AI Visibility

Do Press Releases Help With AI Visibility? What 25 Million AI-Cited Links Reveal

Muck Rack classified 84% of its May 2026 cited-link sample as sources brands neither owned nor paid for. Press releases show up mainly in industry-trend answers, not as a general buyer-intent citation engine.

Jaxon ParrottMar 1, 2026

Press releases can help AI visibility when they give engines a fresh, factual notice about a real announcement. They are weak evidence for the broader job most PR budgets are judged on: shaping recommendation, comparison, and buyer-intent answers.

Muck Rack's May 2026 Generative Pulse article says its team analyzed more than 25 million links cited by ChatGPT, Claude, and Gemini responses across 17 industries. In that observed cited-link sample, Muck Rack classified 84% as earned media: sources brands neither owned nor paid for. Boundary: Muck Rack's 84% figure measures a broad non-owned, non-paid source category in its observed cited-link sample; it is not a journalism-only share, a universal share of every AI citation, a provider-selection mechanism, a recommendation guarantee, or a pipeline result.

The answer to "do press releases help with AI visibility?" is therefore: yes for specific announcement and industry-trend contexts, but usually as a notice layer or catalyst for coverage, not as the main authority layer. The operating question is not whether to publish releases. It is what job you expect them to do, and what separate evidence proves that job happened.

What Muck Rack Actually Measured

The strongest public source for this page is Muck Rack's May 2026 "What Is AI Reading?" update. Its measured unit is a cited link/source category inside observed AI responses, not a brand outcome.

Key source boundaries:

Population: more than 25 million cited links from ChatGPT, Claude, and Gemini responses across 17 industries in the May 2026 edition.
Taxonomy: Muck Rack's 84% label is the broad earned-media category — sources brands neither own nor pay for. It includes journalism but is not the same as journalism.
Journalism subset: Muck Rack reports journalism alone at 27% in the May 2026 cited-link sample, with journalism citations ranging from 25% to 27% across three editions.
Press-release unit: Muck Rack's FAQ says press releases include releases published over the wire, on a brand's own website, or syndicated by another news outlet. That definition measures the content type wherever it appears; it does not isolate wire-service distribution as a provider mechanism.
Trend-query behavior: Muck Rack's FAQ says about 1% of industry-trend responses contain a press-release citation, about 3.5x the rate for best-of questions.

That is useful operating evidence. It says press releases appear more often for trend discovery than for comparative buying questions. It does not say a release, a wire package, or a syndication bundle will make an engine recommend a brand.

How AI Engines Treat Sources: What the Research Supports

The EMNLP 2025 paper cited below is not a PR study. It is a controlled study of political-news citation generation. The researchers tested source labels by removing media-source identifiers and by swapping source names between left- and right-leaning articles while holding article text constant. Their finding supports a narrower claim: in that political-news citation experiment, source identity had a large effect on citation behavior. Boundary: the EMNLP paper does not study press releases, wire services, buyer-intent PR queries, brand recommendations, revenue, or a production-engine ranking rule.

A separate arXiv inventory by Kai-Cheng Yang analyzes AI Search Arena data: more than 24,000 conversations, 65,000 responses, and more than 366,000 citations from systems across OpenAI, Perplexity, and Google. About 9% of those citations referenced news sources, and those news citations concentrated among a small number of outlets. Boundary: the 366,000-citation inventory is a news-source citation study across arena conversations; it does not measure press-release performance, earned-media ROI, or a brand's probability of being recommended.

Together, those studies support disciplined source planning: know which source types and outlets appear in the answers that matter. They do not support deterministic language like "AI engines trust X," "wire syndication is spam," or "a tier-one placement creates recommendations."

What the Press-Release Evidence Says

Press-release visibility is real but bounded. In Muck Rack's December 2025 report, press releases rose from 0.2% in July 2025 to about 1% of cited links by December 2025. In the May 2026 FAQ, press releases are described as disproportionately present in industry-trend answers: about 1% of those responses contain a press-release citation, roughly 3.5x the rate for best-of questions.

That means the best press-release use case is narrow:

Publish factual notices for announcements engines may need to resolve: funding, leadership, product availability, research, partnerships, policy, compliance, or incident response.
Include dates, named entities, product names, figures, geographies, and source links that can be extracted without interpreting marketing copy.
Use the release as one input to journalist outreach, analyst updates, owned documentation, and entity-cleanup work.
Measure whether the release is indexed, retrieved, cited, and accurately represented before calling it an AI-visibility win.

Boundary: Muck Rack's press-release findings measure observed citation/source-type presence in its samples; they do not establish that press-release syndication volume causes citation, that duplicated syndicated copies create source diversity, or that a paid wire placement improves recommendation, referral, pipeline, or revenue.

Which Sources Show Up Most Often?

Muck Rack and Nieman Lab both show concentration, but the unit matters. Nieman Lab reported on Muck Rack's July 2025 data and found that journalistic content appeared in more than 27% of citations overall and 49% of citations for recency-implying queries. Reuters, the Financial Times, Time, Forbes, and Axios appeared in both ChatGPT and Gemini top-outlet lists, and Axios appeared in ChatGPT's top three cited domains across 13 of 17 industries.

