Press releases do get cited by AI search engines, rarely, and almost never as the wire page itself. In the largest published count, a study of more than 25 million citations put press releases at about 1.1% of the total, and a stratified audit of 250 releases found that only 19% were cited by any engine inside 90 days, with award and partnership releases at zero. The releases that worked were the ones a publication picked up, and the citation then pointed at the publication rather than the wire. So ask whether anybody would write about the release rather than whether to distribute it. A funding round, a filing or an acquisition clears that bar; a certification does not. Spend the budget on the pitch and on something worth quoting, because paid content is a rounding error in what engines cite.
Press releases are cited by AI search engines occasionally, and they are a small share of everything engines quote. Muck Rack analysed more than 25,000,000 links cited by ChatGPT, Claude and Gemini and found press releases accounted for about 1.1% of all citations, with paid and advertorial content at about 0.3%, in "What is AI Reading?" (May 2026, read 19 November 2026). The same report notes the press release share fell from 6% in its December 2025 report and sits near the 1.2% it recorded in July 2025, mostly because ChatGPT cites releases less often than it did.
Counting from the other direction gives a number that sounds better and means something different. EPR Research took a stratified random sample of 250 press releases from public PR Newswire and Business Wire feeds in the first quarter of 2026, checked five engines at 30, 60 and 90 days, and found 47 of the 250, or 19%, cited by at least one engine within 90 days, with 203 never retrieved, in "The Press Release Retrieval Study 2026" (12 August 2026, read 19 November 2026). Its cumulative figures were 8% at 30 days, 14% at 60 and 19% at 90. Everything-PR discloses that its own ownership overlaps with 5W Public Relations, so a study of what the wire delivers comes from a publisher connected to the agency side of the same market.
Hold both numbers at once and the answer is not confusing. A release has roughly a one in five chance of being retrieved at all, and the whole category is about one percent of what engines cite, so the wire is a minor channel in which most individual items do nothing.
Which fifth gets through is answerable, and the published answer is specific enough to act on.
The kinds of press release that get cited are the ones that report a fact a third party would have reported anyway. In the EPR sample, the retrieval rate ran from 40% down to zero depending entirely on what the release was about, and the pattern is the most useful thing in the study.
| Release type | Cited within 90 days | What that tells you |
|---|---|---|
| Merger or acquisition | 20 of 50, 40% | Externally consequential, reported by others |
| Regulatory or SEC filing | 15 of 50, 30% | Required disclosure, treated as a source of record |
| Product launch | 7 of 50, 14% | Cited when the product itself is newsworthy |
| Executive appointment | 5 of 50, 10% | Occasionally picked up in trade coverage |
| Partnership, award or other | 0 of 50, 0% | Nothing an engine has a reason to quote |
The zero row is the finding worth taking to a budget meeting. Fifty releases announcing partnerships, awards and similar milestones produced no citations on any of the five engines across three checks. On that evidence the award announcement is a ritual rather than a channel. EPR puts an illustrative cost on that block of spend using its own assumed price per release, and flags the assumption as its own rather than a published rate.
Read the top two rows the same way. Acquisitions and filings are cited because they are facts the market needs, which is a property of the event and not of the writing. No amount of structure turns a certification into an acquisition.
Citations point at the publication because the engine is quoting the article somebody wrote, not the announcement you paid to syndicate. In the EPR sample, of the 47 releases that were cited anywhere, Perplexity accounted for 40 of them with only 7 wire URLs among them, ChatGPT for 28 with 2 wire URLs, and Gemini, Claude and Google AI Overviews cited none of the wire URLs at all.
Editorial pickup is also the single strongest predictor in the same study. Releases picked up by a top-50 publication were cited 38% of the time against 6% for those that were not, a gap of roughly six times, which EPR reports alongside smaller effects for a citable number early in the release and for named entities in the body.
So the wire is doing the job of a delivery mechanism rather than a destination. The release gets your facts in front of a reporter and into a few syndicated copies; what an engine later retrieves is the version with a byline on a domain it already trusts. Budgeting for distribution and not for the pitch inverts the mechanism.
That also explains why the duplicated syndication footprint helps less than it looks. Dozens of identical copies on low-authority sites give an engine many versions of one document and no new information, which is close to how our write-up on whether AI-generated content gets cited describes the same problem from a different direction.
Wire services report higher citation rates because they count a different thing, and because the party selling distribution is the party publishing the study. Two of the figures circulating in 2026 come from a distributor's analysis of releases sent through its own wire, which is a sample of exactly the releases that were worth paying to send.
Three denominators are in play and they are not comparable. The share of all AI citations that are press releases, which Muck Rack's study measured at about 1.1% across its 25 million links. The share of releases that get cited at least once, which EPR measured at 19% of 250. And the share of a single wire's own releases that appear anywhere in an engine's answers. Distributors publish that third number, and it says nothing about how often a release is the source an answer rests on.
We are not quoting the distributor figures here, because the standard on this site is that a number goes into an article only once somebody has opened the source. Both primary pages we tried, the distributor's own write-up and its release on the wire, returned HTTP 403 to us on 19 November 2026. Where the only high number in a market comes from the company selling the service and the page will not open, treat it as a sales claim until it does.
