The Pages AI Engines Cite on AI-Visibility Questions Say 'AI Visibility' (95%) and 'GEO' (83%). None of 613 Third-Party Pages Says 'Machine Relations'

The pages AI engines cite when buyers ask about AI visibility say "AI visibility", "GEO" and "AEO". They do not say "Machine Relations". We took the cited pages from the Machine Relations Index answer ledger, fetched them on 2026-11-08 and searched the text. Of the 350 third-party pages cited on at least three of the Index's 42 AI-visibility and GEO buying questions, 95.4 percent used "AI visibility", 82.6 percent used GEO or "generative engine optimization", 73.1 percent used AEO or "answer engine optimization", and 0 used "Machine Relations". Of a random sample of 263 other cited third-party pages, 0 did either. That is 0 of 613 pages, with an upper bound of 0.6 percent.

This lines up with what the engines themselves say. In their answers, none of 36,836 answers to questions that had not named the term used it. The pages they read do not use it either. This page counts the words on cited pages. It does not show that the sources cause the answers' wording, and it does not say which term should win.

What was counted #

• Population. Successful runs in the Index's buying-question panel on the AI-visibility and GEO vertical (42 questions: tools, agencies, comparisons and how-to-choose questions), runs not flagged as naming a brand in the prompt, ChatGPT, Claude, Gemini, Google AI Mode and Perplexity (Google AI Overviews left out), run dates through 2026-11-07. That is 12,965 cited-URL events from 1,350 answer runs and 6,004 distinct addresses. The ledger export is dated 2026-11-08 08:10 UTC.
• Top sample. Every address cited on at least three of the 42 questions, excluding YouTube, Reddit, LinkedIn, Facebook, X and Quora (816 events, 6.3 percent of the total, which are video and social posts and were not read as pages). That is 380 addresses. 351 were readable and 350 were third-party. They carry 3,323 of the 12,149 non-social citation events (27.4 percent) and sit on 209 domains.
• Tail sample. 300 addresses drawn at random, fixed seed, from the 5,159 non-social addresses outside the top sample. 265 were readable and 263 were third-party.
• Third-party. The three authoritytech.io pages that fell into the samples are counted separately below. No machinerelations.ai page fell into either sample.
• Reading a page. Each address was fetched once on 2026-11-08, between about 15:36 and 15:42 UTC, with a plain HTTP request and a browser-style user agent. Scripts, styles and tags were stripped, the text was lower-cased, and pages with under 2,000 characters of text were set aside. Pages that blocked the request (403, 429), failed to connect or returned an error were set aside too: 29 of 380 in the top sample and 35 of 300 in the tail.
• Matching rules. "Machine Relations" is the phrase machine relations or the domain machinerelations.ai. "AI visibility" is ai visibility. GEO is the word geo or generative engine optimization (either spelling). AEO is the word aeo or answer engine optimization. Earned media is earned media. Digital PR is the phrase digital pr.

What the cited pages call the work #

Share of third-party pages that contain each term at least once (95 percent Wilson intervals):

In the top sample, weighting each page by how often it was cited changes little: the pages carrying "AI visibility" account for 96.4 percent of the 3,323 events, GEO 83.0 percent, AEO 75.9 percent and Machine Relations 0. 212 of the 350 pages use all three of AI visibility, GEO and AEO.

The tail sample reads lower on every term because it is less on-topic. A page cited once on one question is more likely to be a general page that touches AI visibility in passing.

The neighbors of the discipline #

Digital PR appears on about one top-sample page in six (16.0 percent) and earned media on about one in eight (12.9 percent). Those are the nearest ideas to Machine Relations in the sources, and they sit beside the dominant vocabulary, not in place of it. The cited pages are mostly software listicles and measurement guides: the most-cited addresses in the vertical are ranked lists of AI-visibility tools.

Where AuthorityTech appears #

Three authoritytech.io pages were cited often enough to land in the samples (one in the top sample, two in the tail). All three use the phrase "Machine Relations". They are the only pages among the 616 read that do.

Two third-party pages name AuthorityTech without using the phrase. One cites "Authority Tech's 2026 study" for a Perplexity conversion figure and the other credits AuthorityTech for average-sources-per-answer figures in a benchmark table. That is 2 of 613 pages (0.3 percent).

