Press Release SEO: Does AI Cite Press Releases?

For fifteen years, press release SEO meant one thing: buy a wire distribution, seed a few backlinks, and hope Google noticed. Most SEO practitioners quietly wrote off press releases years ago, once the backlinks stopped counting for much.

Then the question changed. Buyers stopped asking Google "best payroll software" and started asking ChatGPT. And ChatGPT, it turns out, reads press releases very differently than Google's link graph ever did.

A July 2026 study put a number on it. It is a large number, and it comes with a catch most marketers will miss.

Does AI cite press releases?

Yes. AI models cite press releases, and they do it fast. A July 2026 Notified/PRWeek report analyzing roughly 250 million AI citations found that 99.3% of 8,000 syndicated press releases were cited by ChatGPT or Claude, with an average time to first citation of about eight hours. The catch: that rate applies to widely syndicated releases, not a PDF posted on your own newsroom.

That last sentence is the whole game. The 99.3% figure is real, but it measures releases distributed across a wide network of high-authority domains. The syndication is doing the work, not the press release format itself.

Read the two claims side by side and the nuance sharpens:

What the headline says:

• •Press releases get cited 99.3% of the time.
• •AI picks them up in about eight hours.

What the data actually says:

• •Syndicated releases, seeded across many trusted domains, get cited near-universally.
• •The citation window is perishable, because recency is one of the four signals models weigh.

Press releases are not back. Syndicated, structured, factual releases are back. That is a narrower and more useful claim.

Why AI started citing press releases again

Press releases lost their SEO value when Google discounted the links inside them. AI search brought them back for a completely different reason: models do not count links, they read text and pick sources. Four shifts explain the turnaround.

Reason #1: AI reads what the wire syndicates, not your newsroom

A press release on a wire like GlobeNewswire or PR Newswire lands on dozens of publisher sites within hours. That syndication footprint is exactly the kind of distributed, consistent reference AI models pull from. The Notified/PRWeek report is candid that its 99.3% rate reflects GlobeNewswire-distributed releases. The mechanism is the network, not the document.

Reason #2: Third-party sources carry most of the citation weight

For SaaS brands, an Elevarus analysis found 84% to 93% of AI-citation weight comes from third-party sites, not the brand's own domain. A syndicated release is one of the few ways to plant a factual, on-message reference on many third-party domains at once, on your schedule rather than a journalist's.

Reason #3: Freshness is a first-class ranking signal

The Notified report names its selection framework SOAR: Structure, Originality, Authority, and Recency. Recency is not a tiebreaker, it is a primary lever. Our own tracking across 34,000-plus AI answers shows how quickly source pools rotate, with the leader in a given answer flipping in 24% of editions. Fresh, dated content wins slots that stale pages lose.

Reason #4: Speed matters when the query is new

The eight-hour time-to-citation is the sharpest number in the study. When a topic is new, whoever publishes a clean, factual, widely-syndicated reference first can seed the answer before competitors show up. A press release is one of the fastest ways to get a dated, quotable claim onto the open web.

What actually gets a press release cited

The SOAR framework is a useful checklist, but most releases fail on the same two signals: originality and structure. A model cannot quote a sentence that says nothing specific, and it cannot extract a claim buried in three paragraphs of adjectives.

It leads with a factual, extractable answer

AI extracts 40-60 word passages, not whole documents. The first paragraph should state the news as a clean, self-contained fact a model can lift verbatim. Bury the news under a quote from your CEO and you hand the citation to a competitor who led with the number.

It carries one net-new number nobody else has

Originality is the signal most releases skip. "Leading provider announces new features" is not quotable. "Payroll runs dropped from 40 minutes to 9 after the update, across 1,200 accounts" is. Give the model a first-party statistic it cannot get anywhere else and you become the source, not a mention.

It names a real person and a primary source

Authority in the SOAR sense is not domain rating. It is whether the claim is attributable and verifiable. A named spokesperson, a link to the underlying data, and a boilerplate dense with entity facts all raise the odds a model treats the release as a citable source rather than noise.

It gets syndicated, not just posted

This is the difference between the 99.3% and near-zero. A release on your own site is one page on one domain. The same release on a wire is dozens of references across domains AI already crawls. If you only publish to your newsroom, you are testing the wrong variable.

