Press Releases for AI Citations: When It's Worth It

Search the subject and you will find two sets of numbers pointing in opposite directions. Newswire suppliers publish material saying the overwhelming majority of their releases are indexed or accessible to AI platforms. Independent analyses of which sources assistants actually name put press releases in the low single digits as a share of cited sources. Both can be stated accurately and still describe entirely different things.

Three claims get used interchangeably, and separating them is the first useful step.

Crawled and indexed. The release is reachable, parseable and present in a corpus or a live search index. That is a distribution property. It says nothing about whether anything will ever retrieve it.

Retrieved. At query time, the system pulls a set of candidate passages. Your release is in the pool for a specific question. Still no citation.

Cited. The answer names the source, and the user can see it. That is the only one a buyer ever notices.

Indexing is not citation. A release can be crawled by every platform in the market and never be named, because retrieval ranks candidates at the moment the question is asked, and it usually prefers the independent write-up to the originating copy. Read any supplier claim about AI indexing as a reach metric until it is proven at answer level.

Share-of-citation figures carry their own caveat. They depend entirely on the prompt set used. A release about a funding round competes against business press, Companies House data and investor databases. A release about a product specification competes against documentation, review sites and forum threads. Change the prompts and the share moves. That is why two credible studies can disagree by an order of magnitude without either being dishonest.

What to ask your newswire supplier

Ask for a like-for-like answer rather than a reach number. Four questions get you there: which platforms were checked, what prompts were used, how many answers named the release itself as a cited source rather than a downstream article, and over what period. If the response comes back as an indexing percentage or an estimated audience figure, you have been given distribution data dressed as visibility data.

Syndication actively works against you. One release becomes forty or fifty near-identical copies across aggregator domains of varying quality. From a retrieval point of view there is no obvious canonical version, the duplicate text competes with itself, and the strongest-looking copy is often on a domain nobody trusts or links to. Distribution breadth and source clarity pull in opposite directions here, and clarity wins.

Then there is the copy itself. A release built from positioning language, with no named spokesperson, no dated event and no numbers, gives a model nothing to extract and nothing to attribute. Assistants quote facts. If the only quotable sentence is an adjective about market leadership, you will not be quoted.

Entity confusion does quiet damage. The company is "Acme Ltd"in the boilerplate, "Acme"on the website, "Acme Technologies"in a directory and something else again on LinkedIn. No single durable page says plainly what the business does and who it serves. Models resolve ambiguity by picking whichever description is most consistent across independent sources, and inconsistency is how you end up in the wrong category.

Technical issues finish the job. Newsrooms that render only through JavaScript, sit on a subdomain with no internal links, or clear releases after ninety days all remove the asset from retrieval. Undated or re-dated copies add a further problem: a claim that cannot be pinned to a date looks unverifiable, and freshness matters in categories where assistants prefer recent sources.

Write for extraction, not for applause. The first 150 words should carry a named spokesperson with their role, a dated event, at least one specific number, and a plain-language description of the product or service that a reader outside your sector would understand.

Include one original data point or definition that the rest of the web does not already hold. A survey finding, a benchmark, a price, a specification, a defined term. Without it, your announcement is paraphrasable, and anything paraphrasable gets attributed to whoever wrote the better version.

Host the authoritative version on your own site and point the wire copies back to it. Add structured data so the publisher, date, author and organisation are machine-readable. Use the same entity name everywhere: the release, the About page, directories, professional bodies, social profiles. Clarity and consistency sit at the centre of the 6 Pillars of AI Visibility we work to, and they cost nothing except discipline.

Finally, give the announcement somewhere permanent to point. A reference page, a methodology page, a product specification page, a data explainer. News cycles end in days. Reference pages get retrieved for years.

Retrieval tends to favour independent write-ups, trade titles and reference pages over the originating release. That is the central practical point. The release is a mechanism for producing citable sources. It is rarely the citable source itself.

So target the places that already shape answers in your category. UK trade press, sector round-ups, professional body publications, niche newsletters and the community discussions your buyers actually read. Three trade titles covering the same facts independently carry far more weight than one release sitting on fifty aggregator domains.

Source weighting also differs by platform. The same release can appear behind Perplexity answers and never surface in ChatGPT or Gemini, because each assistant weights freshness, domain trust and discussion differently. Teams who check one assistant and draw a conclusion are how the industry ended up with two confident, opposing views on whether press releases work.

