Public relations can contribute evidence that AI search retrieves and cites. The source mix varies by category and question; the measurements below describe that mix, not the causal effect of a PR campaign. Across six answer engines over 135 days, editorial publications carry 11.97 percent of the citations we observe, vendor-owned domains carry 9.53 percent, and the entire wire and press-release class carries 0.21 percent across ten domains. Those three numbers come from our own instrument, and each of them points somewhere different from where the usual advice points.
The measurements below are read from the Machine Relations Index release generated 2026-09-28: 17,868 monitored runs of 1,010 buyer prompts across ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and Google AI Mode, observed over 135 days, producing 24,639 cited domains.
The Denominator, Stated Before the Findings #
Every share on this page is computed against one quantity: 123,924 domain-run citations, being the sum of cited runs over all 24,639 domains the Index observed. Those domains partition exactly into the nine source classes the Index assigns, each domain assigned to exactly one class, so the nine shares sum to 100 percent.
This matters because "share of AI citations" is the least comparable number in this field. A share of cited domains is a different quantity from a share of citations, which is different again from a share of responses containing a citation, and published figures for earned media's share of AI citations range across an order of magnitude mostly because they count different things. These figures describe the dated release, not the latest rolling Index. Domain classifications do not identify who commissioned an article, whether it was earned or sponsored, which brand benefited, or what a campaign cost. Those require placement-level evidence.
Two limits on reading it. First, 1,231 of the 1,340 editorial domains sit below the Index's evidence floor of ten observations across seven distinct days and carry the grade collecting rather than a confidence letter. The class totals are sums and hold; no individual tail domain's rate should be read as a stable rate. Second, class boundaries are our taxonomy's, not a law of nature, and one of them matters here: prnewswire.com sits in the editorial-publication class at rank 28 with 191 cited runs, not in wire distribution. Read the wire figure as the ten domains named below. Counting prnewswire.com with them raises the wire family to 448 cited runs, 0.36 percent, still the smallest class in the Index by an order of magnitude.
What Six Engines Actually Cite #
| Source class | Domains | Cited runs | Share | Median runs per domain |
|---|---|---|---|---|
| Other observed source | 20,917 | 76,790 | 61.97% | 2 |
| Editorial publication | 1,340 | 14,834 | 11.97% | 2 |
| Vendor-owned source | 823 | 11,807 | 9.53% | 3 |
| Market and company database | 718 | 6,391 | 5.16% | 2 |
| Community and social platform | 27 | 4,591 | 3.70% | 15 |
| Academic and government source | 431 | 4,229 | 3.41% | 2 |
| Analyst and consulting research | 364 | 3,468 | 2.80% | 2 |
| Search or media platform | 9 | 1,557 | 1.26% | 2 |
| Wire and press-release distribution | 10 | 257 | 0.21% | 6.5 |
The first row is the one most source-mix charts leave out. Nearly 62 percent of what these engines cite goes to the 20,917 domains the Index records in its uncategorized_source class — 85 percent of every domain observed being cited. A model of this market built from six or eight tidy buckets is describing the 38 percent that is easy to label, and a PR strategy built on that model is optimising against a minority of the citation surface.
The Wire Class Is Ten Domains, and That Is the Finding #
Wire and press-release distribution holds 0.21 percent of citations. The reflexive reading is that wire distribution underperforms. The data says something more specific and more useful.
| Wire domain | Cited runs | Citation rate | Days cited | Engines | Grade |
|---|---|---|---|---|---|
| businesswire.com | 106 | 0.59% | 55 of 135 | 6 | B |
| globenewswire.com | 79 | 0.44% | 41 of 135 | 5 | C |
| financialcontent.com | 37 | 0.21% | 25 of 135 | 4 | C |
| einpresswire.com | 17 | 0.10% | 11 of 135 | 2 | collecting |
| signalgenesys.com | 8 | 0.04% | 7 of 135 | 1 | collecting |
| prweb.com | 5 | 0.03% | 5 of 135 | 2 | collecting |
| clarityqst.com | 2 | 0.01% | 1 of 135 | 2 | collecting |
| 10bestpr.com | 1 | 0.01% | 1 of 135 | 1 | collecting |
| newsfilecorp.com | 1 | 0.01% | 1 of 135 | 1 | collecting |
| tuatupr.com | 1 | 0.01% | 1 of 135 | 1 | collecting |
Businesswire and GlobeNewswire are 72 percent of the whole class. Both are cited by five or six engines. Businesswire is cited on 55 separate days across the window. The median wire domain carries 6.5 cited runs against the median editorial domain's 2.
