The Machine Relations Stack: Five Layers, in Order

The Machine Relations stack has five layers: earned authority, entity clarity, citation architecture, distribution and measurement. They depend on each other in that order, and the mistake I see most is working them from the top down. I coined Machine Relations in 2024, and the stack is how I think the discipline actually runs. The category reference for it is the Machine Relations stack at machinerelations.ai.

This page is my call on each layer: what it is, why it sits where it does and what I would do first.

What is Machine Relations, in one line?

Machine Relations is the evolution of PR, from earning trust with people to earning it with the machines people ask first. PR still works. The placement is raw material, and performance now means the machines repeat it. The stack is what turns a placement into something a machine repeats. My direct answer to the definition question is in What Is Machine Relations?

Why five layers, and why this order?

Each layer gives the next one something to work on.

An engine answering a buyer has to decide who to trust, which company it is looking at, what it can quote, where to retrieve it from and, after the fact, whether any of it worked for you. That is five questions, and the stack is the five answers. Skip the first and the rest have nothing to build on. Skip the second and good coverage lands on the wrong company.

This is also why the single-tactic labels fall short. GEO, AEO, AI SEO and LLMO each describe a slice, mostly in the middle of the stack. I use those terms where people search them, and I place them inside Machine Relations. The comparison is here.

Layer one: what is earned authority?

Earned authority is independent coverage in sources that engines trust, produced by editorial decisions that are not yours to make. It is the foundation, and it is the layer most teams want to skip.

Independent evidence and first-party content have different jobs. A company establishes its own account; outside reporting can examine and corroborate it. My PR and AI search essay separates that mechanism from what citation studies can actually establish.

My call: start here, and treat every placement as raw material. I do not claim a given placement causes a given citation. The evidence supports the pattern, not a guarantee for a single piece. I made the wider argument in Entrepreneur, PR worked for humans, and now it has to work for machines.

Layer two: what is entity clarity?

Entity clarity is a consistent public record of who you are, so a machine can tell your company, founder and category from everything with a similar name. The glossary names it entity optimization; I call it entity clarity because the point is for the machine to be clear, not for you to be optimized.

It matters because coverage only helps if it resolves to you. The metric for this layer is entity resolution rate: how often engines correctly identify your brand as one distinct company. Google documents how structured data helps it understand the entities on a page.

My call: fix inconsistency before you buy anything else. It is the cheapest layer to repair and the most expensive to ignore.

Layer three: what is citation architecture?

Citation architecture is building your own content so a machine can lift a clean, attributable answer out of it: answer-first paragraphs, claims that carry sources, clear structure and modular sections.

Research suggests this is a two-part job. Zhang et al.'s study of citations across ChatGPT, Google AI Overviews and Perplexity found that being selected as a source and having your language absorbed into the answer are separate stages. A page can win the first stage and still not be quoted.

My call: write the answer first and put the evidence next to it. Make the claim and its supporting evidence understandable without forcing the reader to reconstruct the whole article.

Layer four: what is distribution?

Distribution is being present on the surfaces engines retrieve from, with content they can access and crawl. This is where GEO, AEO and SEO do their work.

None of them is optional. Google states that its AI features have no additional requirements beyond SEO best practices, which tells me the base layer still matters and nothing replaces it. The original GEO research paper reported visibility gains from certain content changes in its benchmark, with results that varied by domain.

My call: do this layer well and do it after the three beneath it. Distribution without earned authority is polishing a claim nobody has corroborated.

Layer five: what is measurement?

Measurement is knowing whether engines name and cite you when buyers ask. The metric I use is share of citation, the percentage of the cited-source pool your brand occupies in a set of answers, measured under a disclosed question set, engine panel and window.

My call: disclose the method every time. A number with no stated method is a mood. And measure the same set over time, because a change you cannot reproduce is noise.

What should I do first?

My call, in order:

1. Ask the engines your buyer's question. Not your brand name. Write down who gets cited.
2. Check identity. Do the engines agree on who you are? Fix that first.
3. Look at your earned coverage. Is independent coverage describing you in your buyer's terms?
4. Then fix your own pages and distribution.
5. Measure the same set again, on the same terms.

Gartner predicted in 2024 that traditional search engine volume would fall 25% by 2026 as chatbots and virtual agents take over queries. The stack is the work that follows from that shift.

FAQ

What are the five layers of the Machine Relations stack?

Earned authority, entity clarity, citation architecture, distribution and measurement. They depend on each other in that order.

Where do GEO and AEO sit in the stack?

Mostly in citation architecture and distribution. They are practices inside Machine Relations, which is the umbrella over them, AI SEO, LLMO, AI PR and digital PR.

Which layer should I start with?

Earned authority, after a quick identity check. Everything above it depends on independent coverage, and coverage only helps if engines can attach it to the right company.

Who coined Machine Relations?

I did, in 2024. The short answer is at who coined Machine Relations.

Is the stack a product?

No. It is a framework for the discipline, published at machinerelations.ai, the discipline's reference home. AuthorityTech, which I founded in 2018, puts the discipline into practice.