Press release distribution for AI companies: what you can claim
· 6 min read
Two sentences that appear in most AI launch releases are now checkable in a way they were not a year ago, and both are worth fixing before you distribute. Article 50 of the EU AI Act began applying on 2 August 2026, so a release describing a synthetic-media or emotion-inference product without mentioning disclosure is inconsistent with the regulation that covers the product it announces. And "NIST AI RMF certified" describes a document that does not exist: the framework is voluntary and certifies nobody.
The EU transparency duties started applying this month
Article 50 of the EU AI Act began applying on 2 August 2026. It obliges providers to ensure that people are informed they are interacting with an AI system, unless that "is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect." It obliges providers of systems generating synthetic audio, image, video or text to ensure outputs are "marked in a machine-readable format and detectable as artificially generated or manipulated," where technically feasible. And it obliges deployers of emotion-recognition and biometric-categorisation systems to inform the people exposed to them.
The relevance to a press release is direct rather than incidental. A launch release is frequently the first public description of what a system does, who it addresses and how it labels its output. If the release describes a synthetic-media product without mentioning provenance marking, or describes an emotion-inference feature without mentioning disclosure, the release is now inconsistent with a regulation that applies to the product it announces. That is a bad document to have published under your own dateline, and no distributor will flag it — a wire reviews for its own risk, not yours.
This is not legal advice. Whether Article 50 reaches your system, and in what role, depends on the Act's definitions and on where you place the product; the version that matters is the one your counsel signs.
"NIST AI RMF certified" describes a document that does not exist
The second claim is a governance one. NIST's AI Risk Management Framework is, in NIST's own words, "intended for voluntary use." It is guidance for managing risks to individuals, organisations and society. NIST publishes no conformity-assessment scheme against it and issues no certificates. There is nothing to be certified to.
"Aligned to the NIST AI Risk Management Framework, with our mapping published at /trust" is a claim a reader can check. "NIST-certified AI" is a claim about a credential that has no issuer. The same applies to "ISO 42001 certified" where no accredited certificate exists, and — a habit carried over from the wider software sector — to "SOC 2 certified," which describes an examination report rather than a certification.
Nothing here is a wire rule. Every self-serve desk in this survey will publish either sentence without comment. The exposure is with procurement teams, with journalists who now ask for the certificate, and with competitors.
The awkward finding about AI answers
The commonest reason an AI company is sold a press release in 2026 is that the release will supposedly train or feed the assistants. It is worth being straight about the evidence.
Press releases are a small and contested share of what assistants cite. Muck Rack's continuing analysis of links cited by AI models puts the three tier-one wire domains at around one per cent of citations, and Meltwater reported press-release citation share in AI search falling between April and May 2026, halving month on month. We are citing both as reported rather than measured: Meltwater's figure comes to us second-hand, and Muck Rack's underlying report sits behind a download form. Treat the direction as informative and the decimal places as not.
What we can say from our own August 2026 sweep is narrower and more useful: some syndicated pages are retrievable and some are not, and the difference tracks whether the outlet keeps a native permalink or serves the release from a shared hub that assistants rarely surface. We wrote that up in do press releases show up in AI answers, including the outlets where we found nothing at all.
If discoverability inside assistants is the reason you are buying, buy the outlets whose pages are retrievable and check them afterwards. Do not buy an outlet count.
What the wires do and do not restrict
AI is not a restricted category anywhere in this survey. No newswire publishes a rule against announcing a model, a funding round for an AI company, or a benchmark result. Two adjacent rules do bite.
The first is that PRWeb refuses venture funding announcements — "content related to venture capital announcements -- e.g., 'series A funding,' etc." — and refers them to PR Newswire membership. Given how much AI news is funding news, this catches a lot of releases at the cheapest self-serve door.
The second is the benchmark claim. No wire polices it, but a release stating a leaderboard position without naming the benchmark version, the evaluation harness and the date is a claim that ages badly in public. Name the eval and the date, or state the result as internal.
The links, since someone will ask
Almost every syndicated press link is nofollow or rel="sponsored", and Google's own documentation says sponsored marks "advertisements or paid placements". We measured this per outlet rather than per publisher and found the treatment varies within a single publisher: TechBullion carries a plain dofollow link on business releases and rel="sponsored" on crypto ones. If an AI company is buying distribution on a search-authority argument, read why that argument is usually wrong first, and check any placement yourself in about ninety seconds.
What we sell into this sector
We are a reseller: distribution bought from wire partners and marked up, which is the ordinary structure of this market. Our Headline package is $899 and Launch is $599. What we add is that we open every published URL and report what went live, marked as measured or as supplier-reported, rather than forwarding a pickup list. We do not promise rankings, permanent pages, or that any named outlet will publish.
For an AI company, the honest framing of what a syndication buy does is: it produces a dated, citable record of an announcement on third-party domains, and a set of logos you may reference accurately. It does not produce trade-press coverage, and it does not produce placement in an assistant's answer. What a press release cannot do is the longer version of that sentence.
What we could not establish
No wire publishes an AI-specific content policy — not one, across the seven networks surveyed in August 2026. Given how quickly the regulatory position is moving, the absence is worth noting rather than reading as permission: policy silence means the decision is made case by case at an editorial desk, and it can change between your first release and your second.
Where these figures came from
- EU AI Act, Article 50 (transparency obligations) — the duty to inform natural persons they are interacting with an AI system, the machine-readable marking requirement for synthetic content, and the 2 August 2026 application date: https://artificialintelligenceact.eu/article/50/
- NIST AI Risk Management Framework — "intended for voluntary use"; NIST publishes no conformity assessment or certification against it: https://www.nist.gov/itl/ai-risk-management-framework
- PRWeb Editorial Guidelines — venture capital and Series A announcements refused and referred to PR Newswire membership: https://www.prweb.com/editorial-guidelines/
- Google Search Central, "Qualify your outbound links to Google" — rel="sponsored" marks advertisements or paid placements: https://developers.google.com/search/docs/crawling-indexing/qualify-outbound-links
- Muck Rack research on links cited by AI models, and Meltwater’s April–May 2026 press-release citation-share figure — both REPORTED, not measured by us. Muck Rack’s full report is gated behind a download form: https://muckrack.com/resources/research/state-of-journalism
- AICPA, SOC 2 — an examination of controls reported against trust services criteria, not a certification: https://www.aicpa-cima.com/topic/audit-assurance/audit-and-assurance-greater-than-soc-2
- Link attributes measured per outlet in our August 2026 placement sweep, including TechBullion’s differing treatment of business and crypto releases.
We distribute press releases to 300+ outlets, then open every published link and report what actually went live.
View packages