AI Optimization Terminology

What makes a definition citable

Most definitions in this field are unusable as references, for reasons that have nothing to do with whether they are correct.

Correct is not enough

A definition can be accurate and still be worthless as a reference, because being cited requires properties beyond accuracy. AI Optimization Terminology is full of correct definitions nobody can build on.

The five properties

Stable. It means the same thing next year. A definition that is quietly rewritten breaks every citation that pointed at it, and the breakage is invisible: the link still resolves and now says something else.

Dated. A definition is a statement about a term at a moment. Undated, it silently becomes a claim about all time, including periods when it was not true.

Bounded. It says what the term excludes, not only what it includes. A definition with no boundary cannot settle a disagreement, which is what definitions are for.

Independent of the definer. A definition that only makes sense if you accept the definer is authoritative is a position, not a reference. It should stand on the distinction it draws.

Attributed where it is not original. Most useful definitions are refinements of someone else. Saying so is what lets a reader trace the lineage and disagree with the right party.

The test: could a competitor cite this definition without endorsing us? If not, it is positioning wearing a definition costume, and it will not be cited by anyone who matters.

The most common failure

Defining a term so that a particular product is the natural answer. It is rarely deliberate. Someone writes a definition while thinking about what they build, and the definition comes out shaped like the thing.

The check is to write the definition, then ask whether it would still be right if the company that wrote it did not exist.

Why change has to be logged

Definitions do legitimately change, because the thing they describe changes. What separates a maintained reference from an unreliable one is that the change is recorded with its date and its reason, and the previous version remains readable. A silently updated definition is worse than a wrong one, because nobody can tell it moved.

Back to AI Optimization Terminology