47 of 250 wire-distributed press releases were cited by at least one AI engine within 90 days. 203 were never retrieved. Not once. Not by any engine. Not in any prompt window.
That is the headline finding of the Press Release Retrieval Study 2026, published today on Everything-PR. It is the first study to measure whether wire-distributed press releases actually appear in AI engine answers and what separates the ones that do from the ones that don't.
I have been writing and distributing press releases since the 1990s. I have watched the press release survive every format shift the industry threw at it. Social, mobile, the collapse of print, the influencer era. It survived all of them because the underlying architecture worked: structured facts, distributed through a reliable channel, picked up by editors who needed material on deadline.
That architecture is now being tested by a reader that does not have a deadline, does not skim, and does not care about your adjectives. The AI engine reads the release -- or more accurately, it reads the journalist who read the release. And 81% of the time, the release never makes it into the answer at all.
The Data
250 press releases. Stratified random sample from PR Newswire and Business Wire public feeds, Q1 2026. Fifty each across five release types: M&A, regulatory/SEC filing, product launch, executive appointment, partnership/award. Tested against five engines -- ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews -- at 30, 60, and 90 days. Two prompts per release per engine per window. 15,000 individual observations.
The retrieval matrix:
| Release Type | Sample | Retrieved | Rate |
|---|---|---|---|
| M&A / Acquisition | 50 | 20 | 40% |
| Regulatory / SEC Filing | 50 | 15 | 30% |
| Product Launch | 50 | 7 | 14% |
| Executive Appointment | 50 | 5 | 10% |
| Partnership / Award / Other | 50 | 0 | 0% |
Read that bottom row again. Zero of fifty partnership and award releases were cited by any engine. Not one. Fifty press releases, each costing $3,000 to $5,000 to distribute on a national wire. Total AI-retrieval value: zero.
The Wire Is Not Broken. Your Release Is.
The counter-argument writes itself: the wire is dead, move on. That is wrong. M&A releases retrieved at 40%. Regulatory filings at 30%. The wire still works -- when the release carries structural news that earns editorial pickup.
The gap is not budget. Not distribution tier -- national vs. regional made no statistically significant difference. Not multimedia -- AI engines cannot process embedded video or images, so multimedia releases performed identically to text-only.
The gap is five predictors. Five variables that separate the 19% from the 81%.
| Predictor | When Present | When Absent | Multiplier |
|---|---|---|---|
| Editorial pickup by top-50 publication | 38% | 6% | ~6x |
| Citable number in first 100 words | 28% | 9% | ~3x |
| 5+ named entities in body | 27% | 11% | ~2.5x |
| Regulatory / SEC-required | 30% | 15% | ~2x |
| Schema-marked boilerplate | 24% | 13% | ~2x |
Editorial pickup is the dominant variable and it is not close. A ~6x multiplier. The engines are not reading the wire -- they are reading the journalist who read the wire. The release is the input. The editorial coverage is the citation anchor.
This is the structural insight the industry needs to absorb: the wire is a distribution mechanism, not a retrieval mechanism. The release gets the story to the journalist. The journalist's article gets the story into the AI engine. Two steps. Most communications teams optimize only for the first.
The Retrieval Window Is 60 Days -- Then It Is Over
At 30 days, 20 of 250 releases had been cited. At 60 days, 35. At 90 days, 47. The curve flattens hard after day 60. Only 12 new citations appeared in the final 30-day window.
What this means operationally: if a release has not been cited within two months, it will not be. The retrieval window for a press release is approximately 60 days. Measure there. Do not wait six months hoping for a long tail that does not exist.
Who Is Actually Reading the Wire
| Engine | Cited (of 47) | Cited Wire URL | Cited Editorial |
|---|---|---|---|
| Perplexity | 40 (85%) | 7 | 33 |
| ChatGPT | 28 (60%) | 2 | 26 |
| Gemini | 22 (47%) | 0 | 22 |
| Claude | 17 (36%) | 0 | 17 |
| Google AIO | 14 (30%) | 0 | 14 |
Perplexity is the only engine that reads wire syndication URLs directly. The rest cite the journalist -- not the wire. In 38 of 47 retrieved releases, the engine cited the editorial coverage, not the wire URL. The wire URL was the cited source in only 9 cases.
No release was cited by all five engines. Maximum was 4 of 5 -- achieved by 7 releases, all M&A with named dollar figures and Reuters or Bloomberg pickup.