The wire class contains ten domains in this taxonomy, and several are repeatedly cited. That establishes observed use of those domains. It does not establish channel saturation or a ceiling on future use. A citation URL on a wire domain can still support a claim about the company in the announcement; measuring that benefit requires the answer text and exact cited placement. Domain ownership alone cannot settle brand impact.
Distribution reach, disclosure needs and AI answer use are different outcomes. This citation table measures the last of those at domain level; it cannot determine whether a release met its other purposes.
Owned Domains Are Cited More Than the Earned-Media Story Predicts #
The conventional framing sets earned media against owned media and gives owned media almost nothing. Our measurement does not support that.
Vendor-owned sources — company domains, product documentation, corporate research pages — hold 9.53 percent of citations across 823 domains. That is 45.9 times the entire wire class and 79.6 percent of the volume carried by all 1,340 editorial publications. The median vendor-owned domain carries 3 cited runs against the median editorial domain's 2.
Cross-engine reach separates them further. Of 1,340 editorial domains, 61 are cited by all six engines — 4.55 percent. Of 823 vendor-owned domains, 87 are — 10.57 percent, more than twice the rate. The domains doing this are ordinary: microsoft.com at rank 13 with 333 cited runs and grade A, ibm.com at 14, paloaltonetworks.com at 21, hubspot.com at 30, stripe.com at 41. That is owned material: documentation, product pages and company-published research that answers a question directly enough to be retrieved.
The honest version of the earned-versus-owned claim is therefore narrower than the usual one. Earned coverage is one route to citation among several. This dataset contains no campaign costs and cannot determine which route is more expensive. What earned coverage does that owned content cannot is supply an account of your company written by someone who is not you, which is the thing a buyer asking "is this vendor legitimate" is actually looking for. That is a different argument from a citation-share argument, and it survives this data. The citation-share argument does not.
Editorial's Share Is a Category Fact, Not a Constant #
"Earned media dominates AI citations" is a claim whose value moves with the category. Measured within each category's published strata, editorial's share of citations ranges from 5.4 percent to 17.7 percent, and vendor-owned's from 0.1 percent to 21.7 percent — and the two move independently.
| Category | Cited runs | Editorial | Vendor-owned | Community | Wire |
|---|---|---|---|---|---|
| Consumer finance | 6,489 | 17.7% | 2.5% | 4.3% | 0.0% |
| Family software | 3,424 | 16.4% | 0.3% | 4.0% | 0.1% |
| Cybersecurity | 9,772 | 14.6% | 16.5% | 3.6% | 0.3% |
| Healthcare services | 8,247 | 14.6% | 5.5% | 2.6% | 0.3% |
| Martech and advertising | 4,599 | 14.4% | 19.8% | 2.1% | 0.2% |
| HR and talent | 7,784 | 14.3% | 11.6% | 3.2% | 0.3% |
| Consumer products | 5,963 | 12.9% | 0.1% | 4.3% | 0.1% |
| Fintech | 8,372 | 10.1% | 11.9% | 2.5% | 0.5% |
| Enterprise software | 8,203 | 9.6% | 21.7% | 3.0% | 0.1% |
| Consumer health | 7,210 | 9.0% | 0.1% | 2.3% | 0.0% |
| AI infrastructure | 5,291 | 8.3% | 9.3% | 7.7% | 0.0% |
| Education and training | 3,878 | 5.6% | 2.2% | 5.4% | 0.0% |
| Deep tech and hardware | 4,112 | 5.4% | 0.9% | 6.6% | 0.2% |
Two patterns give PR teams hypotheses to test.
In consumer health and consumer products, vendor-owned sources hold one tenth of one percent of domain-run citations while editorial holds 9 to 13 percent. That is a reason to inspect the cited editorial pages and the brands they discuss. It does not establish what proportion of answers includes owned material or whether PR caused the editorial citations.
In enterprise software and martech, vendor-owned exceeds editorial — 21.7 against 9.6, and 19.8 against 14.4. Inspecting the cited vendor pages can show which documentation, research and product information answers those questions. The class totals alone do not identify the page types or establish competition between a brand's PR and its own pages.
Cybersecurity, fintech and HR have both classes in double digits. In all these categories, the useful next measurement is how the specific coverage and owned pages contribute to answers about the brand.
