Framework
The Machine Relations Stack
The Machine Relations Stack has five layers, in build order: earned authority, entity clarity, citation architecture, distribution and measurement. Jaxon Parrott coined Machine Relations in 2024; each layer is a hypothesis to operate, measure and refine.
01Earned Authority
The foundation layer
Independent editorial coverage that gives AI answer systems third-party sources to evaluate — for example Forbes, TechCrunch, or The Wall Street Journal.
Why it matters
Observation boundary: Muck Rack's published editions reported earned-media shares of 89% in July 2025, 82% in December 2025, and 84% in May 2026. Fullintel/UConn separately reported that 47% of citations in its sample came from journalistic sources. Those are source-composition findings, not proof that earned coverage produces citations.
In practice
A fintech startup securing a TechCrunch feature. When someone asks ChatGPT "What are the best AI fintech tools?" the article becomes one independent source an answer system can retrieve, evaluate, and potentially cite.
When it is weak
The site is strong, but no independent source explains the brand's role, so an engine has nothing outside the brand to lean on.
How to check
Search the category plus the brand in the sources buyers trust. If only the brand's own site explains its role in the market, this layer is weak.
How it connects: Supplies third-party source evidence for Entity Clarity and Citation Architecture to organize, mark up, and test against real answer results.
Layer Evidence
02Entity Clarity
The identity layer
A consistent public record of who the brand is (name, founder, category and claims), so AI systems can resolve it as one distinct entity and describe it the same way across platforms.
Why it matters
Entity resolution is an operating hypothesis to test: consistent definitions, schema, knowledge graph records, and external profiles should reduce ambiguity when answer systems encounter the brand. It is not immediate confirmation and not an assurance of citation.
In practice
A SaaS company aligning Organization schema, a knowledge panel, Wikidata, Crunchbase, LinkedIn, and media references around the same name, category, and claims. Measurement then tests whether engines resolve the brand correctly.
When it is weak
Engines blend the brand with adjacent companies, old descriptions or generic category language, so it appears sporadically and is described inconsistently.
How to check
Ask ChatGPT, Perplexity and Gemini to describe the company, its founder, its category and what sets it apart. If the answers disagree materially, this layer is weak.
How it connects: Organizes earned and owned references into one coherent entity record. Gives Citation Architecture a clear brand identity to structure around and Measurement a resolution hypothesis to evaluate.
Layer Evidence
03Citation Architecture
The content layer
Structuring content so AI systems can extract and cite it — standalone statistics, one-sentence definitions, answer-first paragraphs.
Why it matters
AI answers usually cite specific passages, tables, definitions, and attributed statistics rather than entire articles. Clean fragments make extraction and attribution easier to observe; they do not promise that any engine will cite the source.
In practice
A cybersecurity company publishes an answer-first paragraph with a sourced statistic, a one-sentence definition, and a dated methodology note. The passage is easier to quote, verify, and measure than a buried narrative claim.
When it is weak
The content reads well but is too buried, narrative or vague for an engine to lift a clean, attributable passage.
How to check
Open the top pages and read the first paragraph under each heading. If those passages cannot stand alone as direct answers, this layer is weak.
How it connects: Uses Entity Clarity and Earned Authority as context, then structures evidence so Distribution can carry pages to answer surfaces and Measurement can test whether engines extract them accurately.
Layer Evidence
04Distribution
The distribution layer
Being present, crawlable and parsable on the surfaces answer engines retrieve from. Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO) are the practices inside this layer.
Why it matters
GEO and AEO make pages reachable, parsable, and answer-friendly across search and answer surfaces. They are distribution and optimization practices, not closed-engine levers and not assurance that a page will appear in any specific AI product.
In practice
A B2B SaaS company exposes comparison pages and research reports through crawlable HTML, structured metadata, updated sitemaps, and concise answer sections. Measurement then checks whether those pages appear in Perplexity or AI Overview results.
When it is weak
The brand appears on one engine and the team calls that AI visibility; buyers who use other engines never see it.
How to check
Run one fixed question set across the major engines. If the brand appears on one surface but not the others, this layer is weak.
How it connects: Carries earned authority, entity context, and citation-ready content into accessible surfaces so Measurement can observe retrieval, inclusion, citation, and attribution gaps.
Layer Evidence
05Measurement
The feedback layer
Tracking citation frequency, recommendation rate, and brand share of voice across AI platforms.
Why it matters
Impressions, AVE, and media mention counts measure reach to human readers. They do not measure AI citation frequency, recommendation rate, entity resolution, or share of voice across ChatGPT, Claude, and Perplexity — metrics that help evaluate whether a brand is appearing in answer surfaces.
In practice
A startup tracks 50 high-intent queries weekly, records which engines name or cite the brand, and compares those observations with competitor results. The data suggests which layer to investigate next rather than proving a single cause.
When it is weak
Work ships across the other four layers, but nobody can show which tracked questions moved, so the team repeats what it cannot see.
How to check
If the team cannot show citation and naming over time for a fixed, tracked question set, this layer is weak.
How it connects: Closes the loop by showing where the operating hypotheses held, where citation gaps remain, and what to test next.
Layer Evidence
How the stack compounds
Machine Relations is not linear. Each layer gives the next layer a clearer operating hypothesis to test: independent sources to evaluate, entity records to resolve, content fragments to extract, surfaces to crawl, and measurements to compare. The loop does not prove a closed-engine causal path or promise the next citation; it gives teams a disciplined way to improve source architecture from observed results.
The Flywheel