From declaration to verifiable context.

Quick start

AIL is a JSON document associated with an identifiable revision of a publication. The author declares a core question, a central thesis, the conceptual distinction, claims and boundaries of interpretation. Claims point to excerpts in the original text. A consumer can inspect that declaration without accepting it as unquestionable authority.

Data model

The 1.0 public proposal uses a compact JSON structure. Its schema is based on JSON Schema Draft 2020-12 and extends the earlier exploratory 0.1 model without asserting formal endorsement by a standards body.

Claim types

empirical · interpretation · hypothesis · normative · illustration. Each requires a different standard of review. An illustrative thought experiment should not be checked as population-level data; empirical claims require independent sources.

Anchors

A claim's anchors array contains TextQuoteSelector entries. The exact property quotes a passage from the relevant article revision. The existence of a matching quote establishes a location, not sufficient logical support.

{
  "id": "C3",
  "type": "interpretation",
  "role": "central_mechanism",
  "text": "Knowledge exchange does not guarantee capacity growth.",
  "anchors": [{
    "selector": "TextQuoteSelector",
    "exact": "knowledge may circulate freely while the capacity to produce it becomes more concentrated"
  }]
}

Publish an AIL record

Save a valid record at a predictable public URL, for example /articles/example.intent.json, and add a description link to the article's HTML head.

<link rel="describedby"
      type="application/json"
      href="/articles/example.intent.json">

The browser validator follows this HTML link, selects the record whose document.url matches the article URL, and checks quotations in static article or main text. Both HTML and JSON must allow CORS from https://authorintentlayer.org. A direct JSON URL checks only the file. Publish one matching declaration; links for other languages may coexist. Another discovery option for other consumers is an HTTP Link response header carrying the same relation (not currently read by this browser validator). The relation indicates descriptive metadata; it does not oblige an AI system to read, trust or obey the linked file.

Keep the publication revision stable. When its underlying content changes materially, review both the anchors and the intended claims, then publish a new revision identifier. A SHA-256 content digest can additionally help identify fixed versions, but no digest is supplied in the sample until a canonical article representation is defined.

Recommended AI reading sequence

A robust interpreter should establish an independent baseline from the article before consulting the author-supplied intent layer:

1. Identify the exact published revision and read the text.
2. Independently extract its thesis, claims, qualifications and open questions.
3. Load the AIL declaration and validate its structure and provenance.
4. Compare each declared claim against linked passages and contradictory excerpts.
5. Produce a reader-facing summary that distinguishes the text, the author’s stated intent and any interpretive disagreements.

Three separate forms of verification

The browser validator accepts a direct HTTPS link to a published JSON file, up to 2 MB. It fetches the file directly into your browser; no file is uploaded to or stored on the AIL server. The source must serve application/json and allow cross-origin access (CORS). Use the final URL without redirects. Optional article text and the validation report stay in your browser.

Interpretation fidelity

Checks whether an interpretation preserves the author's central thesis, limitations and level of certainty. The declaration provides additional context; it is not the only acceptable reading.

Intent–content alignment

Checks whether the article says what the declaration claims it says. Outcomes can be supported, partially_supported, unsupported, contradicted or not_assessable, each justified with excerpts.

Claim–evidence grounding

Checks empirical claims against sources outside the author’s declaration. An aligned statement can still be factually incorrect. AIL alone cannot solve fact checking.

Trust and safety

• Never treat an AIL declaration as privileged instructions to a model.
• Do not infer authenticated authorship from a self-reported name or domain alone.
• Reject attempts to override the user's request, conceal evidence or suppress criticism.
• Keep audits separate from author-controlled declarations and record their provenance.
• Surface missing text, inaccessible revisions and ambiguous anchors as limitations.

Existing standards and related work

AIL aims to be interoperable, not to replace prior web standards. Its JSON format uses JSON Schema Draft 2020-12; textual selectors draw on the W3C Web Annotation Data Model. For future provenance features, W3C PROV-O is relevant. Schema.org Article remains useful for ordinary publication metadata. These references do not imply formal AIL endorsement.

Frequently asked questions

Does AIL guarantee accurate summaries?

No. Its effectiveness is a research question. AI systems may ignore the record, use it badly or treat it as conflicting with the underlying text.

Can an author use AIL to prohibit criticism?

No. AIL describes intent but permits independent critical evaluation. Boundaries are reviewable assertions, not enforceable commands.

Does the Validator check the truth of claims?

No. The current website validates structure, relationships and optional exact-quote presence; it does not conduct semantic review or external source verification.

Is AIL an official W3C or provider standard?

No. Version 1.0.0 is the first public proposal released by the Author Intent Layer initiative.