AODM (AI-Optimized Data Markup) is a small set of free, open tags you add to data you already have. Schema.org and JSON-LD describe what a thing is. AODM describes what it is and how much to trust the claim — where it came from, how certain it is, when it was true, and whether anyone observed it or a machine inferred it.
The problem
Knowledge moves between systems as prose, ad-hoc JSON, or vectors in a database, and the context is lost in transit. A number arrives without its unit. A statement arrives without its source. A fact that expired two years ago arrives looking exactly like one confirmed this morning. Every consuming system reinvents its own conventions for patching this up, and none of them interoperate.
The concrete difference
The same claim in JSON-LD:
{
"@context": "https://schema.org",
"@type": "PropertyValue",
"name": "Temperature rise under load",
"value": "1.2"
}A consumer learns there is a value of 1.2. Not the unit, not who measured it, not how confident anyone is, not when, not whether a later revision superseded it. Every one of those has to be carried out-of-band.
The same claim in AODM:
<aodm:fact id="temp-rise" about="engine" valid-from="2026-01-15">
Temperature rise under sustained load.
<aodm:value number="1.2" unit="Cel" tolerance="0.1"/>
<aodm:source uri="https://example.com/report-24" retrieved="2026-01-15"/>
<aodm:confidence value="0.9"/>
</aodm:fact>A system reading this can decide whether to rely on it. A system reading the JSON-LD above can only decide whether to believe it.
What the alternatives do not model
| Capability | Schema.org / JSON-LD | AODM |
|---|---|---|
| Type and structure | Yes, far broader | Yes, deliberately small |
| Per-fact provenance | No | Yes, source |
| Per-fact confidence | No | Yes, confidence |
| Validity window | No | Yes, valid-from / valid-to |
| Units and tolerance | Free text | Yes, structured value |
| Known-false vs. unknown | No | Yes, polarity |
| Inferred vs. observed | No | Yes, derived-from |
| Machine-proposed vs. established | No | Yes, origin |
| Supporting passage, quoted | No | Yes, evidence |
| Search engine rich results | Yes | No |
And the rest of the landscape
| Format | Relationship | Where AODM differs |
|---|---|---|
| RDF / OWL | Complementary | More expressive, with a mature reasoning ecosystem. Doing provenance and uncertainty properly means named graphs, reification or RDF-star, and an ontology stack. AODM treats both as core, and maps onto RDF when you need that power. |
| JSON-LD | Complementary | A serialisation of linked data with an @context indirection. AODM's JSON form is a fixed, closed schema, so a consumer needs no context resolution to know what a field means. |
| Vector databases | Different job | Answers “what text resembles this query”, not “is this still true” or “who said it”. Embeddings also bind knowledge to one model; re-embed with another and the representation is gone. |
| Microformats | Alternative | Overloads HTML class names. AODM uses namespaced elements with a real schema, validation, nesting and typed values. |
Raw JSON or XML in <script> tags | Alternative | No shared schema, so every publisher invents a shape and every consumer writes a bespoke parser. |
AODM does not produce rich results in Google and is not trying to. It complements Schema.org, and using both on the same page is a reasonable thing to do.
When the machine starts proposing things
Once a language model is reading your documents and suggesting claims, the question stops being “what does the system know?” and becomes “which of this did a person establish, and which did a machine guess?” AODM separates observed knowledge someone recorded, derived knowledge a rule guarantees, and generated knowledge a system merely proposed, with the passage that prompted it quoted and located. A reviewer accepts or rejects in seconds.
Anyone can implement it
The specification, both schemas, the validation rules and the conformance tests are published together, and nothing about AODM requires this project’s code. The four reference parsers also test the specification: if two disagreed about the same document, the ambiguity would be in the specification rather than the code. The conformance suite runs all four over the same inputs and requires identical results.
Nobody re-authors their data
The XBRL converter turns financial filings — which every listed company already produces for regulators — into AODM automatically, carrying across the reporting periods, units, dimensions and stated precision that most pipelines discard.
Frequently asked questions
What is AODM?
AODM (AI-Optimized Data Markup) is a small set of free, open tags that make web content and structured data directly readable by AI systems. Beyond describing what a fact says, it records where the fact came from, how confident the publisher is, when it is valid, and whether it was measured or inferred.
Does AODM replace Schema.org or JSON-LD?
No. Schema.org describes what a thing is and earns rich results in search; it is mature, universally supported, and AODM does not replace it. AODM describes how much a claim should be trusted. Use AODM when something downstream will act on your data and losing the provenance would be expensive. Using both on the same page is a reasonable thing to do.
How does AODM help my site get found by AI search?
Generative engines answer questions by extracting claims from pages and deciding which ones they can safely repeat. AODM attaches a source, a confidence value and a validity window to each claim, so an AI system can verify a statement instead of guessing at it — which makes your content materially safer to cite.
Why not use RDF and PROV-O?
RDF can express all of this, and the objection is cost rather than capability. Doing it properly means named graphs, reification or RDF-star, an ontology stack and a team who knows all of it — so most pipelines carry no provenance at all. AODM makes the common case cheap and maps cleanly onto RDF if you outgrow it.
Why not just put everything in a vector database?
A vector store answers "what text resembles this query." It cannot tell you whether a claim is still true, who said it, or what else breaks if the source turns out to be wrong. Embeddings also bind knowledge to one model; re-embed with another and the representation is gone.
Is AODM free to use?
Yes. AODM is released under the Apache License 2.0. The specification, both schemas, the reference parsers and the conformance test suite are free for commercial and non-commercial use.
How long does it take to add AODM to a site?
Minutes for the embedded profile. You add one data-aodm attribute to HTML you have already written — no scripts to load, no build step, and no measurable effect on page speed.