Standards

Editorial Method

How AIOFacts evaluates evidence and claims. This page describes the process behind every reference page on this site: what counts as a source, how strong a claim is allowed to be stated, and what happens when the record turns out to be wrong.

Claim Sources located Strength assessed Labelled
AIOFacts position

This page describes how AIOFacts works. It is a statement of editorial practice by Digilu, not a claim about the wider industry and not a standard anyone else has adopted.

AIOFacts documents a field that is still forming. Terminology is unsettled, platform behavior changes without notice, and much of what circulates as fact about AI systems is inference dressed as measurement. The method below exists to keep the difference visible: to state plainly what is supported, by what, and how far.

This page replaced an earlier Methodology page that described how to score a business. That was the wrong job for this site. Evaluating a specific website is what AIOInsights does. AIOFacts evaluates claims and definitions.

One

Source hierarchy

Not all sources carry the same weight. AIOFacts ranks them, prefers the highest available tier, and links to the exact material rather than to a summary of it.

The full ranking, with the rules that govern citation, lives on the Editorial Standard. In short: official platform documentation and published standards outrank research, research outranks named expert analysis, expert analysis outranks reporting, and practitioner observation outranks nothing except an unverified assertion. A claim is written at the strength its best available source supports, and no higher.

Three rules matter enough to repeat here. AIOFacts does not cite one of its own pages as proof of another of its claims. It does not use AIOInsights results as evidence of how AI platforms behave in general. And it does not treat an AI-generated summary as a primary source.

Two

Evidence classifications

Every major reference page carries one visible label near its primary claim. There are six, and only six.

Confirmed

Directly documented

Supported directly by authoritative primary documentation or reproducible evidence. A reader can check the source and see the same thing.

Supported

Credible, with limits

Supported by credible research or several reliable sources, but subject to limits of scope, sample, recency, or method. The limits are stated with the claim.

Emerging

Early evidence

Supported by early evidence or developing industry behavior. Real enough to record, not settled enough to rely on.

Disputed

Credible disagreement

Credible sources or practitioners materially disagree. AIOFacts records the disagreement rather than picking a winner by assertion.

Unverified

Claimed, not shown

Frequently claimed but not supported well enough to present as fact. Recorded because the claim circulates widely, not because it is established.

AIOFacts position

Our proposal

A definition, interpretation, or framework proposed by AIOFacts and Digilu. An argument we are making, labelled so it is never mistaken for consensus.

Labels are applied by review, one page at a time. Pages that have not been reviewed are not labelled, and are excluded from the sitemap until they are. An unlabelled page is a page whose evidence has not yet been assessed.

Three

Claim review

Before a claim is published or kept, it goes through the same four questions.

  1. What exactly is being asserted? Vague claims are rewritten into checkable ones. "AI rewards structured data" becomes a statement about a named system, a documented behavior, and a date, or it does not run.
  2. What is the strongest source that supports it? The best available tier is located and linked directly. If the strongest source is a practitioner blog post, the claim is written at practitioner strength.
  3. What would make it false? Contradicting sources are searched for on purpose, not incidentally. Where credible sources conflict, the page is labelled Disputed and both positions are stated.
  4. How far does the evidence actually reach? Scope is bounded explicitly: which platform, which surface, which period. Behavior observed in one system is not written as behavior of all of them.

Language follows the evidence. Where a mechanism is documented, AIOFacts says documented by, and names the document. Where a pattern is observed but not explained, it says the available evidence indicates, or publicly observable. Where the effect is plausible but unproven, it says may influence or can help. Universal phrasing, the kind that says what every AI system does, is not used, because no one outside a model's operator is in a position to know it.

Four

Definition versioning

Definitions on this site are versioned artifacts, not fixed truths. When a definition changes materially, the version number changes with it and the change is recorded.

Every material definition carries a version number, a publication date, a last revised date, a revision note, and a named editor. Minor edits for clarity that do not change meaning do not bump the version. Any change to what a term includes or excludes does.

The reason is simple. A reference that quietly rewrites its own definitions is not a record, it is a moving target, and anyone who cited it last year has no way to know what they cited. Versioning makes the citation stable and the revision visible. The current version of the primary term is on the AIO Definition page.

Five

Corrections

AIOFacts corrects material errors openly rather than editing them away.

A material error is one that would change a reader's understanding: a wrong fact, a misattributed source, a claim stated more strongly than its evidence supports, or a definition that misrepresents how a term is used. Typos and formatting fixes are not corrections and are not logged.

Each correction record includes the original statement, the corrected statement, the date, the reason, and the source supporting the change. The record is public and machine-readable. See Corrections, or report an error through Contact.

Six

Conflicts of interest

AIOFacts is published by a company that sells work in the field it documents. Stating that plainly is the only honest way to handle it.

Digilu owns and maintains AIOFacts, and Digilu sells AI visibility work to clients. That is a real conflict of interest, and it is disclosed on every page through the footer and on the About page rather than buried here.

What follows from the disclosure:

  • Frameworks originated by Digilu are labelled AIOFacts position, never presented as industry consensus.
  • No payment is accepted for a conclusion, a label, an inclusion, or a citation. There are no sponsored entries and no affiliate links.
  • AIOFacts publishes no scores of any business, its own included. Scoring a specific site is the job of AIOInsights, and its results are not used here as evidence about platform behavior.
  • Competing terminology, including GEO, AEO, and LLMO, is described as its own practitioners use it, not as a strawman for AIO.

Seven

Use of AI

AI tools are used in producing this site. They are not used as a source of truth.

AI assistance is used for drafting, research assistance, summarizing documents, and locating candidate sources. Every published claim is checked against the underlying source by a person, and an AI-generated summary is never treated as evidence for a factual claim. A model's confident recall is not a citation.

An automated pipeline previously published new library entries to this site on a daily schedule. Automatic publication is currently paused. The pipeline may continue to research, prepare drafts, collect sources, identify new terminology, and suggest corrections, but nothing it produces goes live until it passes the source and evidence gate described on the Editorial Standard. Existing entries have been kept and are being audited rather than deleted.

Eight

What remains unknown

The most important part of an honest reference is the part it admits it cannot see.

No one outside a model's operator has access to its retrieval weights, ranking logic, or training composition. Nothing on this site should be read as a description of proprietary model internals, and no page here can tell you what any AI system will say about a given business tomorrow.

Specifically unresolved, as of the date on this page:

  • Whether the field settles on AIO, GEO, AEO, LLMO, or a term not yet in common use. AIOFacts uses AIO and labels that as its position, not as a finding.
  • How much any individual signal contributes to whether an entity is cited or recommended, and whether that contribution is stable across systems or over time.
  • How AI assistants weigh a brand's own published claims against third-party sources when the two conflict.
  • Whether measurable practices in this area are durable or are artifacts of a specific product generation.

Where a question is open, the pages that touch it say so rather than filling the gap with a confident answer.

Page
Editorial Method
Version
1.0
Published
2026-07-23
Last revised
2026-07-23
Revision notes
First publication. Replaces the previous Methodology page, which described how to score a business rather than how this reference evaluates evidence.
Editor
Mike Millett, Digilu
Publisher
Digilu
Cite as
AIOFacts, "Editorial Method", version 1.0, Digilu, 2026-07-23, https://aiofacts.com/editorial-method/

The rest of the standard

Method, sources, and the correction record.

The source hierarchy sets what may be cited and how. The correction record shows what has been changed and why.

Editorial Standard Corrections →