Press releases help AI visibility on branded and news prompts. They almost never help on the "best X software" prompts where buyers build a shortlist, unless an editorial outlet picks the story up. That's how two 2026 studies can both be right. Notified found that 99.3% of GlobeNewswire releases were cited by ChatGPT or Claude, while Loganix ran 100 "best X in Y" queries and found zero press-release domains in the answers.
The studies measured different things, and that difference is the answer. If you're weighing a wire release (list prices start around $850–$940 for a national release) against other GEO spend, the useful question is "which of my prompts would a release actually reach?"
Two studies, opposite headlines
The pro-wire number comes from Notified, the Equiniti business that owns GlobeNewswire. It looked at more than 8,000 wire-distributed releases from a single week in May 2026, covering local, national and global circuits and both earnings and non-earnings announcements. It reported that they were "cited by ChatGPT or Claude at a rate of 99.3%," with an average of about eight hours to the first citation. The citation data came from its partner Profound (Notified).
The anti-wire number comes from Loganix, an SEO and link-building provider. In March 2026 it ran 100 consumer-style "best X in Y" category queries on Perplexity, ChatGPT and Gemini, across ten verticals that included SaaS. Press-release domains appeared in zero of them. The citations went to editorial listicles, review platforms and niche comparison publishers such as G2, NerdWallet, Clutch and Wirecutter (Loganix via PR Newswire). Loganix announced that result on a newswire, which hints at what wires are actually good for.
Side by side, the two look like a straight contradiction. They aren't, because each one divides by something different.
Three denominators, three different questions
| Study | What it divides by | The question it answers | Headline figure |
|---|---|---|---|
| Notified (GlobeNewswire) | Releases | Was this release cited for anything, by anyone? | 99.3% |
| Loganix | Category queries | Did a wire domain appear in a "best X" answer? | 0 of 100 |
| Muck Rack, BuzzStream | All citations | What share of everything AI cites is a press release? | About 0.04%–1% |
Counting per release is generous by design. A release counts as cited if it shows up in one answer, for one user, on any query at all. Counting per query is strict. It starts with the prompt your buyer types and asks what shows up. Counting per citation tells you how big press releases are as a source type overall, and the answer is small. BuzzStream analyzed 4 million citations across 3,600 prompts. Direct newswire citations made up 0.21% of the dataset, and syndicated releases on Yahoo and MSN made up 0.04% (BuzzStream).
So the 99% says releases get read and used somewhere. The zero says they don't get used in category shortlists. Both can be true, and for a B2B SaaS team the second one is usually closer to pipeline.
What "cited" probably means in the 99% study
Notified's announcement doesn't define "cited," and it doesn't list the prompts. The method is a guess, but there is a clue. Erik Carlson, Notified's CEO, wrote on LinkedIn that "to align with Profound's study, we focused on ChatGPT and Claude." Analyst Vincent Orleck described that earlier Profound study as tracking "newly published marketing pages that were eventually cited by ChatGPT or Claude agents in Profound's logs" (Ephemeral). If the release study used the same approach, "cited" would mean an agent pulled the release into some answer for someone. It would not mean the release appeared for your buyer-intent prompts. Notified hasn't confirmed this, so treat it as a question to ask, not a finding.
None of this makes the data wrong. Ask any vendor study three things before you set a budget around it: which prompts, which engines, and what counts as a citation. Every study in this post comes from a company that sells something related to the answer, so the method matters more than the headline.
Where press releases do show up: news, trend and brand prompts
Once you sort by query type, the studies agree. Muck Rack analyzed more than 25 million links from ChatGPT, Claude and Gemini for its May 2026 report and found that "press releases appear almost exclusively in industry trend responses, at 3.5 times the rate they appear in best-of queries" (Muck Rack). Its follow-up FAQ puts the trend figure at about 1% of responses (Muck Rack FAQ). By our arithmetic (Muck Rack didn't publish this), that leaves best-of answers at around 0.3%.
BuzzStream's data has the same shape. Newswire citations were "most common in exploratory/informational prompts, but still just 0.37% of the time." Loganix's own test, the one that produced the zero, found something similar on brand queries such as "What is [Brand]?". There, the Yahoo Finance /news/ copies where wire releases syndicate were cited.
