Press releases in AI search: 4.85K wire citations across 7.43M
4.85K wire-host citations across 7.43M measured citations. Compare wires, engines, and announcement title groups.
Originally published October 7, 2026
Selected wire hosts supplied 0.065% of measured citation occurrences. Downstream editorial pickup is outside that direct count.
July 9–October 6, 2026 (UTC) · 4,850 wire-host citations out of 7,429,354 citations
- Direct wire share
- 0.065%Citation-occurrence denominator
- Wire citations
- 4,850Repeated citations count repeatedly
- Total citations
- 7,429,354Both measured engines
Direct wire citations were a small part of this AI-answer sample. The selected wire hosts contributed 4,850 of 7,429,354 citation occurrences, or 0.065%, from July 9 through October 6, 2026. That result measures direct source use; it does not measure all editorial coverage an announcement may have prompted.
The measurements come from Parse, the AI answer index Dimitry Apollonsky also runs; the analysis and interpretation are Signals'. This is an observational study, not a test of a Signals service.
The panel covers ChatGPT Search and Google AI Mode and excludes Parse and Soar source domains. Signals runs an aged Reddit account marketplace plus an editorial network for AI brand mentions across Reddit, Quora, Product Hunt, and Threads.
Wire hosts supplied a small direct citation share
The 0.065% share divides selected wire-host citation occurrences by all eligible citation occurrences in the frozen panel. It is not the proportion of answers citing a wire, the percentage of releases ever cited, or a measure of the press-release industry's total impact. Each of those questions would require a different denominator.
A page cited in several answers contributes several occurrences. The definition covers the selected wire domains and their subdomains, which can host more than conventional release pages. We therefore use the term wire-host citations rather than claiming to have manually classified every URL as a press release.
The practical implication is to ask for evidence when a distribution offer promises AI recommendations. The small direct share does not make an announcement worthless, but it also cannot be transformed into a campaign success rate. The press-release guide explains how to separate the communication objective from the AI measurement.
PR Newswire led the selected host table
The main prnewswire.com host accounted for 2,582 citation occurrences, ahead of businesswire.com and globenewswire.com in this observed cut. The figure ranks selected hosts by occurrences and preserves the host names. Subdomains remain separate in the raw output so the result can be checked without silently changing the counting unit.
PR Newswire led the selected wire-host counts
Wire-host citation occurrences · citations · July 9–October 6, 2026 (UTC)
View data table
| prnewswire.com | 2,582 |
|---|---|
| businesswire.com | 1,169 |
| globenewswire.com | 614 |
| newswire.com | 171 |
| einpresswire.com | 125 |
| prweb.com | 113 |
| prlog.org | 34 |
4,850 wire-host occurrences; leading hosts shown. Source: Parse index, analyzed by Signals.
Host membership does not guarantee every URL is a conventional press release.
The ranking lacks the number of releases published by each wire and the proportion relevant to the tracked questions. It therefore cannot estimate citation probability per release, quality per dollar, or the effect of choosing one distributor over another. More observed citations can reflect a larger or differently matched supply of material.
For a buyer, the table is a source-discovery aid. Inspect actual URLs and determine whether the material resembles the planned announcement. An observed citation to a wire's editorial resource is not the same evidence as a citation to a paid release, even when both share part of the hostname.
Both engines cited wire-host pages
The engine figure reports wire-host citations as a share of each engine's eligible citation occurrences. This keeps Google AI Mode's and ChatGPT Search's different citation volumes from becoming an implicit preference claim. The denominators are the engines' observed citations in the same calendar window, not matched sets of release pages supplied to both.
A difference in share can reflect the questions covered, source selection, or the type of answers produced. The pooled window also spans changes in engine behavior. This figure is a descriptive split, not evidence of a stable future rate or an experiment showing that a particular wire suits one engine better.
Use the result to keep engine-specific reporting in a campaign plan. If a release is later cited, save the engine and full answer with the source URL. Reporting “AI cited us” without that context makes it harder to reproduce the observation or compare it fairly with a later check.
Launch wording appeared in part of the cited material
Launch-related title patterns accounted for 1,203 wire-host citation occurrences. Funding and partnership patterns formed separate groups, with the remainder kept as other or unclassified. These labels come from an ordered title classifier. They describe the observed wording rather than a manual determination of the underlying news event or its importance.
Most cited wire material fell outside the selected title patterns
Wire citations by ordered title pattern · citations · July 9–October 6, 2026 (UTC)
View data table
| Funding | 206 |
|---|---|
| Launch | 1,203 |
| Other or unclassified | 3,150 |
| Partnership | 291 |
4,850 wire-host occurrences. Source: Parse index, analyzed by Signals.
Funding is matched first, then launch, then partnership; the remainder is retained.
A page about a funded launch can match more than one idea; the SQL assigns the first matching category. The group totals also weight repeated citations, so they are not counts of unique announcements. A frequently cited page can contribute multiple observations without indicating that many similar releases would receive the same treatment.
The finding is useful for inspecting what kinds of source material appeared, not for inventing a winning release template. A real announcement should state what changed and support its claims. Adding launch language to an ordinary company description does not become an evidence-based AI tactic because that word appears in this sample.
