AI Optimization
Owned-domain floors, and why we refuse them
The scoring decision that most affects whether an evaluation is worth anything, and the reason our own properties score badly on the largest measure.
What an owned-domain floor is
A rule, usually unstated, that guarantees a minimum score for something the evaluator has an interest in: its own properties, its clients, or businesses that bought the paid tier. It is rarely built deliberately as fraud. It arrives as a reasonable-sounding adjustment that happens to run one direction.
How it gets built without anyone deciding to
A scoring model produces an unflattering result for a client. Someone reasonably observes that the model is missing context. A correction is added. The correction is genuine and it is only ever applied when the result was too low, because nobody raises a complaint about a score being too generous.
Repeat that a dozen times over a year and the model has a floor nobody chose and nobody can point to.
The single test: can this evaluation produce a bad result about the company that built it? If not, it is not a measurement. It is marketing with a numeric interface.
What we do about it
Every ecosystem domain we own grades 0.5 on independent corroboration, because references between sites controlled by the same party are navigation rather than evidence. Sixteen properties agreeing with each other adds close to nothing, and counting it would be the largest available thumb on the scale.
That decision costs us the most flattering number on our own report, and it is the only reason any other number here is worth reading.
The subtler version
A floor does not have to be a minimum score. It can be a choice of question, a choice of comparison set, or a decision about which signals are checked. Each of these can be defended individually and each can be selected to produce a favourable outcome.
This is why method has to be published rather than summarised. A published weighting can be argued with. A stated commitment to fairness cannot.
What to ask of any AI optimization score
Where is the vendor in its own rankings, and does it publish that? Are the weights disclosed? Has it ever published an unfavourable result about itself or a client? Does the score change when the business changes, or only when the relationship does?
The right response to all of this is to check rather than to believe, which is also the correct response to this page.