The kind of news matters as well. Everything-PR followed 250 Q1 2026 releases from PR Newswire and Business Wire across five engines for 90 days. M&A releases were cited 40% of the time (20 of 50), and regulatory or SEC releases 30%. Partnership, award and "other" releases were cited 0 times out of 50 (Everything-PR). Engines cite releases that contain facts people ask about, like who acquired whom or what was disclosed. They don't cite announcements nobody searches for.
For a SaaS founder, this is easy to apply. When you announce a funding round or an acquisition, the release can become the source engines use to answer "Who acquired X?" or "What does X do?", and quick, accurate brand answers are worth having. When you announce a new partner badge, the data suggests no buyer prompt will ever lead to it.
Where they don't: the "best CRM for small teams" prompt
Shortlist prompts give a very different picture. DerivateX ran 100 buyer-intent B2B software queries across 20 categories through Google AI Overviews in June–July 2026 and logged 1,259 citations. Third-party "best-of" blogs got 63.4% of them, vendor sites 16.8%, review sites such as G2 and Capterra 5.0%, and news and trade press 1.2% (DerivateX). Press releases weren't common enough to get their own category.
Overthink Group's July 2026 data points the same way. It ran "What is the best [category]?" prompts for 250 niche B2B software categories on ChatGPT, Gemini, Perplexity and AI Overviews. 70.8% of citations went to URLs with "best," "top," "leading" or "popular" in the title, and G2 was the most-cited single domain at 5.8% (Overthink Group).
Say the prompt you most want to win is "best CRM for small sales teams." In these datasets, that slot goes to listicles and review platforms almost every time. This matches what we found when we audited how concentrated AI citation sources are, and which share is earnable. Loganix reached the same conclusion: "Citation URLs on category queries is the wrong success metric — press content does not serve that layer."
If you want your own site to compete for that slot, how to write a fair comparison page that AI search engines will actually quote covers what makes those pages get cited.
That leaves one way for a release to reach a category answer: someone else has to write about it.
The multiplier is editorial pickup, not the wire
In Everything-PR's study, the factor most closely tied to citation wasn't the wire service. It was editorial pickup. Releases picked up by a top-50 publication were cited 38% of the time, compared with 6% for releases without that pickup. Across all 250 releases, only 47 (19%) were cited by any of the five engines within 90 days: 8% by day 30 and 14% by day 60. The study's conclusion is blunt: "After 60 days, a release that hasn't been cited is unlikely to be."
When a release was cited, the engines mostly cited the journalist's version:
| Engine | Cited releases | Via wire URL | Via editorial copy |
|---|---|---|---|
| Perplexity | 40 | 7 | 33 |
| ChatGPT | 28 | 2 | 26 |
| Gemini | 22 | 0 | 22 |
| Claude | 17 | 0 | 17 |
| Google AI Overviews | 14 | 0 | 14 |
As the study puts it: "The engines are not reading the wire. They are reading the journalist who read the wire." Perplexity was the only engine in the sample that cited wire URLs in any real number. Gemini, Claude and AI Overviews cited none. Meltwater, which sells PR software, reached a similar view from its own data: "Yes, press distribution can move the needle, but it rarely works like a switch" (Meltwater).
What gets a release picked up, and then cited, is much the same as what gets any page cited. In Everything-PR's sample, releases with a citable number in the first 100 words were cited 28% of the time, against 9% without one. LeadCoverage, a supply-chain go-to-market agency, published roughly 13 weekly releases over one quarter. Its data-led releases with a point of view earned 3.5 times more citations. The company's data was cited 1,058 times, about nine in ten of them by ChatGPT, while announcements, personnel changes and awards "earned almost no citations" (AJOT). CEO Kara Brown summed it up: "AI cannot invent a number, so it cites whoever published one."
That's the same mechanism behind why original research wins AI citations and opinion posts don't. The number is what gets cited, and the wire is one way to deliver it. Whether it's worth paying for that delivery is a question of cost.
The cost math for a B2B SaaS team
Start with list price. Everything-PR checked rates in October 2026: a 400-word US national release starts at $850 on PR Newswire and $940 on Business Wire. For the full cost, including writing, add-ons and targeting, the study uses an illustrative $3,000–$5,000 per release. That's its own estimate, not a published rate. With only 19% of releases ever cited, it works out to roughly $16,000–$26,000 per cited release. By the same estimate, the 50 partnership, award and "other" releases that drew no citations cost about $150,000–$250,000 combined.
