You Have Seven Reputations
AI systems, knowledge records, search, reviews, media, profiles, and legal filings. Seven layers, each with distinct sources, distinct decay rates, and distinct failure modes — and a single blended score that hides all of it.
The word reputation is doing too much work. It refers to at least seven different systems that happen to describe the same entity, and that share almost nothing else — not their sources, not their update cadence, not their tolerance for error, not the remedies available when they go wrong.
Treating them as one thing is the reason so much reputation spending produces so little observable change.
The layers
The AI layer is what language models say when asked about you. Fast to change, hard to inspect, and increasingly the first impression anyone forms.
The knowledge layer is the structured record — the entity graph entries, the machine-readable properties. Small, slow, quiet, and load-bearing for everything above it.
The search layer is the first page of results, which still governs what most people believe they discovered themselves.
The review layer is the platform-specific aggregate that dominates your category, whichever one that is.
The media layer is coverage with institutional weight — durable, highly citable, and effectively permanent.
The profile layer is the scattered set of directory and social entries that nobody thinks about until one of them contradicts the others.
The legal layer is filings and public records: rarely read directly, disproportionately trusted when they are, and impossible to argue with.
Why a blended score is useless
Average the seven and you get a number that cannot be acted on. A composite of 70 might mean six healthy layers and one catastrophic failure, or seven mediocre ones. Those situations demand completely different responses, and the score cannot distinguish them.
Worse, blending hides the only structure that matters, which is that the layers are not independent.
Nothing in the stack fails in isolation. The knowledge layer breaks quietly, and four layers above it start repeating the break as fact.
The cascade
Dependencies run upward from the structured layers. The knowledge layer feeds the AI layer and the search layer, because both treat structured records as high-confidence ground truth. The media layer feeds the AI layer, because institutional coverage is exactly what a model is trained to weight heavily. The review layer feeds the search layer through rich results.
This means a defect low in the stack surfaces as several apparently unrelated problems higher up. A missing property in a structured record shows up as a wrong AI answer, a malformed knowledge panel, and a competitor outranking you for your own name — three symptoms, one cause.
It also means most reputation work is misdirected by construction. The symptoms are visible and the cause is not, so effort flows to whichever layer is loudest rather than whichever layer is upstream.
Where to intervene
Diagnose every layer separately, with its own sources and its own measures. Never report a blended number to anyone making a decision.
Then work upstream first, always. The knowledge and legal layers are the cheapest to fix and the most expensive to leave broken, precisely because they are boring enough that nobody checks them. The AI and search layers are the most alarming to look at and the least productive to attack directly, because they are largely downstream of the layers underneath.
The instinct is to fight where the damage is visible. The stack is telling you to fight where the damage originates. Those are almost never the same place, and the gap between them is where most reputation budgets quietly disappear.