The asymmetry is worth naming for any vendor claim in this category, ours included. A study by a seller is not worthless, and it is also not the same as an independent count: the method, the sample and the definition of a citation all have to be visible before the figure means anything.
A release becomes citable when it contains something quotable in the first few lines and enough named detail to be checked. The EPR study found that a citable number inside the first 100 words was associated with a 28% citation rate against 9% without one, five or more named entities in the body with 27% against 11%, and schema-marked boilerplate with 24% against 13%, which is the same ordering our own framework uses on any page.
Question type matters as much as structure, and it is outside your control. Muck Rack found that press releases appear disproportionately in answers to industry-trend questions, with about 1% of trend responses containing a press release citation, roughly 3.5 times the rate for best-of questions. A release is a plausible source when a reader asks what is happening in a market and an implausible one when they ask which vendor to choose.
Three habits follow. Put the number, the date and the named party in the first paragraph rather than the third. Name the people, places, products and figures precisely enough that a reporter could verify them without calling you. Then pitch it to the three publications that actually cover your category, because the pickup is the mechanism and the wire is the paperwork.
If you want to know which publications the engines are already reading for your category before spending anything on either, a free visibility assessment samples your buying questions across the engines and lists the sources the answers were built from. The list usually contains two or three outlets nobody on the team had on the media plan.
Publishing a release to influence AI answers is ordinary public relations, and it becomes a rules problem at the point where the content misrepresents who wrote it. Anthropic's usage policy, effective 12 November 2026, prohibits manipulating the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship or independence, in its published usage policy (read 19 November 2026). The example it gives is sites posing as unaffiliated sources to corroborate the same claims.
A labelled release on a wire, carrying your name and a contact, is not that. A network of plausible-looking independent blogs repeating your claim is, and so is a syndication strategy whose purpose is to make one source look like twelve. The line is authorship and independence rather than intent to be found.
Keep the distinction in mind when a vendor offers volume. The offer that arrives as hundreds of placements on sites you have never heard of is selling the shape of the thing the policy describes, and the measured value of paid placement is small enough that the risk is badly priced: about 0.3% of citations in the Muck Rack study.
Our longer answer on the paid route sits in whether you can pay to get cited by AI search engines, which covers sponsorship, syndication and the places where paying genuinely does work.
Spend the PR budget on the things engines cite most, which are earned coverage and material other people want to quote. Muck Rack's study put earned sources at about 84% of citations, with journalism alone at about 27%, and owned plus paid content together at about 16%, so the budget allocation that matches the evidence is weighted towards being written about rather than towards publishing.
Four line items do that work. Original research from data you already hold, which gives reporters a reason to call and gives other sites something to cite. Expert availability, so a named person with a title answers a reporter the same day. Inclusion in the comparison and best-of pages that cover your category, which is a pitch rather than a placement. And a small number of releases reserved for events that are genuinely consequential.
Keep the wire for the filings, the funding and the acquisitions. Those are the categories with measured retrieval rates, and in those cases the release is also the document of record, which is a second reason an engine has to reach for it.
Our write-up on whether you need third-party mentions or fixing your own site is enough covers how to tell which side of that split your problem is on, and our two anonymised early case studies show what the on-site half of the work looks like when it moves, with the page's own caveat that the audit score is an input metric rather than the outcome.
Communications and demand generation teams ask these five questions most often about press releases and AI search.
Occasionally, and less than it used to. Muck Rack's May 2026 study of more than 25 million cited links put press releases at about 1.1% of all citations and noted the decline was driven mostly by ChatGPT citing them less often than in its December 2025 report. The release helps most when a publication picks it up, because the citation then points at the article rather than the wire.
Yes for the release types that get retrieved, which in the EPR sample were acquisitions, regulatory filings and newsworthy product launches, and no for the ones that did not, where partnership and award releases returned zero citations across five engines. The wire is a distribution mechanism for facts reporters want, not a visibility product in itself.
Longer than the news cycle it was written for. The EPR study's cumulative figures were 8% of releases cited at 30 days, 14% at 60 and 19% at 90, so most of the eventual citations arrived after the first month. Judging a release on week-one visibility measures the wrong window.
Mostly the coverage. Of the 47 cited releases in the EPR sample, Perplexity accounted for 40 with only 7 wire URLs among them, ChatGPT for 28 with 2, and Gemini, Claude and Google AI Overviews cited no wire URLs at all. Plan the release as an input to coverage and measure it by the coverage it earns.
Rarely, on the published evidence. Paid and advertorial content accounted for about 0.3% of citations in the Muck Rack study, and duplicated copies give an engine many versions of one document rather than new information. A syndication strategy designed to make one source look like several independent ones also runs into Anthropic's usage policy on misrepresenting origin and independence.
Related guides
A free visibility assessment samples your buying questions across ChatGPT, Perplexity, Google AI Overviews and Claude, and reads your public pages the way a crawler sees them.
Request a free visibility assessment →