Meaning deserves
more than a summary.
A machine-readable layer that lets authors express what they mean — and lets AI systems test whether that meaning is actually supported by the text.
What is meant
What is expressed
What is understood
Between what we write and what machines repeat, meaning can drift.
AI increasingly mediates how people encounter articles, research and ideas. A summary may sound fluent while quietly weakening a claim, removing a qualification or attributing a conclusion the author never made.
Intended meaning
The question, the central claim, the distinctions and the limits an author meant to communicate.
Expressed meaning
What the actual published text supports, qualifies, contradicts or leaves unanswered.
Inferred meaning
What a reader or AI extracts — sometimes faithfully, sometimes incompletely or incorrectly.
A semantic layer, not a script for AI.
AIL makes authorial claims inspectable. It does not instruct a model to agree, override the reader, suppress critique or certify the author's position as true.
Each declaration is connected to a document version and, where appropriate, exact passages. This makes disagreements visible and reviewable.
How AIL works →{
"intent": {
"central_question":
"What did the author mean?",
"central_thesis": {
"id": "T1",
"text":
"Meaning should be testable."
}
},
"claims": [ /* anchored statements */ ],
"verification": {
"status": "not_run"
}
}Clearer reading requires more than one test.
AIL distinguishes interpretation, support in the text and evidence beyond the text.
Interpretation fidelity
Did an AI preserve the author's central idea, qualifications and scope when summarising?
Intent–content alignment
Does the published article actually substantiate the intent it declares?
Claim–evidence grounding
Are factual claims supported by independent, relevant and credible evidence?
“The purpose is not to tell AI what to think, but to help it distinguish what the author intended, what the text actually says, and what the evidence allows us to conclude.”
One format. More transparent interpretations.
Read the specification, inspect the schema and test the first published example. AIL 1.0 is a public proposal open to technical scrutiny and independent evaluation.