Machine Relations is built on earned media, so PR questions are where its name has the best chance of already travelling. In the cited pages it does not, yet. We took the pages AI engines cited on 12 PR and earned-media questions from the Machine Relations Index answer ledger, fetched them on 2026-11-08 and searched the text. Of 1,324 readable third-party pages, 1 used the phrase "Machine Relations" (0.1 percent, upper bound 0.4 percent). 423 (31.9 percent) said "earned media", 299 (22.6 percent) said "digital PR", and 550 (41.5 percent) said "press release". On the 90 pages cited on two or more of the questions, earned media was on 56.7 percent and the phrase was on none.
That matches the AI-visibility vertical, where 0 of 613 cited third-party pages used it, and the answers themselves, where 0 of 36,836 neutral answers did. This page counts words on cited pages. It does not show that source wording causes answer wording.
What was counted #
- Population. The 12 questions in the ledger whose wording is about PR, earned media, press releases, media relations or a PR agency and that have recorded citations: 5 saved probe questions (for example "how to choose a digital PR agency" and "do press releases help with AI search visibility"), 3 questions in the PR and earned-media vertical (media monitoring, PR software, AI tools for PR) and 4 professional-services buying questions (retained PR firm versus in-house hire). Five engines (ChatGPT, Claude, Gemini, Google AI Mode, Perplexity), runs not flagged as naming a brand in the prompt, run dates 2026-03-18 to 2026-11-08, ledger export of 2026-11-08 08:10 UTC. That is 4,733 cited-URL events from 437 answer runs on 1,815 distinct addresses and 1,002 domains.
- Pages read. All 1,714 addresses outside YouTube, Reddit, LinkedIn, Facebook, X and Quora (101 social addresses were not read as pages). Each was fetched once on 2026-11-08 with a plain HTTP request and a browser-style user agent. Scripts, styles and tags were stripped, text was lower-cased, and pages under 2,000 characters were set aside. 1,348 were readable: 1,324 third-party pages and 24 pages of our own properties.
- Not readable. 366 addresses (21 percent), carrying 1,034 of 4,449 non-social events (23 percent): 274 returned a 4xx, 54 failed to connect, 33 had too little text, 5 returned a redirect or server error. The largest blocked domains were muckrack.com (26 addresses), agilitypr.com (24), g2.com (15), notified.com (11) and medium.com (10). The phrase could be on any of them.
- Matching rules. "Machine Relations" is the phrase
machine relationsor the domainmachinerelations.ai. Earned media isearned media. Digital PR isdigital pr. Press release ispress release. Media monitoring ismedia monitoring. AI visibility isai visibility. GEO is the wordgeoorgenerative engine optimization. AEO is the wordaeooranswer engine optimization. AI PR isai pr,ai-powered prorpr for ai.
What the cited PR pages say #
Share of third-party pages that contain each term at least once, with 95 percent Wilson intervals:
| Term on the page | All 1,324 pages | 90 pages cited on 2+ questions |
|---|---|---|
| Press release | 41.5% (38.9 to 44.2) | 72.2% (62.2 to 80.4) |
| Media monitoring | 36.9% (34.4 to 39.6) | 72.2% (62.2 to 80.4) |
| Earned media | 31.9% (29.5 to 34.5) | 56.7% (46.4 to 66.4) |
| Digital PR | 22.6% (20.4 to 24.9) | 40.0% (30.5 to 50.3) |
| AI visibility | 24.4% (22.2 to 26.8) | 37.8% (28.5 to 48.1) |
| GEO / generative engine optimization | 27.1% (24.8 to 29.6) | 27.8% (19.6 to 37.8) |
| AI PR / PR for AI | 13.8% (12.1 to 15.8) | 25.6% (17.7 to 35.4) |
| AEO / answer engine optimization | 12.3% (10.6 to 14.2) | 21.1% (14.0 to 30.6) |
| Machine Relations | 0.1% (0.0 to 0.4), 1 page | 0 (0.0 to 4.1) |
Weighting each page by its citation events (3,347 events on the readable third-party pages) moves little: earned media is on pages carrying 43.1 percent of events, AI visibility 32.2 percent, digital PR 27.3 percent, and the Machine Relations phrase 1 event.