Editorial Citation Share Differs by Question #
The same measurement across question shapes is sharper than the category cut.
| Question shape | Cited runs | Editorial | Vendor-owned | Community |
|---|---|---|---|---|
| News topic | 45,188 | 16.4% | 16.8% | 3.1% |
| Best X | 14,049 | 12.5% | 6.4% | 3.7% |
| Top list | 14,178 | 10.0% | 5.7% | 2.8% |
| Is X worth it | 11,679 | 10.3% | 6.1% | 4.8% |
| X versus Y | 10,552 | 8.4% | 6.1% | 4.8% |
| Problem first | 9,688 | 8.1% | 5.1% | 6.1% |
| How to choose | 12,034 | 7.2% | 4.8% | 4.1% |
Editorial's share peaks at 16.4 percent on news-topic questions and falls to 7.2 percent on "how do I choose" — a 2.3-fold drop between the question PR is built to win and the question a buyer asks when they are ready to spend money. On how-to-choose questions, 76.1 percent of citations go to the unclassified long tail.
This is a useful distinction when designing a measurement plan. A campaign measured on coverage of your announcements is being measured on the question shape where editorial placement performs best and where purchase intent is weakest. The shift is not from PR to something else; it is from announcement coverage toward editorial that answers comparison and selection questions, which is a different pitch, a different outlet list and a different brief to the writer.
Most Citation Relationships Are Single-Engine #
Of all 24,639 cited domains, 16,749 — 68 percent — are cited by exactly one of the six engines. Only 368 domains, 1.49 percent, are cited by all six.
Within editorial publications the concentration is nearly as severe: 773 of 1,340 domains are cited by one engine, and 61 by all six. The measured domains therefore have uneven observed reach across engines; the reach of an individual placement must be checked at URL level. Measuring AI visibility as a single aggregate score hides which engine the coverage actually reached.
The engines differ because their retrieval paths differ, and each documents its own. Google states that its AI features draw on the same web index and standard eligibility as the rest of Search, and documents the eligibility requirements in its own guidance. OpenAI operates distinct crawlers for distinct purposes — a search crawler, a user-triggered fetcher and a training crawler — so a site can be readable to one and closed to another. Anthropic's Claude reaches the web through an explicit web-search tool that returns cited sources for the query it ran rather than browsing continuously. Perplexity exposes domain-level filtering to callers, making the source pool itself a parameter. Microsoft's Copilot family retrieves through the Bing index. Four different retrieval architectures produce four different source pools, which is what a 68 percent single-engine rate looks like from the inside.
The underlying mechanism is ordinary retrieval augmentation: a query is embedded, passages are retrieved from an index, and a generator writes an answer conditioned on them, as in dense passage retrieval and the retrieval-augmented generation architecture. That pipeline reads a press release as it reads any other page: a passage, its retrievability and the corpus it was drawn from — and web-scale corpora such as Common Crawl contain the wire pickup page and the trade article on equal terms, with the difference emerging at retrieval, not at ingestion.
What This Means for a PR Budget #
The discipline itself is unchanged. PRSA defines public relations as building mutually beneficial relationships between organisations and their publics; answer engines are now one of those publics, and they read everything the others read. What changes is where the money goes.
Inspect the category before allocating budget. In this sample, consumer health, consumer products and family software have low vendor-owned citation shares, while enterprise software and martech have higher vendor-owned shares. Use those differences to choose which sources and coverage to investigate. They do not establish the return on the next dollar spent on PR or owned content.
Pitch the selection question, not the announcement. Editorial's citation share nearly halves between news-topic and how-to-choose questions. Coverage that answers "how should a buyer in this category choose" is cited in the moment that decides a purchase; coverage of your funding round is cited in the moment that does not.
Treat each engine as a separate outcome. Two thirds of cited domains reach exactly one engine. Report coverage by engine reached; a single visibility score hides differences among the six measured engines.
Evaluate wire distribution at placement level. The ten-domain class records where citation URLs reside. Inspect whether an answer uses the release to describe or recommend the announced company before judging its AI-discovery value. Keep that outcome separate from the release's distribution and disclosure purposes.
Expect the traffic and the citation to come apart. Pew Research Center found that users who encountered an AI summary were less likely to click through to other sites than users who did not, in its March 2025 analysis of real browsing behaviour. Coverage can be cited into an answer read by thousands and send you nothing measurable in analytics. A PR programme judged only on referral traffic will conclude its best-performing work failed.
