| The thesis: human reputation fades with time. A machine's does not, because a machine has no native sense of your timeline. Every version of your company you have ever been, the pivot you abandoned, the pricing you dropped, the you from three years ago, exists in its mind as simultaneously and equally true. There is no then and now. There is only an eternal present in which all your contradictory selves coexist. |
Human memory has a mercy built into it: it fades. The market's recollection of your worst quarter softens, your abandoned strategy slips out of mind, the version of you from years ago quietly recedes to make room for the current one. Time sorts your history into a sequence, with the past marked clearly as past. We take that sorting so for granted that we never think to ask whether the machines do it too.
They do not. And that single missing faculty, a sense of before and after, reshapes how your entire history behaves.
A machine has no timeline
When a model learns about you, it absorbs statements from across your whole existence without a reliable sense of when each was true. A description from your earliest days and a description from last month can sit side by side in its understanding with no clear marker that one has been superseded by the other. It knows both things were said. It often does not know, or does not weigh, that one of them stopped being true.
So your history does not arrive at the machine as a story with a beginning, middle, and current chapter. It arrives as a pile of simultaneous claims, all wearing the present tense. The company you were and the company you are do not line up in sequence. They overlap, and the machine reads the overlap as a single, slightly contradictory now.
Every version of you, at once
Picture what that means concretely. The positioning you retired is still current. The product you sunset still ships. The market you left is still your market. The leadership that changed is still in place. Not as history, but as live, present-tense fact, coexisting with the true current versions and contradicting them. You are, to the machine, all of your past selves and your present self at the same time, an overlaid exposure of every era you have passed through.
This is why a company that has genuinely transformed can watch the machines describe a stranger. Nothing was deleted. Every former self is still in the record, still speaking in the present tense, and the machine has no forgetting curve to retire them. Your past does not fade. It accumulates, and it all reports as now.
This is not legacy, and it is not latency
It is worth being precise, because this sits next to two ideas we have written about and is distinct from both. Legacy interpretation is about old signals persisting, the past refusing to leave. Strategic latency, which we framed as being read before you are met, is about the delay before your current reality propagates.
Both of those still assume a timeline exists and something is lagging along it. The eternal present is stranger: it is the loss of the timeline itself. The problem is not that the old version is slow to leave or that the new one is slow to arrive. It is that the machine cannot reliably tell which is which, because it does not experience your history as a sequence at all.
Why the machine cannot sort it without help
You might expect dates to solve this, and they help less than you would hope. A model does not reliably reason about your chronology the way a person reading a timeline would. Recency is not something it feels; it is something it has to be told, clearly and consistently, and even then older, heavily repeated material can outweigh a correct but quieter update.
Volume can beat recency. The version of you that was true longest, or was written about most, can dominate the present tense simply because there is more of it, regardless of whether it is current.
So left to itself, the machine does not default to your latest chapter. It defaults to your loudest one, whenever it happened. The present it reports is not the present. It is a weighted average of all your presents, tilted toward whichever era left the biggest mark.
Establishing the now is its own discipline
Here is the practical turn, and it is easy to miss. Most brand correction work is about fixing what is wrong. This is a different job: establishing what is current. Even when every individual statement about you is accurate, you can still be misrepresented if the machine cannot tell which accurate statement describes the present.
Correcting falsehood and asserting recency are two separate disciplines, and the second is the one almost no one practices. You have to actively, repeatedly signal what is true now, and deliberately retire the past selves still reporting in the present tense, or the eternal present will keep collapsing your eras into one.
That ongoing assertion of the now is part of the Protect work in Model Response Optimization, and it is why each new model release is a chance to reassert which chapter is current before the overlay re-forms.
Questions this raises
Why does AI mix up old and current information about my company?
Because a machine has no native sense of your timeline. It absorbs statements from across your whole history without reliably knowing which have been superseded, so former and current descriptions coexist in the present tense. It knows both were said; it often does not weigh that one stopped being true.
Isn't this the same as outdated information persisting?
It is related but distinct. Legacy interpretation is old signals lingering, and latency is the delay before your current reality propagates; both assume a timeline with something lagging along it. The eternal present is the loss of the timeline itself, so the machine cannot reliably tell current from former in the first place.
Won't dates and timestamps fix this?
They help less than you would expect. Models do not reliably reason about chronology the way a person reading a timeline does, and recency has to be told, clearly and repeatedly, rather than felt. Even then, older, heavily repeated material can outweigh a correct but quieter update, so volume can beat recency.
Why does the machine default to an old version of us?
Because left to itself it defaults not to your latest chapter but to your loudest one, whenever it happened. The present it reports is a weighted average of all your past presents, tilted toward whichever era was written about most or was true longest, regardless of whether it is current.
What do we do about it?
Treat establishing the now as its own discipline, separate from correcting what is wrong. Actively and repeatedly signal what is true today, and deliberately retire the past selves still reporting in the present tense. Reassert current-state with each major model update, since the overlay of eras re-forms over time.
The market used to forget, and forgetting was a gift: it let you outgrow your past.
The machines do not forget, and worse, they do not sequence, so every version of your company you have ever been keeps speaking in the present tense at once. You cannot rely on time to retire your old selves, because to the machine there is no time, only an eternal present crowded with all your former chapters. The work is to keep declaring, clearly and constantly, which chapter is the current one, and to usher the old ones off the stage yourself. Nothing else will, because nothing else remembers that they are old.
Find out how many versions of you the machines still hold as current. Request your Brand Interpretation Assessment, or read the Interpreter Architecture whitepaper for the full thesis.