Your press releases aren't press releases anymore. They're training data. Every one you publish this quarter either feeds an AI engine's next answer about your category or gets ignored by it, and a new digital PR study just showed, with real numbers, which kind you're writing. The gap between the two isn't tone, distribution list, or send time. It's whether the release contains a number a model can actually use.
We've covered the earned-media side of this before: Muck Rack's data showing 82% of AI citations trace back to earned media, not owned or paid content. That number has held up as the strongest argument for digital PR as an AI search strategy, and it's been enough on its own to get budget reallocated away from paid placements and toward journalist relationships. LeadCoverage's new study, distributed on GlobeNewswire on July 27, doesn't contradict that finding. It sharpens it, and it hands the industry a second, more dangerous conclusion at the same time: if a number gets you cited, everyone's about to start publishing numbers, whether or not those numbers mean anything.
That second conclusion is the one most digital PR teams will skip past, because the headline number is more fun to repeat. A 3.5x lift is the kind of stat that ends up in a slide deck within a week. What belongs in that same slide deck, and usually doesn't make it, is the discipline required to earn the lift honestly, and what happens to the whole channel once that discipline gets treated as optional.
What LeadCoverage's digital PR experiment actually found
LeadCoverage LLC, run by CEO and co-founder Kara Brown, ran a controlled test over one quarter: one press release per week, roughly thirteen releases total, tracking how often AI tools cited the company's data afterward. Some releases led with a specific, economically relevant number. Others covered the usual press-release filler: a new hire, an award, a company update. Same cadence, same distribution, same company. The only variable that mattered was whether the release contained a number worth citing.
What makes this study useful instead of just another self-serving vendor stat is the control. Most digital PR case studies compare one company's results to an industry average, which buries the actual variable under a dozen confounders: distribution list quality, outlet relationships, brand size, timing. LeadCoverage held all of that constant and only changed the content of the release itself. When the citation rate moved 3.5x on that single change, it's a much harder result to explain away as noise.
The releases that lost weren't badly written. They were just the wrong kind of news. Company announcements, personnel changes, and award wins earned almost none of the 1,058 citations, no matter how cleanly they were pitched or how wide the distribution list. The releases that won led with a specific number, not a range, not a vague trend claim, an actual figure with enough context to be economically relevant to whoever was asking the question.
That's a hard reset for how most PR calendars get built. A typical quarterly plan mixes a product update, a hire announcement, an award submission, and maybe one data point if research had time to run a survey. Under this data, three of those four items are functionally invisible to an AI engine, no matter how well they land with a human editor. The data point isn't one release among several. It's the release doing almost all of the AI-citation work, and everything else on the calendar is there for other reasons entirely, which is worth saying out loud to whoever owns the PR budget.
“AI cannot invent a number, so it cites whoever published one.”
That's Brown's explanation, and it's the whole mechanism in one sentence. A language model can paraphrase a claim, restate an opinion, or summarize a trend in a dozen different ways without ever needing a source. It cannot manufacture a statistic. The moment a query needs a number, the model has to go get one from somewhere, and it reaches for whoever published it first, cleanest, and most repeatedly. Brown made a second point worth sitting with too: "The source cited today is hard to unseat tomorrow, because these systems reward freshness and repetition." Early, original, and specific beats prestigious. That 83% jump in search impressions during the test window is the visible half of that compounding effect; the citation count is the half that keeps paying out after the news cycle ends.
The obvious risk: press-release spam built to bait AI citations
Here's the problem with a finding this clean: it's a recipe, and recipes get copied badly. "Publish a number, get cited" is a one-line playbook, and one-line playbooks are exactly what a press-release mill turns into volume. The moment this kind of result becomes common knowledge across the digital PR industry, expect a wave of releases engineered to look data-led without actually containing anything defensible. A made-up survey of nine respondents. A percentage with no denominator. A "proprietary index" nobody can audit. All of it dressed up to trigger the same citation behavior LeadCoverage measured, none of it earning the citation honestly.
That's not a hypothetical risk, it's the default outcome any time an algorithm rewards a legible pattern. Give an industry a one-line playbook and a distribution wire that charges the same flat fee whether a release contains real research or three made-up numbers, and the flood is not a matter of if. It's a matter of how many quarters before an inbox full of "proprietary index" releases makes every data claim in a press release worth slightly less than it is today, the same tragedy of the commons that ruined cold email deliverability and is already reshaping how PR distribution gets priced.