How to make your press releases citable

The diagnostic half explains why releases get cited. This half is the sequence to run them so they actually do. Treat it as a repeatable motion, not a one-off launch.

Step 1: Audit which sources AI already cites in your category

Run your top buyer prompts through ChatGPT, Perplexity, and Google AI Mode and record every domain the answers cite. That list tells you which wires and publishers feed the models in your space. Pair it with a brand mention audit to see where your name is missing.

Step 2: Write the release around one quotable statistic

Before you draft, decide the single number the release exists to plant. Lead with it in the first 60 words. Everything else, the quote, the boilerplate, the context, supports that one extractable claim.

Step 3: Syndicate through a wire the models actually crawl

Post-only-to-newsroom is the most common mistake. Use a distribution network that seeds your release across the third-party domains your Step 1 audit surfaced. The syndication is what turns a single page into the distributed footprint AI rewards.

Step 4: Date it and refresh it on a cadence

Recency decays, so a release cited today can drop out in weeks. This is citation drift, and it is normal. For evergreen announcements, re-issue an updated release when the citation window fades rather than assuming one push holds.

Step 5: Measure citation lift on a 14-day window

Check your prompt set before the release, then again at day 14 or later. Early citation can appear within hours, but the durable pattern takes a week or two to settle as retrieval pipelines re-crawl. Teams that check at day three will misread a working release as a dud. A managed GEO agency can run this loop for you if the measurement burden is too much in-house.

Where press releases fit in your AI-visibility mix

Here is the honest limit. Press releases are fast and syndicate wide, but they are perishable and they are not the highest-yield channel. Reviews and community threads still carry more durable citation weight for most B2B brands. Trustpilot's analysis of 800,000 AI answers found review-active brands cited in 75.3% of answers versus 1% for brands with no reviews, which is why the broader digital PR for AI search playbook ranks a generic wire below them.

Use press releases for what they are good at, and do not overspend on them:

Press releases do this well:

• •Seed a dated, factual claim fast when a topic is new.
• •Reach many third-party domains on your own schedule.
• •Plant a first-party statistic AI can quote as the source.

Press releases do not do this:

• •Build the durable review and community footprint models trust most.
• •Hold a citation without refresh, because recency decays.
• •Manufacture authority a thin, adjective-heavy release never earns.

The brands winning here treat the release as one input into a standing program, measured on a two-week clock, not a launch-day trophy. If you want that program built and tracked across every AI surface, our GEO services are designed for exactly this.

FAQ

Does AI cite press releases?

Yes. A July 2026 Notified/PRWeek report found 99.3% of 8,000 syndicated press releases were cited by ChatGPT or Claude, with an average time to first citation of about eight hours. The rate applies to releases distributed across a wide network of trusted domains, not a release posted only to your own newsroom.

Is press release SEO still worth it in 2026?

For AI search, yes, with a caveat. Press releases no longer earn ranking value from backlinks, but a syndicated, structured, factual release is one of the fastest ways to plant a citable claim across third-party domains AI models read. It works as part of a program, not as a standalone tactic.

How long does it take AI to cite a press release?

The Notified/PRWeek data reported an average time to first citation of roughly eight hours, with most citations occurring within 24 hours of distribution. The durable pattern takes longer to settle, so measure citation lift at day 14 or later rather than reading the first day as the final result.

Do press releases help with ChatGPT and Perplexity citations?

They can, because both engines lean on third-party sources and a wire release lands on many at once. But reviews and community threads generally carry more durable citation weight. Use press releases to seed fresh, dated claims, and pair them with review and community work for lasting share.

What makes a press release get cited by AI?

Four signals, framed by the report as SOAR: Structure (a clean 40-60 word extractable answer), Originality (one net-new first-party number), Authority (named sources and verifiable claims), and Recency (a clear date and regular refresh). Wide syndication is what turns those signals into near-universal citation.

What to do with this

Press release SEO is not dead, but the old version is. The wins no longer come from links in the release. They come from a syndicated, structured, factual claim that a model can quote within a day, on a channel that reaches many trusted domains at once.

Treat the release as a fast way to seed one quotable number, syndicate it wide, date it, and refresh it when the citation window decays. Then measure on a two-week clock. Run it that way and press releases earn a place in your AI-visibility mix. Run it the old way, as a one-off wire blast with no number worth quoting, and you get the same result you got in 2020: nothing.