AI Optimization
Signals decay, so a result is a statement about a date
Nothing measured here holds still. What decays fastest, what holds longest, and why every finding carries the day it was taken.
Why dating is not a formality
An AI optimization result describes a relationship between a business and a set of systems, and both sides move. The business changes its site. The systems change what they weigh. A finding without a date is a claim about an unspecified moment, and it will be quoted long after it stopped being true.
What decays fastest
Selection results, continuously. Whether an engine names a business varies between samples and between weeks. Only a trend across many samples means anything, and a single result is out of date almost immediately.
Retrieval, abruptly. It does not decay gradually, it breaks. A configuration change, a plugin update, a security product. Fine on Monday, invisible on Tuesday, with nothing on the site looking different.
Third-party listings, quietly. Directories change formats, close, or fall out of use. A citation that existed last year may be gone with no notification to anyone.
What holds longest
Independent editorial coverage. Once written it persists, and it keeps working for years. This is a large part of why corroboration is weighted so heavily: it is the slowest to earn and the slowest to lose.
Authoritative reference records. Durable by design.
Structural clarity on your own site. Stable until someone redesigns, which is also the moment it is most often lost without anyone checking.
The migration rule: re-evaluate immediately after a site migration, a rebrand, a domain change, or a change of business name or address. These are the moments things actually break, and they are the moments everyone is looking at whether the site renders rather than whether it can still be read.
Why the method carries a version
The measurement itself has to change as the systems do. That creates a second decay problem: a reading taken under an old method is not comparable to one taken under a new one.
So every result records the version that produced it, and a comparison across a version change is withheld rather than shown. A number that improved because the ruler changed is worse than no number, and it is most tempting to show precisely when it flatters everyone involved.
What this means for the work
AI optimization cannot be finished. That is not a sales position, it is a property of measuring a relationship where one side is a set of products being actively developed by other companies. What can be finished is a specific fix. What cannot be finished is the condition.