You already own a cheaper channel. BuzzStream found that internal press releases and newsroom content made up 18.15% of ChatGPT's citations, against 2.98% across all platforms. Search Engine Journal's write-up of the same data notes that the distributed version of a release "rarely becomes the version AI platforms cite" (Search Engine Journal). Publish on your own newsroom first, on a stable URL with the key number near the top. For ChatGPT, that copy may matter more than the wire copy.
For the wider picture of which sources ChatGPT cites in 2026, including the domains that took over slots Reddit used to hold, see ChatGPT citation sources in 2026 and who replaced Reddit.
One more point, for SEO: wire releases don't work as a link-building tactic. Google's spam policies list "links with optimized anchor text in articles, guest posts, or press releases distributed on other sites" as link spam (Google Search Central). Even setting the policy aside, backlinks act as a threshold for AI citations rather than a driver, so release links were never going to do much here.
A short decision rule
Use the wire when:
- It's news people actually search for: funding, M&A, regulated disclosures, original data.
- You want fast, accurate answers to brand and news prompts.
- You have a plan for editorial pickup, such as pitching and outreach, because pickup is what multiplies the effect.
Skip it when:
- The news is a partnership, award or hire. In Everything-PR's sample those went 0 for 50.
- The goal is "best X" prompts. That budget does more in original research, earned coverage, and listicle and review-site presence.
How to test it on your own prompts
None of these studies ran your prompts. Before you renew a wire contract or cancel one, measure what a release does for the questions your buyers actually ask.
- Split tracked prompts by intent. Put branded and news prompts ("What does [you] do?", "Who acquired [competitor]?") in one group and category prompts ("best X for Y") in the other. Our free Prompt Ideas Generator turns a domain into 20 buyer-intent prompts tagged by funnel stage, a quick first split.
- Mark distribution day and pickup day. Without a date on the chart, any later movement is a guess.
- Read the result at week four, not day one. Stacker, which runs its own distribution newswire, measured brand presence four weeks after distribution and advises against trusting week-one readings (Stacker).
- Track mention rate and citation rate separately. A release can raise citation rate on brand prompts without changing mention rate on category prompts. The two metrics really do diverge: our scan of notion.so found Notion named in 92% of answers but cited in 3%.
- Treat editorial pickup as the leading indicator. No pickup in the first few weeks means, on Everything-PR's numbers, low citation odds.
Geoptimizer is built for this. It runs every tracked prompt on ChatGPT, Gemini, Claude and Grok and reports mention and citation rate per engine over a 7-day rolling window with a confidence band, so a post-release bump isn't mistaken for noise. You can mark the release on the trend chart and see whether the two prompt groups move differently. The limits, stated plainly: Geoptimizer doesn't track Perplexity, the engine most likely to cite wire URLs directly; none of the studies we found covered how Grok treats press releases; and a tracker shows correlation, not proof of cause.
FAQ
Do press releases help AI visibility for B2B SaaS companies?
On branded and news prompts, yes, especially for hard news like funding or M&A. On "best X software" prompts, rarely. In the B2B software studies above, listicles and review sites take most of those citations, and news and trade press made up 1.2% in DerivateX's AI Overviews sample.
Does ChatGPT cite press releases?
Sometimes, and mostly through copies other than the wire version. In Everything-PR's study, ChatGPT cited 28 releases, and 26 of those citations went to editorial copies rather than wire URLs. BuzzStream found that ChatGPT leans unusually heavily on companies' own newsroom pages, at 18.15% of its citations.
How long does it take for a press release to get cited by AI?
It depends on how you count. Notified reports an average of about eight hours to first citation for releases that were cited. Everything-PR found 8% of its releases cited by day 30, 14% by day 60 and 19% by day 90, and releases still uncited after 60 days were unlikely to be cited later.
Is a newswire worth the cost for GEO?
If the news is real and you'll pitch it for editorial coverage, it can be. If it's an award, hire or partnership announcement, the data suggests it isn't. List prices start around $850–$940 for a national release, and Everything-PR's illustrative all-in estimate comes to roughly $16,000–$26,000 per release that actually gets cited.
The bottom line
The 99% and the zero answer different questions. Wire releases get cited quickly on news, trend and brand prompts, and most often through journalists' copies of the story. They rarely reach the category prompts where B2B buyers make a shortlist. Spend wire budget on real news, put the shortlist budget into data and earned coverage, and measure the two prompt groups separately. If you want to see which of your prompts a release actually moves, start with Geoptimizer's free plan. It covers all four engines, and you don't need a credit card.