That flood runs into a wall that's already been measured, though. Ahrefs' Ryan Law studied 331,000 pages this year and found no evidence that Google detects or penalizes content for being AI-written; what it does penalize, consistently, is thin, unhelpful content, regardless of who or what produced it. Swap "AI-written" for "citation-baiting press release" and the same logic applies. A number with no methodology behind it is thin content wearing a data costume. It might earn a citation or two before an engine's retrieval systems catch on, the same way it might earn a temporary click from a human skimming a headline. It won't hold up, and it won't compound the way a real, defensible data point does.
| DATA-LED DIGITAL PR | GENERIC PRESS-RELEASE SPAM | |
|---|---|---|
| What it publishes | A specific, sourced, economically relevant number | A vague claim, a made-up percentage, or no data at all |
| Why AI cites it | It's the only concrete figure a model can point to | Superficially resembles a citable claim, until checked |
| Durability | Gets re-cited repeatedly as the freshest, most repeated source | Cited rarely, dropped fast, or never picked up at all |
| Cost to produce | Real research or internal data, plus editorial rigor | A press-release template and a distribution fee |
| Outcome at scale | Compounding citations and rising search impressions | Noise that erodes trust in a brand's future releases |
The verdict: Muck Rack and LeadCoverage agree on the same thing
Put LeadCoverage's 3.5x lift next to Muck Rack's 82% earned-media figure and it's tempting to read them as two separate arguments for the same conclusion: publish more press releases. That's the wrong read, and it's the one that will get a lot of digital PR budgets wasted over the next year. Look closer and both studies are measuring the same variable from different angles, and it isn't volume. It's whether the release actually contains something worth citing.
Muck Rack's 82% figure isn't a case for earned media as a category, it's a case for earned media as evidence of rigor. A journalist doesn't cover a company announcement or an award; they cover a claim that survived their own scrutiny, which is precisely why that kind of coverage gets trusted by an AI engine's retrieval systems in the first place. LeadCoverage's data shows the same filter operating one level upstream, before a journalist ever gets involved: a press release with a real number passes a version of that same scrutiny test on its own, because the model checking it can't invent the number itself and has no reason to doubt one that's specific and sourced. Different mechanism, same gate. What gets through in both cases is the release with real data and real methodology behind it, not the release that showed up.
That's also why the two findings don't cancel each other out, and why we treat digital PR as durable in our comparison of digital PR against Google's Preferred Sources: both point at earned trust that compounds across engines, not a platform setting that can be redesigned away. A well-sourced data release earns citations the same way a well-reported article does, because both are surviving a scrutiny test an AI engine is quietly running on every claim it considers repeating. Press-release volume without that discipline never survives the test. It just looks, for a news cycle, like it might.
It also changes what a PR team should be reporting to leadership next quarter. Placement count and reach were always rough proxies for value. Citation count and re-citation rate are a more honest one, and they reward a completely different kind of work: fewer releases, better sourced, published on a rhythm an engine learns to trust. A calendar built to maximize the number of releases going out the door is optimizing for the wrong denominator. A calendar built around one real, defensible number per release, on a predictable cadence, is optimizing for the one LeadCoverage and Muck Rack both just confirmed actually moves the citation count.
How to run digital PR that survives the citation-gaming wave
None of this is a call to publish less. It's a call to publish differently, and it's the same discipline behind our link building and digital PR engagements: original data, defensible methodology, and a cadence that gives an engine repeated reason to trust the source. It also matters where that trust gets built, given how much of the open web's referral relationship with Google is already shifting; the same pressure that's pushing publishers toward a more adversarial stance on AI Overviews is exactly why owning a defensible, original data asset matters more now than it did two years ago, when a press release only had to satisfy a human reporter's inbox.
Do this next
Pull your last four press releases and check them against one question: does each one contain a specific, sourced, economically relevant number, or is it company news dressed up as an announcement? If it's the latter, it's not earning AI citations, and LeadCoverage's quarter-long PR citation study shows exactly why. Then build the next quarter's calendar around one real data point per release, published on a cadence, with the methodology shown, not hidden. That's the version of digital PR that compounds. Everything else is about to get drowned out by its own imitators.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.