One boundary worth naming: everything above concerns earned and owned material. Paid placement is a different instrument with its own disclosure obligations, and the FTC's endorsement guides apply to it whether or not a machine is the reader. Coverage bought without disclosure is a legal exposure before it is a citation strategy.
How PR Affects AI Search Visibility in Machine Relations Terms #
Machine Relations is the practice of building the conditions under which machine readers recognise, trust and cite an organisation. PR is one instrument for creating that evidence. The 11.97 percent share belongs to domains classified as editorial publications in this release; it is not the share of citations caused by PR. Editorial domains publish different kinds of material, and brands can appear in answers through several source classes.
What the measurement also shows is that the surface is larger and stranger than the earned-media framing suggests. Sixty-two percent of citations go to domains the Index holds in its uncategorized class. Ten domains hold the entire wire class. Owned documentation is cited across all six engines at twice the rate editorial publications manage. The audience for all of it — as the Reuters Institute's Digital News Report tracks on the human side — is moving toward intermediated discovery, where what gets quoted matters more than what gets clicked.
The actionable finding is that source composition differs by engine, category and question. Measure a campaign against its own baseline and cited placements to determine what changed for the brand. The Index supplies context for that investigation, not an estimate of campaign lift.
What Changed in This Analysis #
This page was first published on 2026-07-23 built on figures licensed from third-party visibility vendors and PR firms. Two of those figures could not be reconciled with each other: an earned-media share of 84 percent and one of 37.6 percent appeared on the same page as though they described the same quantity, and a heading asserted a 420-to-1 ratio that the paragraph beneath it computed as 188 to 1. Neither number could be checked against its source, because both underlying datasets remain proprietary to the firms that sell them.
The analysis has been rebuilt on the Machine Relations Index, which publishes its release, its window, its evidence floor and its per-domain records so that every figure here can be verified against the Index. This is a different dataset and taxonomy, not a like-for-like revision of those external percentages. It shows substantial vendor-owned citation and variation across categories. Related re-sourcing of our analysis of citation rates by industry vertical is recorded there.
FAQ #
Does PR directly affect AI search visibility? #
PR can add coverage that an engine retrieves and cites. In the 2026-09-28 Index release, editorial-publication domains hold 11.97 percent of 123,924 domain-run citations. That source-composition finding does not measure a PR intervention; campaign impact requires a baseline, the actual placement and repeated answer observations.
Are press releases effective for AI visibility? #
Wire and press-release distribution holds 0.21 percent of citations across ten domains, of which businesswire.com and globenewswire.com are 72 percent. Those domains are cited — businesswire on 55 of 135 days, by all six engines. The cited URL belongs to the wire service; the answer may still use it as evidence about the company announced. Effectiveness for that company requires inspecting the placement and the answers that cite it.
Which type of PR generates the most AI citations? #
This dataset does not rank PR interventions. Editorial's citation share is 16.4 percent on news-topic questions and 7.2 percent on how-to-choose questions. To choose a campaign, identify the questions that matter to the buyer, inspect the pages those answers cite, and test whether new coverage changes the brand's presence. More citations and more commercially valuable citations are separate outcomes.
How should PR teams measure AI search visibility? #
Per engine, and against a stated denominator. Sixty-eight percent of cited domains are reached by exactly one of the six engines, so a single aggregate visibility score can hide differences among the six measured engines. Report which engines cited which placement, on how many distinct days, and state whether a share describes citations, cited domains or responses — those three quantities differ by an order of magnitude and are routinely quoted interchangeably.
Does owned content compete with PR for AI citations? #
In some categories it beats it. Vendor-owned domains hold 9.53 percent of citations overall — 79.6 percent of the volume carried by all editorial publications — and 10.57 percent of them are cited by all six engines against editorial's 4.55 percent. In enterprise software, vendor-owned holds 21.7 percent against editorial's 9.6. In consumer health it holds 0.1 percent. Check the category and actual answers before designing a campaign test; citation share alone does not establish comparative ROI.
October 7, 2026 interpretation clarification #
The dated tables are unchanged. The interpretation now separates domain citation share from PR campaign causality, cost efficiency and brand benefit. It also clarifies that a wire-hosted citation can support a claim about the announced company. The Index does not measure those placement-level outcomes from domain totals alone.