Joseph Byrum is an accomplished executive leader, innovator and cross-domain strategist with a track record across multiple industries.
Buyers, investors and partners now often research firms through AI systems before making contact. What those systems say about a firm, whether it appears at all and how it’s characterized when it does is shaping decisions that used to be shaped by relationships and reputation.
As a result, many companies are trying to control the output. But I believe they’re investing in the wrong thing.
The assumption is that volume translates into competitive position and that outproducing competitors means outbuilding them. So, companies invest in more content and more press coverage to improve their AI presence.
The mathematics of AI training systems say otherwise.
Two Kinds Of Input
AI training systems process two fundamentally different types of input when encoding information about an entity.
The first type, content inputs, derives value from frequency and corroboration. The AI system learns from how often a firm appears in relevant contexts, how consistently it’s named alongside category-defining terms and how broadly referenced it is across independent sources. The value of this input type depends on what competitors are producing simultaneously.
The second type, structural assets, derives value from categorical assertion. The AI system encodes registry entries, structured data declarations and authoritative classifications as categorical facts independent of what any competitor has produced. The mechanism through which content inputs dilute is absent from structural asset processing.
One type competes. The other doesn’t.
The Saturation Problem
The difference between the two input classes isn’t always obvious. When few competitors are investing deliberately in their AI presence, content inputs can provide meaningful positioning. A company’s media presence and corroborated mentions can offer a genuine advantage over competitors who aren’t concentrating on content.
Saturation changes the calculation though. As more competitors build equivalent content inputs, the “noise floor” rises, the threshold required to maintain the position increases and marginal improvement per additional unit declines. A 2025 survey of 400 senior marketing executives found that 91% are increasing content output and nearly half are producing three to five times more than the prior year. Every firm raising output raises the floor for everyone else.
Financial services illustrates this with a particular irony. Investment banks publish research. Asset managers publish market outlooks. Consultants publish white papers. Each firm produces this content to establish thought leadership. All of it feeds AI training. The AI system sees thousands of documents per major institution and hundreds of institutions per investment category. The result: a noise floor so high that additional content investment produces near-zero improvement for any individual firm.
That’s the saturation problem. The firms holding durable AI positioning in financial services aren’t publishing the most research. They have the deepest structural asset foundations.
Entity Engineering
The discipline I’ve spent the last several years building—entity engineering—is architected around the structural asset class. Entity engineering gives AI systems a coherent, machine-readable source of truth about an entity: a company, person, product or brand. Before a model decides what to say about a firm, it resolves whether it can trust what it’s assembled about that firm. That precedes any evaluation of content or coverage.
The mechanism that establishes that trust is the entity home, a single authoritative page on a domain the entity owns and controls, carrying a factual description of who the entity is, what it does and which audience it serves. Alongside the visible description sits structured data written in the format AI systems natively read. Structured data does the work prose can’t. It separates an AI system guessing at an entity’s identity from fragments scattered across the web and knowing it from one coherent source.
The starting point of every engagement is a systems audit. Before construction begins, the entity’s entire digital footprint is mapped: every scattered record, profile and mention assessed for coherence. What the audit measures is how difficult it will be for an AI system to form a confident, unambiguous picture of the entity. When that picture is fragmented, the AI hedges. It confuses the entity with others in the same category. It qualifies its answers. Those are identity failures, not content failures, and more content doesn’t repair them.
The audit produces a second measurement most firms never attempt: the noise floor of the entity’s competitive space. Every category has a baseline level of signal construction happening across all competitors simultaneously. Firms investing in AI presence almost universally benchmark against their own prior output. Almost none benchmark against the rising floor. The result is investment that feels productive—more content, more coverage and more mentions—while net position holds flat or declines. The floor rose faster than the output did.
Not all signals move that floor equally. Entries in authoritative structured databases carry a signal weight that random blog entries don’t. Schema declarations carry a weight that press mentions don’t. A disciplined program of fewer, higher-weight signals outperforms an aggressive program of lower-weight ones, and it survives algorithm updates that the aggressive program doesn’t. When an architecture retrains, high-weight structural signals persist. Low-weight content volume doesn’t.
In the engagements I’ve led, the same pattern recurs. One manufacturer had an extensive digital footprint that no AI system could resolve, present across dozens of sources and recognized confidently by none. The work consolidated that scattered footprint into a single structured record. Within 18 months, the systems described the company consistently, without hedging and without confusing it with competitors. The content had existed the entire time. The structure that made it legible to the machine hadn’t.
Conclusion
Digital marketing is no longer solely an art or a craft. What worked in the SEO era—volume, velocity and keyword density—doesn’t translate to the LLM era. Firms that hold durable AI positions build diagnostic systems capable of distinguishing high-weight signals from low-weight noise. These days, it doesn’t really matter how much you produce—it’s knowing what counts.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