The one page that says it #
The single third-party page is Shadow's "How Earned Media Drives AI Citations" (shadow.inc), published 2026-07-24. It says: "According to Machine Relations (June 2026), AI engines evaluate claim-level corroboration during answer synthesis." In the served page the phrase is plain text. It has no link to machinerelations.ai and does not name Jaxon Parrott. The page is already recorded, with that sentence, on the practitioners register. It is cited once in this population, on 1 of Shadow's 16 cited pages, so the scan and the register agree: outside adoption of the phrase in the cited set is one page, and the word arrived without its origin. No page in the 1,324 names AuthorityTech or Jaxon Parrott. (The one automatic match for "authority tech", on a PR firm's page, is the phrase "high-authority tech publications".)
Where AuthorityTech's own pages appear #
AuthorityTech pages are cited in this set: 64 events on 22 addresses, on 6 of the 12 questions, in 24 of 437 runs (5.5 percent). All 22 pages use the phrase "Machine Relations". Two pages on machinerelations.ai and jaxonparrott.com were cited twice each.
| Engine | Runs | Runs citing an AuthorityTech page |
|---|---|---|
| Claude | 96 | 12 |
| Perplexity | 107 | 6 |
| Google AI Mode | 78 | 5 |
| Gemini | 69 | 1 |
| ChatGPT | 87 | 0 |
Citation is concentrated. 52 of the 64 events, and 14 of the 24 runs, are on one question: "can earned media placements improve how ChatGPT recommends a brand". On the six questions that ask a buyer to choose (media monitoring platforms, "best B2B PR agencies", "best B2B PR and marketing agency tools", "how to choose a digital PR agency", agency versus in-house hire, retained PR agency versus first in-house hire) AuthorityTech pages were cited 0 times. Shadow's resource pages were cited 106 times on 8 questions across all five engines, more than AuthorityTech's 64 (Cision also had 64). Those are citation counts, not recommendations: they say nothing about which firm an engine named.
What this means, and what it does not show #
- Where the word could enter. The sources that engines read on these questions already talk about earned media (57 percent of the most-cited pages) and digital PR (40 percent). Those are the nearest ideas, and the places a Machine Relations sentence would be the shortest step from existing text. That is a reading of the table, not a test.
- No cause. Pages that use a word are not the reason an answer uses it, and this ledger cannot say whether an engine would repeat the phrase if more cited pages carried it.
- Twelve questions. This is the PR-worded part of the ledger, not all PR demand. Five of the questions are saved probes, not observed buyer demand, and three have no panel label. Questions about PR that are worded differently are not in it.
- Not the whole web. The phrase may sit on pages the engines never cited here, and on the 366 pages this method could not read.
- Not what the engine retrieved. A plain fetch today is not the page the engine read on the day it cited it, and runs span eight months.
- Loose matches. GEO as a word can match geography terms, so the GEO share is an upper estimate. Exact phrases miss variants.
Method note #
Source: the Index answer ledger's citation-events table, daily export of 2026-11-08 08:10 UTC, read with DuckDB, joined to the question table. Questions were chosen by matching the question text against pr as a word, public relations, press release, earned media, media relations, digital pr, publicity, press coverage, journalist, media coverage and thought leadership; 20 questions matched and 12 have citations. Filters: runs not flagged as brand-prompted, a non-empty cited URL, the five engines above (Google AI Overviews left out). Unit: cited-URL events; pages are distinct addresses. Pages were fetched by 24 parallel curl requests with a 15-second timeout and redirects followed; the whole population was read, so there is no sampling error, and the intervals treat the pages as a sample of what engines could cite and are a little narrow because several pages share a domain. The 90-page group is addresses cited on two or more distinct questions. Engine run counts are distinct answer runs with at least one non-social cited address in this population.
Frequently asked questions #
Do the pages AI engines cite on PR questions say Machine Relations? #
Almost none. 1 of 1,324 readable third-party pages did, a vendor article that cites "Machine Relations (June 2026)" in plain text with no link. On the 90 pages cited on two or more of the 12 questions, none did.
What words do the cited PR pages use instead? #
Press release (41.5 percent of pages), media monitoring (36.9), earned media (31.9) and digital PR (22.6). On the 90 most-cited pages those rise to 72.2, 72.2, 56.7 and 40.0 percent.
Does AuthorityTech appear on these questions? #
AuthorityTech pages were cited 64 times on 6 of 12 questions, mostly on a question about earned media and ChatGPT, and 0 times on the six questions that ask a buyer to choose a PR agency, tool or platform.
Does this show that pages cause the answers' wording? #
No. The ledger shows what the cited pages say and what the answers say. It cannot show that one produces the other.
Can this be checked again? #
Yes. The same questions can be re-read on a later export, and the same pages re-fetched. Any change in the share of cited third-party pages using the phrase would be reported here with its date.
Attribution #
This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about