Not SEO
SEO tunes for ranking signals. MX provides the structural clarity SEO sits on top of. Build for MX and SEO crawlability improves as a side effect.
An Introduction to MX.
Every page, contract, and PDF a business publishes now has two audiences: the person who reads it and the machine that reads it. Most content was built for the first audience and is unreliable to the second. MX is the instructions that travel with the file.
In the eight days from 5 to 12 January 2026, Amazon, Microsoft, Google, and Anthropic each launched an AI agent with direct commerce or workflow capability. Adobe measured a 700% surge in AI-referred traffic to retail sites over the 2025 holiday period, with those visitors converting at 30% higher rates. The machines are reading now. They are buying.
What MX Is. What It is Not.
Not GEO
GEO targets citations in AI-generated answers. MX is what makes a file readable to the agent before it decides whether to cite it.
Not Accessibility Compliance
Accessibility compliance - the ADA in the US, the European Accessibility Act in the EU, WCAG elsewhere - addresses users with disabilities. MX addresses the same structural layer and accessibility improves as a side effect.
Two Pillars
MX - machine-readable
An open standard. Metadata that records a file's provenance, context, and intended use, written into the file so it travels with it. A machine that meets the file out of context still knows what it is.
REGINALD - machine-trustworthy
The registry layer on top of MX. Publisher-signed claims, tamper-evident records, verifiable evidence. Readable is not the same as trustworthy. REGINALD closes that gap.
Corroboration - machine-authoritative
A claim the reader will act on points to an authoritative source that independently states it. Readable, attributable, corroborated. Three properties, one artefact.
A Concrete Example: the Employment Agency Letter
Employment agencies send acceptance and rejection letters at scale. Many of those decisions involve automated systems. NYC Local Law 144 requires a bias audit and a notice when an automated decision tool is used. The decision letter also has to record that a human authorised it.
Accessible
A properly tagged PDF with a complete structure tree. A screen reader navigates it correctly. It passes the same checks an EAA conformance audit runs.
AI provenance
Which model contributed to the decision, what it was asked, who reviewed the output, the policy version in force at that moment, and the name of the accountable human.
Human-in-Command
The machine operated within a declared boundary, a named person holds authority over it, and the letter holds the signed record that proves it.
One PDF. The candidate reads it. The screen reader navigates it. The regulator verifies it. What the candidate can see under NYC LL144 and what the regulator can inspect under the AI Act are the same artefact.
The Demand for Proof is Global
United States
NIST AI Risk Management Framework (federal baseline), Colorado AI Act (state obligations), NYC Local Law 144 (automated employment decisions). Rules in force now.
Europe
EU AI Act Article 50 requires machine-readable marking on AI-generated content from 2 August 2026. European Accessibility Act applies to public-facing PDFs. UK ICO guidance is in place.
Everywhere else
Similar rules are arriving across Asia and the Gulf. The pressure is not regional and it is not pending. It is arriving from multiple jurisdictions at once.
None of this grants compliance; that stays the business's legal duty. An evidence layer gives any organisation the structured record every one of those regimes expects.
Where This Comes From
I worked on the BBC newsroom script collaboration tool. Many hands touch a script, it changes by the minute, and it has to be right when it goes to air. That work is where three properties took shape: provenance (where content came from and what happened on the way), accuracy (the discipline of getting it right under deadline), and attestation (a signed claim that a named party stands behind it). MX and REGINALD are those three lessons, generalised from a newsroom to any business that has to answer for what its content says and what its machines decide.
The practice is now assembled in three places: the MX book series (the standard and the protocols), The Gathering (the community-led standards body that governs the open parts), and the web-audit tool (which shows a team exactly what a machine sees when it reads their estate).
Further Reading
What is MX?
A plain-language introduction to machine-readable content and why it matters for both humans and agents.
MX is Not SEO, GEO, or ADA
How MX improves SEO, accessibility, and agent-readability as side effects of one implementation.
The Letter the Law Says You Have to Send
NYC Local Law 144, accessible PDFs, and Human-in-Command provenance in one artefact.
Tagged PDFs are MX
Why accessibility law and machine-readability demands converge on the same structure tree.
Who Answers When the Machine Decides?
The accountability gap AI-assisted decisions create, and the record that closes it.
MX: The Introduction (free)
The complete free book. Fill in the short form to download it and pass it to a colleague who wants the full argument.
The MX Book Series
The standard, the protocols, and implementation guides. The full series at mx.allabout.network/books/
The Gathering
The community-led standards body that governs the open MX standard. Anyone can contribute.
REGINALD
The evidence layer built on MX: tamper-evident, verifiable records for regulated industries.
Ready to make your content work for every machine that reads it? Get in touch or email info@cognovamx.com.