A Third Party Defined Machine Relations Without Citing Us. That Is the Exact Problem Machine Relations Exists to Solve.

On July 21, a PR firm published a guide titled "What Is Machine Relations?" They defined the discipline I coined, explained it well, and cited zero of the research that produced it. That is not a grievance. It is a live demonstration of exactly how AI attribution chains break in practice.

The Signal That Matters More Than the Slight

I built Machine Relations because I watched traditional PR collapse against an audience of machines. I published the research. I built the Machine Relations Index, which tracks 15,896 cited domains and 76,087 source events across AI engines.

Then Instant Press Co. published "What Is Machine Relations? (PR for the AI Era)" on July 21, 2026. Their founder defined the concept as "the practice of shaping how AI systems understand, trust, and cite your brand when they answer questions about your category."

That definition is solid. The framework they outline: identity, answerability, corroboration, freshness. Reasonable. There is one problem. They wrote it as though they originated it. No mention of AuthorityTech. No link to the MRI. No citation of the research that produced the concept.

My first reaction was frustration. My second was recognition.

This is what category creation actually looks like in 2026. Someone else explains your idea to their audience, and the machines learn the category from the copy.

85% of Brand Mentions Come From Third-Party Pages

This is not a one-off event. According to AirOps' 2026 State of AI Search report, 85% of brand mentions in AI-generated answers originate from third-party pages, not owned domains. An Omniscient Digital analysis of 23,387 AI citations found that review sites, listicles, forums, and case studies capture 57% of branded evaluative query citations.

When someone asks ChatGPT or Perplexity "what is machine relations," the engine is more likely to pull from a third-party explainer than from the source that built the concept.

This is not a bug. A breakdown of the LLM citation pipeline by TheStacc maps the four stages: retrieval, ranking, extraction, attribution. The best public measurements point at brand-level signals: Kevin Indig's analysis of more than 7,000 citations found brand search volume correlated with AI chatbot mentions at 0.334, and Ahrefs' study of 75,000 brands put branded web mentions at 0.664 against 0.218 for backlinks for brand visibility in Google AI Overviews. Brand strength in this context is measured by how widely a brand name is mentioned across the web, with YouTube mentions an even stronger correlate in Ahrefs' later cross-platform study. Neither study compares originators with explainers, so what follows is my reading, not their finding: in these data, visibility tracks how widely a brand is mentioned, which gives no support to "coined it first" as a signal. Semrush confirmed the same pattern: when AI engines choose between a first-party definition and a third-party explainer, the engine defaults to whoever has more independent corroboration.

There is a measurable reason this happens. According to TheStacc's analysis, heavily cited content averages 20.6% entity density compared to 8-12% for low-cited content. Pages with direct answers, clean headers, and structured data get cited 2-3x more often. The third-party explainer who writes clearly about your category with proper markup will outperform your origin page if your origin page was not built for extraction.

The irony is load-bearing: the discipline I built to solve this exact problem is experiencing this exact problem.

Citation Laundering Is Real. A German Court Just Named It.

Entidex documented a pattern they call "citation laundering": an AI system ingests authoritative data, repackages it as synthesized knowledge, and attributes it to unrelated sources or to no source at all. The numbers are real. The cited sources are not where the numbers came from.

On May 28, 2026, the Regional Court of Munich (LG Munich I, case 26 O 869/26) issued a preliminary injunction against Google that established a principle worth memorizing: synthesis is authorship. When an AI engine evaluates and combines sources into independent statements, that output constitutes the operator's own speech. Search engines cannot claim intermediary protection when generating summaries.

The ruling is German, preliminary, and under appeal. The principle it names is global. When AI synthesizes knowledge, the responsibility for accurate attribution sits with the system operator. Not with the sources it consumed.

For founders building categories, this means the legal ground is shifting under a system that already cannot trace citations back to their origin.

What a Working Citation Chain Requires

The InstantPress article lists identity, answerability, corroboration, and freshness as the framework. Those are the right words. They missed the mechanism that makes them work: a citation chain that traces back to the source.

Four things make a citation chain hold:

Source authority. The original research, data, and definitions live on a domain you control, marked with structured data that identifies the creator, dataset, and methodology. The MRI entity page includes schema.org/Dataset markup identifying the measurement window, cited domains, and API endpoint. When an AI system encounters those signals, it has a machine-readable provenance trail.

Third-party corroboration that cites back. Independent publishers writing about your concept must link to the source. Not out of courtesy. Because the retrieval system uses the link graph to determine who originated what. When 10 third-party articles explain "machine relations" and zero of them link to the research that defined it, the AI engine has no provenance signal. It assigns authority based on clarity and recency. Not origin.

Entity-level consistency. Every mention of the concept across owned properties, guest publications, and structured data must use the same terminology, the same entity references, the same definition. The gap between what you claim and what external sources say acts as a confidence signal for the retrieval model. When the gap is zero, models cite with higher confidence.

Freshness. Stale content loses citation priority. 65% of AI bot traffic targets content published in the last 12 months. A concept defined once and never updated falls out of the retrieval window within months.

The Founder's Move

If you are building a category, hear this: the idea spreading is not the win. The citation chain holding is the win.

Go search "what is machine relations" in ChatGPT, Perplexity, and Google AI Mode right now. Look at what gets cited. Look at whether the citation traces back to the originator or to whoever explained it most recently with the clearest markup.

Then ask yourself the question that applies to every category you are building: if someone published a guide explaining your concept next week, would the AI engines trace it back to you? Or would they learn it from the explainer?

The answer tells you whether you have a category or a catchphrase.

FAQ

What is citation laundering in AI search?

Citation laundering occurs when an AI system ingests data from an authoritative source, repackages it, and attributes it to a different or no source. Entidex documented this pattern, and a German court ruling (LG Munich I, May 2026) established that synthesis constitutes the operator's own speech, not intermediary pass-through.

How do AI engines decide who originated a concept?

They do not read founding dates or press releases. They evaluate the link graph, structured data markup, third-party corroboration, and content freshness. According to AirOps' 2026 analysis, 85% of brand mentions in AI answers come from third-party pages. The measured correlations (Indig's 0.334 for brand search volume, Ahrefs' 0.664 for branded web mentions) point at brand strength built from mentions across the web; neither study tests who said it first, which is my inference. The originator gets credit only when the citation chain traces back through independent sources.

What should founders building a category do about AI attribution?

Build the citation architecture before the category spreads. Mark origin content with structured data (schema.org/Dataset, schema.org/DefinedTerm). Get independent publishers to cite back to the source. Keep definitions consistent and fresh. The Machine Relations Index tracks how these citation chains perform across 15,896 domains.