Neither page shows a third party using AuthorityTech's own measurement. We checked both against AuthorityTech's published articles on 2026-11-08. The 11x conversion gap is Microsoft Clarity's finding for AI-referred visitors in general (1.66 percent sign-ups against 0.15 percent for search), not a Perplexity-only figure and not an AuthorityTech study. Until a correction on 2026-11-08, though, two AuthorityTech pages (a February 2026 article and a curated benchmark) had stated it as Perplexity's, and the third-party page's wording is close to the February article's. We cannot tell whether that page took the figure from AuthorityTech, but the Perplexity framing was AuthorityTech's to correct, and it has been. The sources-per-answer figures (ChatGPT 6.88, Google AI Overviews 12.06, Perplexity 16.35) appear in an AuthorityTech article that takes them from a 2026 academic paper on 602 prompts and says so. These pages show AuthorityTech's name travelling with other people's numbers, which is reach, not yet credit for original research.

Correction, 2026-11-08. An earlier version of this section said the 11x figure in AuthorityTech articles was Clarity's finding, without saying that two AuthorityTech pages had presented it as Perplexity's. This version says so. The counts (0 of 613 pages using the phrase, 2 of 613 naming AuthorityTech) are unchanged.

What this does not show #

• No cause. Pages that use a word are not the reason an answer uses it. The ledger cannot say whether an engine would repeat "Machine Relations" if more cited pages carried it. The Index finds that engines differ widely in how much they copy from cited text, so the link between source wording and answer wording is expected to differ by engine too.
• Not the whole web. This is one vertical's cited pages on 42 questions, not a crawl of the web. The phrase may appear on pages the engines never cited here.
• Not the whole page set. The top sample covers 27.4 percent of non-social citation events. The tail is a random 300 of 5,159 addresses, so it estimates the tail's share and does not enumerate it.
• Not what the engine retrieved. A plain fetch today is not the page the engine read on the day it cited it. Pages may have changed since the run, and a page that builds its text in the browser would read as empty to this method, which is why short pages were set aside.
• Loose matches. GEO as a word can match geography terms ("geo-targeting"), so the GEO share is an upper estimate. "Digital PR" and "earned media" are exact phrases and miss variants.
• Not buyer demand. It does not say what buyers type or which term they would choose.

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 run table (successful runs only) and the question table (buying-question panel, the AI-visibility and GEO vertical). Filters: runs not flagged as brand-prompted, a non-empty cited URL, engines ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, run dates through 2026-11-07. Unit: distinct (answer run, cited URL) pairs, 12,965. Top sample: addresses cited on three or more distinct questions, ordered by question count then events, YouTube, Reddit, LinkedIn, Facebook, X and Quora excluded, first 380. Tail sample: Python random.sample with seed 20261108 over the remaining 5,159 non-social addresses, 300 drawn. Pages were fetched by 24 parallel curl requests with a 20-second timeout and redirects followed. Intervals are Wilson 95 percent intervals on page counts, treating pages as independent; several pages sit on one domain, so the intervals are a little narrow. The upper bound for 0 of 613 is the Wilson upper limit.

Frequently asked questions #

What do the pages AI engines cite call GEO and AI visibility? #

On the 350 third-party pages cited on three or more AI-visibility and GEO buying questions, 95.4 percent used "AI visibility", 82.6 percent GEO or generative engine optimization, and 73.1 percent AEO or answer engine optimization.

Do the pages AI engines cite use the phrase Machine Relations? #

Not third-party pages in this sample. 0 of 350 top-sample pages and 0 of 263 tail-sample pages used it. The only pages that did were three authoritytech.io pages.

Is that why AI engines do not say Machine Relations? #

The two facts agree, and this page does not show that one causes the other. Engines also left the term out of 36,836 answers to questions that had not named it, but the ledger cannot say whether more cited pages carrying the phrase would change that.

Do cited pages talk about earned media or digital PR? #

Some do. Digital PR was on 16.0 percent of top-sample pages and earned media on 12.9 percent. They are the nearest neighbors of the discipline in the sources, and they are a minority vocabulary.

Can this be checked again? #

Yes. The same method can be rerun on a later ledger export, with the same pages re-read. Any later 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