You worked for months to build visibility in AI answers — and now you see competitors who started from zero getting dangerously close again. AI authority is not a milestone that, once reached, stays forever: if you stop feeding it, it decays, and those who keep producing fresh signals overtake you even if they started with a disadvantage against you. It's not a matter of doing everything from scratch: a minimal maintenance plan is enough to protect what you've already built and keep competitors at a distance.
You did everything right. Your brand appears in AI answers, the mentions are there, the credibility signals are in place. Then six months go by without updates, without new publications, without fresh mentions. And one day you ask Perplexity about your industry and in your place there’s a competitor who didn’t exist six months ago.
It’s not an injustice. It’s a mechanism with a precise name: trust decay. And understanding it is the difference between those who keep their visibility in AI answers and those who lose it without even realizing.
Credibility is a score, not a title
There’s a widespread misconception: thinking that once you’ve won authority in the eyes of the AI, it stays there forever. Like a diploma hanging on the wall. That’s not how it works.
In the world of information credibility research, trust is not binary. The research by Srba et al. (2024) formalizes it clearly:
“Credibility assessment follows two steps: detecting individual signals, then aggregating them into a single ordinal credibility label or a numerical credibility score.” Srba et al., 2024
Note the last words: numerical credibility score. Not a yes/no badge. A numerical score that the system computes by aggregating individual signals. And a score, by definition, can go up or down.
Every signal that contributes to that score has a temporal component. A mention in an industry outlet from 2024 carried weight when it was published. In 2026, if it wasn’t followed by other mentions, that signal doesn’t disappear, but its relative weight decreases, because in the meantime the system has ingested millions of fresher signals. Including those of your competitors.
The mechanism: why authority decays
To understand decay, you have to start from how models are updated. I covered this in depth in the article on the knowledge cutoff: every model has a cutoff date beyond which it knows nothing. But the point that matters here is another one.
Jeffrey Cheng et al. (2024) documented that the effective cutoff does not coincide with the declared one:
“Although sometimes it does align with the reported cutoff, in many cases it does not.”
Jeffrey Cheng et al., 2024
In simple terms: the model does not have a uniform memory of everything that happened up to date X. Some resources it recorded in the most recent version, others in a version from months or years earlier. Which means that the “freshness” of your presence in the model depends on when, and how often, your content and the mentions concerning you were included in the training cycles.
From this follows an operational deduction: if you stop generating new signals, your representation in the model crystallizes on a dated version. And with every new training cycle, the competitors who in the meantime have published, earned mentions, updated their content are recorded with fresher data. Your score doesn’t drop in absolute value, but it drops in relative value, because the others rise.
It’s like a league where, if you don’t play the matches, the others’ points overtake you.
A brand with the exact same signals as a year ago, with nothing new, transmits a dormant-source signal.
The double layer: training and retrieval
Decay acts on two fronts at the same time, and this is what makes it particularly insidious.
In the training data, your entity is a map of associations built from all the signals the model encountered during training. Those associations reflect the state of things at the moment of the crawl. If since that moment you’ve generated nothing new, at the next training cycle the model updates your competitors’ associations but yours stay frozen. The map gets rewritten, and you occupy less space.
In RAG retrieval, the effect is even more direct. Systems like Perplexity and Bing Chat retrieve sources in real time and also weigh them for recency. A brand with fresh mentions, updated content, and current structured data transmits an active-source signal. A brand with the exact same signals as a year ago, with nothing new, transmits a dormant-source signal. The system doesn’t penalize it explicitly, it simply prefers whoever is fresher.
Every quarter, verify that the data is current.
Trust boundaries break easily
There’s a further aspect that makes trust decay even more relevant in an ecosystem where AI models interact with one another and with external sources. Xinyan Ma et al. (2026) describe it in a technical context, but the principle applies to reputation as well:
“These interactions breach traditional trust boundaries, where localized malicious inputs or model hallucinations can propagate through the system.”
Xinyan Ma et al., 2026
The traditional boundaries of trust are no longer stable. In a system where information circulates between models, crawlers, and external sources, incorrect or dated information about you can propagate. If you don’t actively maintain your trust signals, you leave room for obsolete information, or worse, for information generated by others that concerns you but that you don’t control.
Maintaining authority is not just a matter of ranking. It’s a matter of control over the narrative the AI builds about your brand.
The trust maintenance plan: what to do in practice
The good news is that recovery from trust decay doesn’t require rebuilding everything from scratch. If you already have an authority base, maintaining it is far less costly than rebuilding it. The problem is that almost no one thinks about it until it’s too late.
Here’s an operational framework that starts from the signals that matter most.
External signals: at least two mentions a month. You don’t need an article in the Financial Times every week. You need consistent mentions on industry sources: updated professional directories, interviews, posts on vertical outlets, third-party citations. Two a month is the minimum to maintain the flow of fresh signals that the model records in the update cycles. I covered this in depth in the article on E-E-A-T: external signals carry more weight than self-declared ones.
Structured data: quarterly review. Schema markup, Google Business profile, Wikidata if you have one. Every quarter, verify that the data is current. Obsolete structured data is not neutral, it’s a negative signal, because it tells the system that the source is not maintained. Brand entity consistency is the glue that holds all the other signals together.
Content: continuous updating, not rewriting. You don’t have to publish an article a day. You have to update the ones you have, fresh data, recent sources, new sections where the topic has evolved. An article published in 2024 and updated in 2026 communicates “this source is active on the topic”. The same article, untouched since 2024, communicates the opposite. Temporal authority is built over time, but it is maintained only with regular updates.
A quick check to understand where you stand
Before building a maintenance plan, you need to understand whether decay is already underway.
Ask your reference AI engine something about your industry, a query that six months ago returned your brand in the answers. Are you still there? Is the information up to date or does it date back months? If the answers talk about you in the past tense, or no longer mention you, decay is already active.
Then check the date of the last relevant external mention of your brand. If more than three months have passed without any mention on third-party sources, the flow of fresh signals has been interrupted.
These are surface checks; to understand the real extent of the decay you need tools that cross-reference your presence in the training data with active citations in RAG systems. But they tell you whether the problem exists and how urgent it is.
Maintenance costs less than rebuilding
The difference between those who maintain AI visibility and those who lose it is not the budget. It’s consistency. Two mentions a month, a quarterly update of structured data, content reviewed regularly, is a modest investment compared to what it takes to rebuild lost authority.
Because recovery is possible, but it’s slow. A brand that has lost positions in the model has to rebuild signals that competitors have accumulated for months. Every month of inactivity is a month that has to be recovered with twice the effort.
AI authority is not a milestone. It’s a continuous process. Whoever understands this first has a structural advantage over everyone who thinks they can stop.
