Press Releases and AI Citations: The Data PR Agencies Won't Show You
Press releases account for under 1% of all AI citations, and on the Machine Relations Index Forbes is cited roughly 3.65x as often as PR Newswire. Here is what the data says about what AI engines actually cite and what PR agencies are not telling you.
The PR industry is racing to sell you AI visibility through press release distribution. At least six agencies and platforms published guides on this in July 2026 alone. They are selling you the wrong thing. The data shows press releases account for under 1% of all AI citations, and on the Machine Relations Index, Forbes is cited roughly 3.65x as often as PR Newswire.
Forbes Outpaces PR Newswire on the Machine Relations Index
Measured on Machine Relations Index release mri_score_v2.0+2026-09-20+47973f373a20 (127-day window, 16,039 answer runs, six engines), forbes.com is cited in 665 answer runs, a 4.15% share, against prnewswire.com's 182, a 1.13% share. Forbes is cited roughly 3.65x as often as PR Newswire. Among classified publications, forbes.com ranks 2 of 1,230 and prnewswire.com ranks 8.
How to read the editorial-class rank. The editorial-class ranks above (Forbes at rank 2 of 1,230, PR Newswire at 8) are read off the Machine Relations Index's source_role field, which in the current release combines source-type evidence with AuthorityTech's placement catalog; 96 of the 1,231 class members are admitted through the catalog, and the Index is moving editorial classification to source-type evidence alone. The class currently includes six platform domains alongside edited newsrooms: medium.com (class rank 1), amazon.com (7), prnewswire.com (8), apple.com (16), wordpress.com (239) and blogspot.com (330). At the same time substack.com, beehiiv.com and dev.to — the same hosted open-publishing product as Medium — are filed as community platforms. Those six hold 1,358 of the editorial class's 13,525 cited runs, 10.04%, and they include the class's top slot. The class denominator also moves between releases (published figures carry 1,025 to 1,230) because it counts the domains classified in each release. Independent of class membership, citation rates, cited runs, days cited, engine breadth and confidence are direct per-domain observations over a fixed run set, and are the primary measure; read the rate first.
That finding undercuts the pitch. PR agencies have been telling founders that press release distribution is now an AI visibility strategy. The pitch sounds compelling. The logic fails twice over: PR Newswire does not outpace Forbes on cited answer runs, and even where a wire domain's raw citation count looks large, it is because its content is structurally extractable, company name, specific claim, date, named source in every sentence, not because distribution itself creates visibility. The wire and press-release distribution class as a whole holds 232 of 112,518 cited-domain observations on the current release, or 0.21% of classified citation. The current release files prnewswire.com, the largest wire domain, as an editorial publication, so that nine-domain class excludes it. Counted with prnewswire.com as wire distribution, the class is ten domains holding 414 of 112,518 cited-domain observations, or 0.37% of classified citation. On either count the finding holds: prnewswire.com's strongest standing inside any published segment is #15 of 448, which leaves the wire class outside every segment's top ten.
What the Broader Data Actually Says
When you zoom out from per-domain counts to share of all AI citations, the picture inverts.
BuzzStream and Xofu analyzed 548,000 retrieved pages across 15,000 prompts and found that syndication accounts for just 6% of AI citations. Newswires specifically come in under 1%. The top 10 domains capture 46% of all citations within a given topic. The top 30 own 67%. The measured unit is source composition in the cited sample. It shows what engines cited, not why, and does not guarantee a future citation.
Presenc AI's citation-source analysis across approximately 5,200 monitored announcement queries found that syndicated press releases appearing on newswire pickup sites are cited directly in AI answers less than 8% of the time.
The disconnect is not a contradiction. It is two different questions. How many citations does a wire domain collect in absolute terms? A modest but real number. What percentage of all AI answers cite press releases? Almost none.
One measures the domain. The other measures the format. A raw per-domain count without a denominator will always favor whoever publishes the most URLs; the format-level share is what determines whether AI engines cite you at all.
What AI Engines Actually Extract From Press Releases
The GEO paper (Aggarwal et al., KDD 2024) paper published in SIGKDD 2024 quantified the mechanism: adding specific statistics to content improves AI engine visibility by 30% to 40%.
Wire service content produces structured facts by default. Company announced X. Revenue was Y. Partnership with Z. Every sentence contains a named entity and a verifiable claim. That is what AI retrieval systems extract.
But here is what PR agencies do not tell you: the structure, not the distribution, is doing the work. A well-structured page on your own domain with specific claims, named entities, and verifiable data points would perform the same extraction function without paying $1,500 per release to a wire service.
Presenc AI's data confirms this. Each independent editorial article generated from a press release creates an estimated 0.6 to 1.1 additional AI citation sources per platform. Five earned articles from a single release can add three to five incremental citation pathways per platform. Trade publication coverage drives 2.8x to 4.2x more AI citations than the same information on a newswire pickup page.
