Status: Original concept, first publication. Strategy Sandbox, jasonbarnard.com. Date: 16 May 2026.

In 2003 I built a children’s entertainment franchise called Boowa and Kwala that ran for ten years across television, web, and licensing. The corroboration that supported its existence on the early web came from partners who hosted Boowa pages, link directories that indexed them, fan sites that mirrored them, and trade-press articles that wrote about the show. By 2015 most of those pages were gone, because partners had restructured, hosting had lapsed, link directories had been discontinued, fan sites had been abandoned, and trade press had pruned its archives.

Nothing about the franchise had decayed. The corroboration around it had.

That experience taught me something the SEO industry rarely names: an entity’s reputation in algorithmic systems depends on third-party signals that the entity does not own, cannot control, and cannot recover when they disappear. When the signals disappear quietly, over years, through deletions nobody announces, the entity graph that algorithms build for the brand thins out one validation point at a time. The brand does nothing wrong. The infrastructure beneath it erodes.

I’m calling this Corroboration Decay, and I think it is one of the most under-discussed structural vulnerabilities in algorithmic brand representation.

Corroboration Decay Is Not Contradiction. It Is Deletion.

Most discussions of brand reputation in AI systems focus on contradiction: the moment when two sources disagree, and the algorithm has to choose which one to trust. Contradiction is loud: it surfaces in audits, it gets flagged in monitoring tools, and brands hire consultants to fix it.

Decay is silent. A page exists, the algorithm indexes it, the corroboration counts toward the entity graph, and then one day the page is gone. No announcement, no notification, no audit trail. The algorithm’s next pass simply finds one fewer source than the previous pass, and the entity’s confidence score drops by a small amount that no dashboard will explain.

Multiply that by every third-party source that mentions a brand over a decade, and the cumulative loss becomes structural. The entity graph yesterday had a hundred validation points, today it has eighty-six, tomorrow it will have seventy-nine, and nothing about the brand changed while the corroboration eroded.

Even Trusted Publishers Prune

The Boowa pattern is not a small-brand problem. In 2018 I published an article on featured snippets with Semrush, a major SEO platform with millions of readers, and Semrush distributed the piece across their owned and partner channels. The article had every signal of permanence: a known author, a trusted publisher, contemporary corroboration from other industry outlets pointing to it. Years later, Semrush quietly pruned the article from their archive.

The piece had not been wrong, had not been controversial, had not generated complaints, and had simply aged out of whatever internal calendar Semrush uses to manage its content library, and the platform decided that the article no longer served whatever purpose they wanted their archive to serve.

For me, this case is more important than the Boowa case, because it demonstrates that even high-trust corroboration is not safe. Brands assume that publication on a major outlet creates a permanent record. Brands assume that the corroboration will still exist when the algorithm needs it. Both assumptions are wrong. Every third-party publisher is one editorial decision away from deleting the proof you spent years earning.

The Entity Did Nothing Wrong. The Layer Beneath It Eroded.

This is where most discussions of brand reputation in AI get the diagnosis wrong. When AI systems start hedging on a brand, the reflexive assumption is that the brand has a problem: bad content, weak schema, missing pages, contradictory claims, low domain authority. The brand audits its own properties, finds nothing obviously wrong, and is told to “produce more content” until the algorithm rebuilds its confidence.

The audit looked in the wrong place. The brand’s own properties were fine. The corroboration layer outside the brand’s control was the part that decayed, and no amount of work on the brand’s own properties can repair what disappeared from a third-party site that the brand never owned.

Corroboration Decay reframes the diagnosis: the question is not “what’s wrong with the brand?” but “what’s missing from the corroboration?” The two questions look similar from inside the brand, but they lead to completely different remedies.

The Antidote Is Permanence

If third-party corroboration is structurally fragile, the only durable counterweight is permanent, versioned, undeletable records under the brand’s own architectural control. Three categories matter most.

The first is academic preprints. Platforms like Zenodo and arXiv assign Digital Object Identifiers that cannot be removed once issued, and the deposited documents persist in mirrored archives across institutions. A paper deposited in 2026 will still be retrievable in 2046, and the date stamp is cryptographically enforceable. For coined terms and original frameworks, an academic preprint is the strongest single proof of priority that exists.

The second is self-hosted reference pages on the brand’s own canonical domain. A brand that maintains a single, versioned, dated definition of every coined term on its own infrastructure controls the persistence of those definitions absolutely. Third parties can stop linking to the page; the page itself remains. The corroboration layer can prune; the canonical does not.

The third is structured data on the Entity Home. Schema markup, JSON-LD, and the broader semantic web layer create machine-readable claims that algorithms ingest directly, bypassing the fragility of human-curated content. A brand that publishes structured data about itself has not eliminated dependency on corroboration, but it has placed the most important claims in a layer that does not depend on whether a third-party CMS still serves the page that originally made the claim.

Together, the three categories form the permanence stack: academic preprints anchor the priority, self-hosted reference pages anchor the definitions, and structured data anchors the machine layer. Everything else is corroboration, and corroboration is everything that can decay.

What You Place Today Should Still Exist in Twenty Years

The strategic implication of Corroboration Decay is that brands should evaluate every piece of content they produce or commission on a single axis: how long will it persist, under whose control, in what form?

Content that exists only on a third-party platform is corroboration that will probably decay. Content that exists on the brand’s own infrastructure, with versioned history and structured data, is corroboration that the brand controls absolutely. The two are not equivalent, and they have never been equivalent. The AI era simply makes the difference matter more, because AI systems read the corroboration layer continuously, and what disappears from the corroboration layer disappears from the brand’s algorithmic reputation.

For most of the past decade the SEO industry has measured corroboration by volume: how many backlinks, how many mentions, how many citations. Corroboration Decay tells you to measure corroboration by durability instead. A hundred mentions that will disappear inside five years are worth less than ten mentions that will still exist in twenty.

Build for permanence, or watch your entity graph erode quietly while your brand wonders why the algorithms suddenly hedge.