Syndicated pickup is outside the direct numerator
A release can be followed by an article on another domain, but that article is not counted as a wire citation here. The study measures direct wire-host references. It does not map the path from a release to independent coverage, determine whether the release caused that coverage, or attribute later brand naming to the announcement.
This boundary cuts both ways. The direct share may omit downstream influence, but the remaining citations cannot be credited to press releases either. Some articles are independent reporting, some repeat an announcement, and some have no relationship to a release at all. A host classification alone cannot distinguish those mechanisms.
A stronger pickup analysis would link each announcement to subsequent pages using content and timing, then classify the relationship. It would preserve independent reporting and copied syndication as separate outcomes. Parse's publisher research offers related evidence about cited publishers, with a different unit and no claim to solve that release-level attribution problem.
Host matching is transparent but imperfect
The wire group uses explicit hostname patterns for PR Newswire, Business Wire, GlobeNewswire, Accesswire, EIN Presswire, PRWeb, Newswire, and PRLog. Matching includes their subdomains. Parse and Soar are excluded from all source tables. The committed SQL makes those choices inspectable and keeps the sample from depending on an undocumented commercial-source label.
Hostname membership is still a proxy. A wire host can serve editorial resources, media assets, or other pages alongside releases. Some press releases also appear on company websites or publications outside the selected host list. They do not enter the wire group simply because their content resembles an announcement.
The study therefore answers a defined source-use question rather than claiming a complete census of press releases on the web. A refresh should retain the host definition for comparability or explain the change explicitly. Expanding the regex and keeping the old headline denominator would make the apparent trend difficult to interpret.
The data cannot supply a release-to-citation delay
The selected citations have observation dates, but reliable publication dates are not available for enough release pages to support a lag study. The first time Parse observes a page cited is not the date the page was published. Using one as the other would collapse an unknown period into an invented measurement.
The Signals ledger also does not provide a complete set of delivered release URLs for this batch. Without those URLs and publication receipts, we cannot construct an order cohort with consistent follow-up or report a service-level time to citation. The missing measurement remains unmeasured rather than being represented as zero days or no effect.
For future work, save the release URL and actual publication timestamp, then record direct citations and subsequent editorial coverage separately. A fixed question panel would make the follow-up interpretable. It would still require a comparison design before the resulting chronology could support a claim about the effect of distribution.
Two counting formulations agree on the headline
We counted wire occurrences with a conditional aggregate over the eligible citation set and reconciled that total by summing the per-host wire groups. The full citation denominator was separately recomputed from eligible answer citation arrays using an existence check against the source set. The returned totals agree at 4,850 and 7,429,354, respectively.
The receipts record the raw query rows and map every figure to those rows. The share is calculated from exact totals and rounded only in the presentation. Smaller host cells remain descriptive; they are not promoted into claims about the performance of an individual distributor or the likely outcome of a future release.
Agreement between formulations checks the counting, not every source classification. Both use the same explicit host definition and answer eligibility rules. The study should be quoted with its window and unit so readers can distinguish a direct wire-host share from answer prevalence, release success rate, or brand-recommendation attribution.
The buying decision needs a communication objective
The measured direct share supports a concrete operator question: what result is the distribution purchase meant to deliver? If the objective is a public announcement, evaluate the audience and delivery. If the objective is brand naming in AI answers, define that observation separately and ask what evidence connects the proposed work to the intended result.
The press-release guide gives the release brief, vendor specification, and follow-up record. Other Signals research articles examine source types with different denominators. Their numbers should not be assembled into a return-on-investment ranking without a common treatment and outcome design that this batch does not contain.
A null or small direct presence is still useful research. It prevents a campaign report from replacing the buyer's question with a distribution count. The honest next step is a clearer objective and better receipts, not a claim that every release fails or that an editorial alternative has a proven causal advantage.
Questions about the wire-citation result
The headline is deliberately narrower than “press releases work” or “press releases do not work.” It reports the share of eligible citations assigned to a named host group in a frozen panel. The questions below cover the common denominator changes that would otherwise turn that useful observation into a claim the analysis did not measure.
Does this measure the share of AI answers citing press releases?
No. It measures citation occurrences. An answer can contain several citations, and a cited page can appear in many answers.
Does the low share prove releases have no downstream value?
No. Editorial pickup and other communication outcomes are outside this direct wire-host measurement.
Which wire should we buy?
The host counts do not establish a per-release success rate or a controlled service comparison. Choose against the announcement's concrete audience and deliverables.
How we measured
Read-only production queries frozen at October 7, 2026. Answers are ChatGPT Search and Google AI Mode in the stated UTC window, excluding gpt-5-mini and gpt-5-3-mini records while retaining blank model labels. Citation IDs are expanded from eligible answers. Parse and Soar source domains and subdomains are excluded. Latest content snapshots are selected strictly before October 7. SQL, raw output, and number trace are committed under docs/research-receipts for this slug. Wire hosts match explicit PR Newswire, Business Wire, GlobeNewswire, Accesswire, EIN Presswire, PRWeb, Newswire and PRLog patterns including subdomains. Announcement groups are title regexes.
What this cannot tell us
- Host membership includes non-release material and excludes releases on other hosts.
- No release-volume denominator, pickup linkage, or reliable publication-to-citation lag is available.
- The tracked question panel is not a random sample of all AI search.
- This observational analysis does not establish a causal effect of a Signals service.
