Do Press Releases Get Your Brand Into AI Answers?

PR teams are being sold "GEO press releases" as an AI visibility channel, and the two best datasets on the question appear to flatly contradict each other. Muck Rack's May 2026 analysis of more than 25 million links cited by ChatGPT, Claude, and Gemini puts paid and advertorial content at 0.3% of all AI citations. Stacker's July 2026 study tracked 245 wire-distributed brand stories and found them present in 9–20% of related AI answers within four weeks. One number says press releases barely register; the other says they reach a fifth of answers. Both are measuring something real, and the gap between them is where the budget decision lives.

Key facts

Key takeaways

Muck Rack's What Is AI Reading? series is the largest published breakdown of what AI engines actually cite: the May 2026 edition covers more than 25 million cited links across ChatGPT, Claude, and Gemini in 17 industries. We have covered its headline finding — 84% of citations point at domains the brand does not own — but the press-release detail matters on its own.

Two numbers frame it. Paid and advertorial content earns 0.3% of all citations, stable across three editions of the study. And press releases, which Muck Rack tracks as a separate category from both advertorials and journalism, "appear almost exclusively in industry trend responses, at 3.5 times the rate they appear in best-of queries."

That second number is the useful one, because it is a map of where the channel works at all. Trend-shaped questions — what's happening in a category, what changed this quarter — pull from news-indexed sources, and a genuinely newsworthy release can be one of them. Evaluation-stage questions — best tools, alternatives, comparisons — pull from review platforms, forums, editorial lists, and product pages instead. Those are the prompts where a recommendation would actually pay, and they are precisely where press-release citations approach zero.

Stacker's study is the only published tracking of wire-distributed content through AI answers, and we used its timing data in how long it takes to get cited by AI. Read for efficacy rather than timing, it says something different.

Stacker tracked 245 brand stories for four weeks after distribution and measured response presence rate: the brand named in the answer, its own domain cited, or a Stacker-distributed copy of the story cited. Average presence ran from 20% on Google AI Overviews and 19% on AI Mode down to 13% on Perplexity and 9% on ChatGPT; medians were roughly half the averages, with ChatGPT at 3%. Google's surfaces spiked early and settled lower, while Claude compounded — 118 of the 132 stories with any Claude presence grew between weeks one and four.

The provenance discount is significant and worth stating plainly: Stacker sells the distribution being measured, and citations to its own syndicated copies count toward presence. The absolute rates therefore describe what a distributed story can touch, not what your domain earns. But even discounted, the shape of the result is informative — distribution produces measurable, mostly temporary presence on news-shaped prompts, strongest on Google's surfaces and weakest on ChatGPT.

Set side by side, the two studies look irreconcilable. They differ on every axis that matters, and lining the axes up is the actual answer to the question.

The reconciliation: a press release almost never becomes the cited source of record for an answer, and never at all outside trend-shaped queries — that is the 0.3% world. But distribution can still put a brand's name in a nontrivial share of answers for a few weeks, often with no citation to any page the brand controls — that is the 9–20% world. The gap between being named and being cited is the same one we measured from the other direction in ghost citations: mention and citation are separate signals, and a wire release is a machine for producing the first without the second.

For a PR team, that means a release is a mention instrument with a decay curve, not a citation asset. Whether that is worth the wire fee depends entirely on whether the prompts you care about are announcement-shaped.

Our weekly research sweep pulls the Google News results for answer-engine-optimization and AI-visibility queries, and the week of 5–12 September 2026 makes the saturation measurable: of the 12 news items surfaced, 8 were newswire press releases — four syndicated through a single financial-news network — including three "best agencies" or "best tool" rankings attributed to research firms with no footprint beyond their own announcements.

The mechanism these are built on is exactly the one in the data above. A wire release lands on dozens of news-indexed domains, those domains are eligible for trend-shaped retrieval, and for a while the sponsoring brand is present in answers about its category. What the mechanism cannot do is cross the query-type boundary: the best-of and comparison answers a "best agencies" ranking is nominally aimed at draw on review platforms, forums, and editorial coverage — the source types where press releases appear least. The tactic buys presence in the answers nobody converts from.

It also carries a policy edge worth knowing. Google's spam policies permit syndicated news content but list "links with optimized anchor text in articles, guest posts, or press releases distributed on other sites" as link spam. Since Google's AI surfaces run on core Search systems, a release written as an anchor-text delivery vehicle is the version that risks the underlying index eligibility everything else depends on.

The same datasets that discount the wire point at what works. Journalism alone accounts for 27% of AI citations in Muck Rack's data — the largest single category, and stable across a year of editions. And the citation value of coverage is unevenly distributed across outlets: the August 2026 Press Ranger and OtterlyAI analysis of 129.3 million citations found pages from OpenAI-licensed publishers collecting 10.2 ChatGPT citations each against 6.9 for unlicensed publishers, a premium we unpacked — caveats included, starting with the fact that it is itself a press release from two vendors — in what licensing deals do to ChatGPT citations.

The practical split for a comms team:

One measurement note ties it together: a release's effect is mostly mentions, and mentions churn more slowly than citations but need the same sampling discipline to read. A single check the morning after distribution measures nothing; presence in Stacker's data was still moving in week four.

Elmo is an open-source AI visibility platform that tracks both halves of this separately: brand mentions and page citations, per prompt, per engine, on a schedule — so an announcement's presence bump and its decay are visible as data rather than vibes, and your earned-media target list comes from the domains actually cited in your category's answers. Self-host it free or use the managed cloud; start with how to track your brand in AI search, and see the AI search glossary for the vocabulary.