The release is the seed. The editorial pickup is the crop. Most firms are counting seeds and calling it a harvest.
The Real Split: Distribution Value vs. Citation Value
Here is the breakdown that should be on every founder's desk before they sign a PR retainer:
| Source Type | Direct AI Citation Rate | Citation Durability | Real Value |
|---|---|---|---|
| Raw newswire syndication (no editorial pickup) | Under 8% | Low | Minimal AI visibility return |
| 1 to 2 trade editorial articles from release | Low for release, moderate for articles | Moderate | Covers niche queries |
| 3 to 5 independent editorial articles earned | Low for release, high for articles | High | Strong citation lift |
| Tier-1 national or major trade article earned | Low for release, very high for article | Very high | Durable across all platforms |
| Data-led release with proprietary statistics | Moderate (data sourced directly) | High | Statistics cited even without the article |
Data-led releases with original statistics get picked up editorially at 2.7x the rate of standard announcements like funding rounds, new hires, or product launches without data. The statistics themselves get cited independently of whether the full article is retrieved.
This is the move most PR agencies miss entirely. They optimize for placement volume. The AI engine optimizes for extractable claims.
Why This Is a Machine Relations Problem, Not a PR Problem
Traditional PR measures placements. How many outlets picked up the release. How many journalists opened the pitch. How many impressions the coverage generated.
None of those metrics tell you whether an AI engine will cite your brand when a buyer asks a category question.
Machine Relations measures citations: how often AI answer engines reference your content when generating responses to buyer-intent queries. The distinction between "getting placed" and "getting cited" is the entire category.
Muck Rack's "What Is AI Reading?" study found that over 95% of AI citations come from non-paid channels, with 85% of those coming from earned sources. Their May 2026 update put the non-owned, non-paid share at 84% across 25 million analyzed links. Earned coverage is what fills those citation slots. Syndicated release text — identical copy pushed to hundreds of outlets — barely registers in the same samples.
The PR firms telling you otherwise are measuring the wrong output. They are optimizing for a metric that stopped being the one that matters.
What to Do Instead of Buying More Wire Distribution
Stop measuring press releases by pickup count. Start measuring by citation rate.
Here is the founder playbook:
Embed original data in every release. Statistics get cited independently. "Our platform processed 2.3 million queries in Q2" is extractable. "We are excited to announce our latest innovation" is not. The Aggarwal et al. research proved this: specific statistics improve AI visibility by 30% to 40%.
Optimize for editorial pickup, not syndication volume. Five earned articles from a single release add three to five citation pathways per AI platform. Two hundred identical syndication pages add almost nothing. 86% of journalists say at least some of their stories originate from PR pitches, but original research and data are among the most valuable elements that make a pitch land. Give them the data.
Build structured content on your own domain. A well-structured page with named entities, specific claims, and verifiable data creates the same extraction signal that makes wire services show up in citation counts. You do not need a wire to be structurally extractable.
Measure citation architecture, not media impressions. Track how often your brand appears in AI-generated answers to category queries. That is the metric that determines whether buyers find you through ChatGPT, Perplexity, Gemini, or Google AI Mode.
FAQ
Do press releases get cited by AI search engines?
Rarely as direct citations. Syndicated press releases are cited directly in AI answers less than 8% of the time. Newswires account for under 1% of all AI citations by volume. The value of a press release for AI visibility comes almost entirely from the earned editorial coverage it generates, not from the syndicated text itself. The measured unit is source composition in the cited sample. It shows what engines cited, not why, and does not guarantee a future citation.
Why does PR Newswire show any AI citation counts if press releases are not effective?
A wire domain's citation count is a function of publishing volume, not distribution effectiveness: named entities, specific claims, and verifiable data in every release make each one easy for AI systems to extract. That structural advantage is a property of the format, not proof that distribution drives citations. As a share of classified citation, the entire wire and press-release class holds 0.21% on the current Machine Relations Index release as it classifies its members, or 0.37% once PR Newswire is counted as wire distribution rather than as an editorial publication — either way well behind editorial publications like Forbes.
What type of press release content gets cited by AI engines?
Releases containing original data, proprietary statistics, or survey results get picked up editorially at 2.7x the rate of standard announcements. The statistics themselves are cited independently of whether the full article is retrieved by AI engines.
Should founders still invest in press release distribution for visibility?
Only if the distribution is a means to earned editorial coverage, not an end in itself. The release should carry original data and be structured for journalist pickup. Measuring success by syndication page count is measuring the wrong output. Measure by earned articles generated and downstream AI citation rate per platform.
What is the difference between PR and Machine Relations?
Traditional PR measures placements: outlet coverage, journalist opens, impression counts. Machine Relations measures citations: how often AI answer engines reference your brand when generating responses to buyer-intent queries. The shift from measuring placements to measuring citations is why Machine Relations exists as a discipline.
Updated 2026-09-23: figures reflect the current Machine Relations Index methodology; see the MRI methodology and update log.