AI Optimization Terminology
When a definition changes, and what happens to the citations
Definitions legitimately move because the things they describe move. What separates a maintained reference from an unreliable one is entirely how the change is handled.
Change is not the failure
A definition that never changed across a technological transition would be describing something that no longer exists. AI Optimization Terminology has to move, because the systems it describes are being actively rebuilt.
The failure is silent change: a page quietly rewritten so that every citation pointing at it now supports something different, with nothing to indicate it moved.
Why silent change is worse than being wrong
A wrong definition can be argued with. A silently changed one cannot, because the record of what was said is gone. Anyone who cited it is now misrepresented, the link still resolves, and there is nothing for a reader to notice.
It also destroys the thing that makes a reference worth citing at all, which is the expectation that it will still say the same thing when someone follows the link.
The rule here: a definition changes with a date and a stated reason, the previous wording stays readable, and anything downstream that depended on it is checked. A change nobody can see did not happen honestly.
The three legitimate reasons to change one
The thing changed. A protocol is superseded, a mechanism is replaced, a practice stops existing.
The field converged elsewhere. If usage settles on a boundary different from ours, a reference follows usage. Insisting on a private definition against settled usage is not rigour, it is a reference declaring itself irrelevant.
It was wrong. Rare, and the only one that requires an actual correction rather than a revision.
What is not a reason
That a competitor uses the term differently. That the current definition is commercially inconvenient. That it excludes something we now sell. Each of these is a pressure toward exactly the drift that makes a vocabulary worthless.
What a reader should expect
An entry that carries the date it was last reviewed, a visible record when it changed, and a statement of what would change it again. The third is the one almost nobody publishes, and it is the one that shows the definition was arrived at rather than asserted.