Stop Chasing AI Protocols: Fix Your Knowledge Base First

TL;DR: Every new AI protocol — llms.txt, MCP, markdown feeds — creates organizational scramble without fixing the underlying problem: incomplete, fragmented knowledge. Operators who build a canonical knowledge base first and publish to formats second will outperform those who implement every new spec on arrival. The format is never the strategy.

The AI FUD Tax Is Real and It Is Expensive

Here is how the cycle runs. An AI visibility vendor sends an audit flagging that your site lacks an llms.txt file. A senior executive reads the report and now needs answers: Is this valid? Why haven’t we done it? How far behind are we? Who owns this? Marketing and engineering get pulled into meetings that produce a ticket, a timeline, and a budget line — for a file format that may or may not move the needle on a single conversion.

Multiply that by every new protocol that surfaces over a quarter and you have what Bill Hunt calls the AI FUD Tax: the organizational cost of repeatedly reacting to the latest external AI audit. The individual implementation is cheap. The decision overhead is not. For operators running performance campaigns at $10K+ per month, that overhead compounds fast — especially when engineering time gets redirected away from actual revenue infrastructure to chase a spec that is still being debated on LinkedIn.

The formats themselves — llms.txt, MCP, schema markup, markdown outputs — are not the problem. Each does something specific: some help machines discover content, some define how systems exchange data, some provide alternate representations of existing pages. Treating them as interchangeable, or treating any one of them as a strategy, is the mistake. They are delivery mechanisms. What they deliver is what matters.

Decision Coverage: The Metric That Actually Diagnoses the Gap

Before any format question, there is a more useful diagnostic: does your organization have the knowledge required to support a real customer decision?

Hunt frames this as Decision Coverage — measuring how completely you have exposed the evidence an AI system needs to evaluate, compare, qualify, and confidently recommend your product or service. A hotel listing that says “beachfront” does not automatically qualify for a recommendation when a user asks for the “best family-friendly beachfront resort in Cancun under $400 a night with a kids’ club.” The AI is evaluating five or six criteria simultaneously. If the property cannot substantiate each one with authoritative evidence, it does not appear — not because it failed a technical audit, but because it failed to provide the evidence required to make the cut.

Transpose that to a regulated vertical. A forex broker answering a prompt about “best brokers for US residents trading indices with low spreads” needs to satisfy: regulatory status, instrument availability, spread data, deposit minimums, platform compatibility, and ideally independent review signals. If four of those six criteria are documented and two are not, publishing those same four through a new protocol does not establish the missing two. It just makes the same gap available in another format.

This is where most AI-readiness audits fail operators. They flag the presence or absence of a file. None of the 100-plus agentic readiness audits Hunt reviewed critiqued the depth or quality of the content inside an llms.txt for sites that had one. Format compliance is auditable. Knowledge completeness requires actual work.

A New Format Cannot Fix Missing Knowledge

The pattern is consistent across verticals. An operator gets a competitive alert that a rival is getting cited in AI-generated answers. The immediate response is: more content, more schema, more structured data, maybe an MCP endpoint. What rarely happens is a structured deconstruction of the decision criteria driving those citations.

An expanded schema can describe relationships between entities. It cannot determine what those relationships should be. MCP can make multiple organizational resources accessible to AI systems. It cannot determine whether those resources contain the answers a customer actually needs. An llms.txt file can point machines toward information. It cannot compensate for information the organization never created.

Alex Moss made a closely related point in a recent Search Engine Journal piece: technical SEO needs to focus on data integrity — accurate entities, explicit relationships, machine-readable formats, reliable perception signals. That argument holds. But data integrity is a downstream problem. Before you can keep information accurate and synchronized, you need to determine what knowledge should exist, who owns it, and which source is authoritative. Knowledge architecture comes before data integrity. Both come before format selection.

For operators in high-CAC verticals — iGaming acquisition, legal lead generation, crypto exchange onboarding — the cost of an AI system failing to recommend you is not a vanity metric problem. It is a direct revenue problem. Running a thorough performance marketing audit that surfaces knowledge gaps, not just technical gaps, is the starting point operators should demand before any format implementation.

Build the Canonical Base Once, Publish Everywhere

Hunt’s operating principle is deliberately simple: build the canonical knowledge base once, then publish to every format from that single source. The foundation contains the facts, relationships, policies, comparative evidence, and customer decision criteria the organization can authoritatively establish. Those elements live in a governed, connected knowledge source — independent of any particular publishing format.

From that base, every delivery mechanism — web content, schema markup, Merchant Center feeds, APIs, markdown, MCP, llms.txt — draws from the same source of truth. Updates happen once. Synchronization problems shrink. New formats become publishing decisions, not reconstruction projects.

Operators running paid media at scale already understand this principle from feed management: a single product feed with clean attributes pushes to Google Shopping, Meta catalog, and TikTok catalog simultaneously. The moment you maintain separate feeds per platform, version drift kills campaign performance. The same logic applies to AI knowledge infrastructure. Maintain one authoritative base. Publish everywhere.

The organizations best positioned as AI search matures will not be those that implemented llms.txt in Q3 2025. They will be those that built knowledge architecture clean enough that supporting the next useful format requires no additional reconstruction.

What This Means for High-CAC Vertical Operators

Operators in forex, crypto, iGaming, and legal share a structural challenge: the decisions customers make before converting are complex, multi-criteria, and high-stakes. An AI recommendation engine evaluating those decisions needs substantially more evidence than a typical e-commerce query.

A CDL driver comparing carriers before signing with a recruiter is weighing home time, pay structure, equipment type, run preferences, and benefits. Operators running CDL driver recruitment who document all five criteria clearly and authoritatively will outperform those who only publish pay rates. The same applies to a prop firm competing for funded trader signups — a forex lead generation strategy built on thin product pages will underperform against a competitor with thorough challenge rules, payout evidence, and independent review data published in machine-readable form.

For crypto operators, the bar is higher still. Token launches and exchange onboarding involve regulatory, security, and liquidity criteria that AI systems will increasingly use to filter recommendations. Operators investing in crypto exchange growth who have not yet mapped their Decision Coverage gaps are building on a weak foundation regardless of which protocols they implement.

Legal operators face the same dynamic. A mass tort intake page optimized for schema but missing qualifying criteria — statute of limitations by state, injury type eligibility, case value ranges — will not fare well in an AI-mediated intake flow. Law firm marketing that documents those criteria explicitly, in authoritative and machine-readable form, builds the evidence layer that protocols then carry to AI systems.

The practical starting point is not a new file format. It is a structured audit of the decision criteria your highest-value customers evaluate before converting, mapped against the evidence your organization currently publishes. Where the gaps are, fill them. Then configure your delivery mechanisms — whichever ones are appropriate — to draw from that complete base. Precision targeting works the same way: the signal quality of your audience data determines what the algorithm does with it. Better inputs, better outputs. No format substitutes for that.

Originally reported by Search Engine Journal, September 2026.