The May 2026 Muck Rack update also reports platform-level differences:

AI platform	Citation inclusion in Muck Rack sample	Average cited links per response	Top cited domain named by Muck Rack	Operating implication
ChatGPT	96% of responses	5	Wikipedia	Test buyer and trend prompts here first if your audience starts with ChatGPT.
Gemini	82% of responses	8	Reddit	Watch community and video-adjacent surfaces as well as editorial coverage.
Claude	55% of responses	13 when it cites	PubMed Central	Expect stronger research, technical, health, government, and institutional source pull in some categories.

Boundary: these are Muck Rack sample rates and domain observations, not stable guarantees for every prompt, model version, geography, or industry.

Press Releases vs Earned Coverage vs Owned Content

A practical AI-visibility program separates source type from outcome:

Asset type	What the evidence supports	Best AI-visibility job	What it does not prove
Earned coverage and third-party coverage	Muck Rack's broad non-owned/non-paid category is 84% of its May 2026 cited-link sample; journalism alone is 27%.	Build an external evidence trail in outlets and industry sources engines already cite.	No universal share, causality, recommendation lift, pipeline, or revenue.
Press releases	Press-release citations are most visible in industry-trend answers and were about 1% of cited links in Muck Rack's December 2025 report.	Publish dated, factual notices and provide material journalists and engines can verify.	No proof that syndication volume creates authority or buyer-intent recommendations.
Owned content	Muck Rack includes owned first-party corporate/blog content as a separate category in its taxonomy.	Clarify entities, claims, product facts, documentation, FAQs, and measurement pages.	No proof that owned pages replace external validation in answers that cite outside sources.
Original research	The KDD 2024 GEO paper found content changes such as adding statistics, citations, and quotations can improve visibility inside its benchmark setup, with up to 40% lift overall.	Give journalists, analysts, customers, and AI systems specific facts worth citing.	No current commercial-engine citation-rate guarantee, no universal 3x/10x multiplier, and no ROI proof.

What Actually Builds AI Citation Readiness?

Treat AI visibility as a measurement pipeline, not a slogan.

Publication: Did the release, owned page, research asset, or earned article publish at a stable URL?
Access and indexing: Can crawlers reach it, and does the page expose clean title, date, author, entity, and canonical metadata?
Retrieval: Does the source appear when you test the prompts, engines, dates, and geographies that matter?
Citation: Is the URL cited, and for which prompt class — trend, how-to, comparison, recommendation, or brand-specific?
Claim support: Does the AI answer use the source to support a correct claim?
Recommendation: Does the answer include, exclude, or position the brand in a shortlist?
Commercial outcome: Do referral, conversion, pipeline, and revenue systems connect the AI session to a business result?

Publication, access/indexing, retrieval, citation, claim support, recommendation, referral, conversion, pipeline, and revenue are separate measurements. A press release can succeed at the first two and fail at buyer-intent recommendation. Earned coverage can appear in a citation and still not move pipeline. The point is to instrument the chain instead of collapsing it into one percentage.

How to Write Press Releases for AI Extraction Without Overclaiming

Useful press releases still matter. Write them so machines can resolve facts and humans can verify them:

Put the announcement, date, entity names, product names, location, and primary figure in the first two paragraphs.
Add an FAQ only when it answers real questions: availability, pricing, eligibility, compliance, methodology, or how the announcement changes a category.
Include original data only when you can publish the methodology, denominator, field dates, geography, sample, and limitations.
Link to durable supporting pages: product docs, research methodology, regulatory filings, newsroom assets, customer proof, or executive bios.
Keep syndicated copies canonical to the same fact pattern; do not treat hundreds of duplicate republications as independent evidence.

The GEO paper supports structured, evidence-rich content only within its benchmark experiment. Muck Rack's FAQ supports practical formatting because factual content gives AI systems more usable material. Boundary: neither source proves that a press-release format, an FAQ, schema, or original-data asset will be cited or recommended by a specific production model.

How to Evaluate a 2026 PR Budget for AI Visibility

Allocate budget by measured job:

Keep funding press releases when they are needed for communications operations: regulatory disclosure, investor relations, major launches, funding, security incidents, leadership changes, partnerships, research releases, or category education.

Do not buy wire distribution as the whole AI-visibility strategy: Muck Rack's data shows press releases appear in a narrow slice of observed AI responses, especially trend queries. The evidence does not support routine wire spend as a buyer-intent citation or recommendation engine.

Fund earned outreach where AI answers already cite the outlet or source class: use prompt logs to identify which publications, trade sources, analyst pages, communities, research repositories, and reference pages appear for your actual category.

Fund owned content where the answer needs entity clarity: comparison pages, changelogs, documentation, methodology pages, executive bios, product pages, support articles, and citation-ready FAQs can correct facts even when they are not the primary citation source.

Measure ROI only at the right layer: a release impression, a pickup count, a cited URL, an answer mention, a referral visit, a demo request, and closed pipeline are different events. A PR budget decision should name which layer moved and which did not.