Citation mixes move as well. Reddit's share of citations inside ChatGPT was reported to fall sharply after a long period of prominence, and the exact numbers vary by tracker and prompt set. The direction is the lesson: a dominant source type can lose weight quickly. Spreading announcement coverage across trade press, owned reference pages and third-party profiles is insurance against that.

Measurement needs a baseline. Before the release goes out, run a fixed prompt set across ChatGPT, Gemini, Claude, Perplexity, Copilot and Meta AI, and record two things: who is named, and which domains are cited. Without that record, any change afterwards could just as easily be a model update.

Track on a schedule rather than spot-checking. Answers vary between sessions, so a single screenshot proves very little. Daily monitoring across the same prompts is what turns anecdote into a trend you can report.

Then run source attribution: list the domains the assistants actually cite for your category and check whether your newsroom, your reference page or the resulting coverage appears. If your release produced a trade article that is now cited, the release worked, even though the wire copy is invisible.

Allow a realistic lag. Several weeks to a few months is normal between publication and any movement in cited sources, depending on how quickly the coverage is indexed and how often the category is queried. Report share of voice and source mix. Pickup counts and potential audience reach tell you nothing about whether a buyer sees you in an answer. Our monthly AI visibility monitoring exists for exactly this pattern of before-and-after tracking, and you can also run a free scan to see where you currently stand.

Compare honestly. Set the cost of one wire distribution against the same budget spent on original research you can publish, a well-built reference page, or direct outreach to three trade editors. In categories where assistants lean on specialist publications, the outreach and the research usually produce more citable material than the circuit does.

A release is genuinely the right call for funding, senior appointments, acquisitions, regulatory news, launches with real specifications, and original survey data. Each of those carries facts other people want to repeat.

A release is the wrong call for a rebrand with no new facts, an award entry, or an attempt to correct what an assistant says about you. That last one matters. Press releases cannot fix misrepresentation on their own. DarkInvader was initially placed in the wrong category by AI answers, and the correction came from the description, the structure of the pages describing the business, and the third-party sources the models were relying on. The announcement only helps once the underlying sources are right. If your problem is a wrong category or an outdated description, start with an audit of how assistants currently describe you rather than a distribution spend.

One governance point. Claims pushed out at scale need to stand up, because they get copied, quoted and eventually attributed. Organisations already working to the ICO's guidance on artificial intelligence and data protection or the NIST AI Risk Management Framework should extend the same evidence discipline to announcement claims: every number in a release traceable to a dated source, every comparative claim reviewable, every statistic attributed.

A short go or no-go checklist

Does the announcement contain at least one fact nobody else has published?

Is there a named spokesperson, a date and a specific number in the first 150 words?

Is the authoritative version hosted on your own site, with structured data and a permanent reference page to link to?

Is the company named identically across the release, the website, directories and profiles?

Do you have three named trade outlets you will approach directly, rather than relying on syndication?

Have you recorded a baseline prompt set and cited-source list before publication?

Four yeses or fewer, and the budget is better spent elsewhere.

Do press releases get cited by AI assistants like ChatGPT and Perplexity?

Occasionally, but independent analyses consistently place press releases low in the mix of cited sources compared with news articles, reference pages and documentation. What gets cited far more often is the coverage a release produces. Treat the wire copy as a mechanism for creating citable third-party sources rather than as the source itself.

Why do newswire companies report high AI indexing rates when citation studies show the opposite?

They are measuring different stages. Indexing rates measure whether content is reachable and parseable, which is a distribution property. Citation studies measure how often a source is named in a generated answer for a given prompt set. Both can be accurate at the same time, so ask suppliers for answer-level evidence with the prompts and platforms stated.

Does paid wire distribution improve the chances of AI citations?

Not reliably on its own. Wide syndication creates many near-identical copies, which can obscure the canonical version rather than strengthen it. Paid distribution helps most when it reaches journalists who then write independent pieces, since those write-ups are what assistants are more likely to retrieve and cite.

How long after publishing a press release should you expect any change in AI answers?

Plan for several weeks to a few months. The lag depends on how quickly resulting coverage is indexed, how frequently your category is queried, and how each platform weights freshness. Without a recorded baseline from before publication, attributing any change to the release rather than to a model update is guesswork.

Can a press release fix incorrect information an AI assistant gives about your business?

Rarely by itself. If an assistant has you in the wrong category or repeats an outdated description, the fix is in the underlying material: your own description and structured data, your About and product pages, and the third-party sources the model relies on. Once those are corrected and consistent, announcements reinforce the change instead of fighting it. If you want help diagnosing where the wrong description is coming from, talk to us about monitoring and source attribution.