stated plainly. Our guide on how to write a press release AI assistants will quote covers the specifics.

Measure, don't assume

Run the questions your buyers would ask in ChatGPT, Perplexity and Gemini. Record which pages get cited, with URLs, and repeat over several weeks. Be wary of any offer of guaranteed AI citations or "guaranteed AI indexing." Nobody controls what a model cites, and the providers making those promises in our research don't publish how they measure them. For what a credible report should contain, see how to read a press release distribution report.

How Word Got Out reports AI citations

Word Got Out is built around the data above. Because the domain a release lands on matters, and because earned media drives most citations, a release is only one part of what we do. We start by mapping which sources AI assistants cite when buyers ask about your market, then publish and reach out where it counts, then measure the same questions again.

We don't guarantee rankings, traffic or AI citations, because no one can. We report what changed, including when nothing did, with a screenshot of every answer. Our measurement method is published.

The free snapshot is the place to start: it shows whether AI mentions you today and which sources it relies on for your category.

Frequently asked questions