The Machine Relations Framework

AI search is now part of B2B discovery. Buyers use AI systems to summarize markets, compare vendors, and explain categories. The sources those systems cite influence which claims and brands become visible.

Machine Relations is the practice of making that evidence legible to machines: earning coverage where it matters, publishing original support where you own the facts, keeping entity data clean, measuring prompt-level visibility, and separating citations from recommendations and revenue.

For press releases, Machine Relations does not mean "publish more releases." It means use releases for factual notice, use earned coverage for independent validation, use owned pages for entity clarity, and measure each layer before claiming AI visibility impact.

Start your AI visibility audit to see which sources AI engines cite for your category, where your brand appears, and whether press releases, owned pages, or earned coverage are doing the job you think they are doing.

FAQ

Do press releases help with AI visibility?

Yes, but narrowly. Muck Rack's May 2026 FAQ says press releases appear most often in industry-trend answers, with about 1% of industry-trend responses containing a press-release citation and about 3.5x the rate for best-of questions. That finding includes releases published on wires, owned sites, or syndicated outlets; it does not prove that wire distribution volume creates buyer-intent recommendations.

How large is Muck Rack's broad earned-media category?

Muck Rack classified 84% of cited links in its May 2026 ChatGPT, Claude, and Gemini sample as broad earned media: sources brands neither owned nor paid for. Journalism alone was 27%, and journalism stayed between 25% and 27% across three editions. Boundary: the 84% number is Muck Rack's observed cited-link taxonomy, not a universal journalism-only share or a revenue claim.

Which AI platform cites the most sources?

In Muck Rack's May 2026 sample, Claude cited sources in 55% of responses but averaged 13 cited links when it cited. Gemini cited in 82% of responses with an average of 8 cited links. ChatGPT cited in 96% of responses with an average of 5 cited links. Those are sample measurements, so teams should retest their own prompt sets.

Which publications do ChatGPT and Gemini cite most?

Nieman Lab's reporting on Muck Rack's July 2025 data said Reuters, the Financial Times, Time, Forbes, and Axios appeared in both ChatGPT and Gemini top-outlet lists. Axios appeared in ChatGPT's top three cited domains across 13 of 17 industries. That concentration supports targeted source mapping, not a guarantee that a pitch to one outlet will produce an AI citation.

Why do AI engines often cite editorial coverage instead of press releases?

The evidence supports a cautious answer: Muck Rack observed far more broad earned and journalistic cited links than press-release cited links, and an EMNLP 2025 political-news experiment found source identity strongly affected citation behavior when source labels were removed or swapped. Those findings do not disclose a universal provider mechanism. They do justify investing in independently verifiable sources, clean owned facts, and prompt-level measurement.

How long does it take to build AI citation authority?

There is no source-backed universal timeline. Press releases can publish and index quickly, but citation and recommendation effects must be tested by prompt, engine, geography, and date. Muck Rack's journalism analysis suggests freshness can matter — many journalism citations were from the previous 12 months, and citation rates were strongest shortly after publication before tapering — but that is not a promise of impact after a fixed number of months.

Sources

Muck Rack. "Earned media still drives 84% of AI citations. Here's what that means for PR." Muck Rack Blog, May 7, 2026. https://muckrack.com/blog/what-is-ai-reading-may-2026. Boundary: this Muck Rack title uses earned media for sources brands neither own nor pay for in its May 2026 cited-link sample; it does not establish a journalism-only share, provider mechanism, recommendation outcome, pipeline, or revenue.
Muck Rack. "7 FAQs about the May 2026 update of ‘What Is AI Reading?’ a study of 1 million+ links cited by AI models." Muck Rack Blog, June 4, 2026. https://muckrack.com/blog/what-is-ai-reading-new-insights-may-faqs
Muck Rack. "What is AI reading?" Generative Pulse, May 2026. https://generativepulse.ai/report/
Dai, Sunhao, et al. "Media Source Matters More Than Content: Unveiling Political Bias in LLM-Generated Citations." Proceedings of EMNLP 2025, Association for Computational Linguistics. https://aclanthology.org/2025.emnlp-main.872.pdf
Yang, Kai-Cheng. "News Source Citing Patterns in AI Search Systems." arXiv, July 7, 2025. https://arxiv.org/abs/2507.05301
Deck, Andrew. "Generative AI models love to cite Reuters and Axios, study finds." Nieman Journalism Lab, July 24, 2025. https://www.niemanlab.org/2025/07/generative-ai-models-love-to-cite-reuters-and-axios-study-finds/
Muck Rack. "What Is AI Reading? December 2025." Generative Pulse, December 2025. https://media.muckrack.com/static/reports/2025/MuckRack-GenerativePulse2025-1.pdf
Aggarwal, Pranjal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande. "GEO: Generative Engine Optimization." KDD 2024 / arXiv:2311.09735. https://arxiv.org/abs/2311.09735
Machine Relations. "Machine Relations." https://machinerelations.ai/glossary/machine-relations
AuthorityTech. "AI Visibility Audit." https://app.authoritytech.io/visibility-audit