Perception Is a Retrieval Problem
Reputation is not a value stored somewhere and looked up on demand. It is assembled fresh, from whatever happens to be available, every single time someone asks. Once you see it that way, most of what the industry sells stops making sense.
The intuitive model of reputation is a ledger. Somewhere there is a number, and events adjust it: good press credits the account, a bad quarter debits it, and your standing is the running balance. This model is wrong in a way that quietly invalidates most reputation strategy.
There is no stored value
Nobody consults a ledger. When a person or a system needs an impression of you, it goes and gets one. It runs a query against whatever sources it has access to, weighs what comes back, and composes an answer on the spot. Then it throws the answer away.
The next query composes a different answer. Same entity, different sources retrieved, different emphasis, different conclusion. Ask two systems about the same company on the same afternoon and you will get two reputations, because you ran two retrievals.
The pool is the product
If every impression is assembled at query time, then the only durable object in the system is the pool of retrievable material. Not the impressions — those are ephemeral. The pool.
This is why campaigns feel so unsatisfying. A campaign produces a burst of favorable material, the pool briefly tilts, and then the burst ages out of relevance while the structural problems in the pool — the wrong founding year in the structured record, the unresolved thread with four hundred comments, the outdated revenue figure that three aggregators copied from each other — sit exactly where they were.
Those structural items are not louder than the campaign. They are simply more retrievable, more often, for longer. Retrievability beats volume, and it beats it decisively.
A campaign changes what exists. Only structural work changes what gets retrieved.
Weighting is not neutral
Retrieval systems do not sample the pool evenly. They privilege sources that are structured, stable, frequently cited, and institutionally legible. An encyclopedia entry outranks a blog post not because it is more accurate but because it is more citable. A structured record outranks a press release because it parses cleanly.
This produces a specific and counterintuitive failure mode: a small, boring, factually wrong record can outweigh an enormous quantity of accurate material, indefinitely, simply because it sits in a format that retrieval systems find easy to trust.
Most reputation damage is not dramatic. It is a bad field in a good record.
What follows
Three things follow, and they are unpopular.
The first is that sentiment scores are close to meaningless. They measure the tone of material in the pool, weighted by nothing in particular. What matters is the tone of material that gets retrieved, which is a different and much smaller set.
The second is that volume metrics are worse than meaningless — they are actively misleading, because they reward flooding the pool with material that will never be retrieved, and flooding is cheap, which is precisely why it is sold.
The third is that the correct unit of work is the source, not the story. You do not fix a narrative. You fix the specific retrievable artifacts that the narrative is being assembled from, in order of how often they are actually retrieved. Everything else is theater with an invoice attached.