The Most-Retrieved Release -- and Why
Nvidia's $4.6 billion acquisition of Run:ai. Cited by 4 of 5 engines within 14 days. Still cited at 90 days.
Why it worked: dollar figure in the headline. Two named entities in the lede -- Nvidia, Run:ai. Named executive -- Jensen Huang. Regulatory angle -- EU and DOJ review. Bloomberg, Reuters, TechCrunch, The Verge, and CNBC published within 24 hours. Every single predictor in the study was present.
Compare that to the bottom of the sample: 50 partnership and award releases. Zero citations. No dollar figure. No structural change. No regulatory trigger. No editorial pickup. Generic boilerplate. The wire cost $3,500 per release. The AI-retrieval value was zero for every single one.
What This Means for How You Write Press Releases
I have written about press release format and structuring releases for AI engines before. This study quantifies what I have been arguing: the release that gets cited is the release that earns editorial pickup, leads with a number, and names enough entities for the engine to anchor against.
Seven rules -- drawn from the data, not from instinct:
1. Lead with a number. "$2.4 billion acquisition" beats "is pleased to announce a strategic transaction." ~3x multiplier. Costs nothing. Rewrite the lede. I covered why numbers-first leads matter in the format guide -- this study proves it with 15,000 observations.
2. Name five entities. Companies, people, products, places, regulators. The engines extract named entities. ~2.5x multiplier. This is the AI press release discipline in one sentence: give the machine something specific to grab.
3. Earn editorial pickup. The dominant variable -- ~6x. The wire delivers the release. The journalist delivers the citation. Pair every wire release with a targeted pitch to 5-10 reporters. The release and the pitch are two documents for two audiences.
4. Do not wire awards, certifications, or minor partnerships. Zero of 50 in this category were cited. Save the wire budget for releases that earn pickup. Everything else belongs on the corporate newsroom with outreach.
5. Build the boilerplate for machines. Schema-marked boilerplate with company name, founding year, headquarters, sector, CEO -- ~2x multiplier over a generic tagline. The boilerplate is the most reliable string a model has for compressing what an entity is.
6. Measure at 60 days, not 6 months. The retrieval curve flattens after 60 days. If the release is not cited by then, it will not be.
7. Write for the machine that summarizes, not the journalist who skims. Declarative sentences. Front-loaded entities. Specific figures. Direct quotes. FAQ sections. Both readers reward the same structural choices -- but the machine is less forgiving when you bury the fact.
The Bigger Picture: Three Sentences on Wikipedia Still Outrank Fifty Press Releases
This study confirms what the 5W AI Lab Founder Reputation Gap Index found earlier this year: press releases on standard wire services do not move AI engine answers on their own. They need editorial amplification to enter the retrieval layer.
The asymmetry is consequential. Most communications functions spend the majority of their budget on inputs the retrieval engines do not weight heavily. The 81% zero-citation rate is the cost of that misallocation -- quantified.
The infrastructure that does work: earned media in tier-1 outlets. Schema-tagged content on owned domains. Wikipedia anchors built through citation-eligible coverage. Consistent entity naming across every surface. Quarterly Citation Share audits across the five major engines.
AI Communications is a mix of journalism, psychology, and engineering. The press release is one component of that system -- not the system itself. The firms that treat the wire as the finish line are funding visibility their buyers will never see.
FAQ
Does this mean press releases do not work?
No. 19% produce AI citations. The 81% that do not share structural characteristics this study documents. The press release format remains the most machine-readable corporate communications format -- when it carries structural news.
Should companies stop using the wire?
Not for M&A, regulatory, or C-suite moves -- 30-40% retrieval. For partnerships and awards -- 0% retrieval -- use the corporate newsroom plus outreach instead.
What is the most actionable finding?
Lead with a number. ~3x multiplier. Costs nothing.
What is the most important finding?
Editorial pickup is ~6x. The wire delivers the release to the journalist. The journalist delivers the citation to the engine. Two steps. Most teams optimize only for the first.
Where can I read the full study?
The complete dataset, methodology, prompt bank, and coding protocol are in the Press Release Retrieval Study 2026 on Everything-PR.
Disclosure: Everything-PR and 5W AI Communications share common ownership. Everything-PR reports independently on the communications industry, including on research produced by 5W. Editorial decisions are made by Everything-PR's editorial team.
Ronn Torossian is the founder and chairman of 5W AI Communications, the AI Communications Firm. He is the publisher of Everything-PR and the author of two best-selling editions of For Immediate Release.