Why "AI-Ready" Content Is The Wrong Goal
Every Learning and Development (L&D) and Knowledge Management (KM) leader with whom I speak right now says they are often asked some version of the same question: "Is our knowledge base ready for AI?" I believe this is the wrong question, and it's sending a lot of good teams down the wrong path.
Being "AI-ready" frames this as a content project where I believe it's a governance problem. "AI-ready" projects usually mean tidying up the wiki, retagging some pages, maybe even running everything through a chatbot pilot to test its performance on a company's internal data. This is what many teams do and find to be a superficial, cosmetic step. Many companies treat AI readiness as a simple "content cleanup" and fail to recognize the importance of addressing deeper, more critical issues of data security, permissions, and long-term governance.
Here's an uncomfortable truth: connecting a Large Language Model (LLM) to your company's Confluence space, SharePoint site, or LMS content library amplifies the existing mess. Instead of cleaning up the clutter, AI tools act as a megaphone for your worst documentation. This creates internal chaos—employees may unknowingly follow old escalation paths, adhere to outdated onboarding modules, superseded compliance training, a course an AI copilot cites as "current" policy when it was retired two years ago. And then there's the risk of hallucination where AI models confidently present outdated or wrong answers as absolute fact…and your people believe them!
Your corporate connected AI will find every unreviewed training content, every "draft that never got finished," and every policy that was superseded months ago—and these become candidate answers the AI hands with high confidence to your employees.
The AI doesn't know the difference between your official escalation procedure and the abandoned draft course someone started before they left the company. It just knows both exist, both use the right vocabulary, and so it treats both as valid input. That's the core challenge: retrieval isn't judgment. The AI can surface what matches, but it can't yet weigh what's true, current, or safe to act on. AI lacks critical thinking—the very discernment my PhD research argued has to be deliberately built into any learning system, not assumed to emerge from the technology alone.
I've spent the last two years leading enterprise knowledge platform rollouts, including establishing a group-wide learning and knowledge function from scratch across a multi-country, multi-acquisition pan-European organization and building the governance layer underneath an AI-connected Confluence environment. The pattern I keep seeing is this: organizations invest heavily in the AI layer and treat the trust layer underneath it as an afterthought. That ordering needs to reverse.
I argue that companies are rushing to buy expensive AI tools while ignoring the messy data and security risks underneath. Organizations must secure and organize their internal information before connecting it to AI, not after. What does this mean in practice? From my experience, companies focus on the shiny, expensive chatbot or software, the so-called "AI layer," where they spend massive budgets because it feels innovative. However, they ignore the underlying data quality, security permissions, content accuracy, and governance policies—what I call the "trust layer."
Many organizations believe that installing AI will magically fix their organizational data. However, real-world experience shows this simply creates a system that confidently feeds employees outdated drafts, incorrect policies, or restricted data because the underlying "trust layer" was treated as an afterthought. If your foundation is weak, you will end up having untrustworthy AI. You need to first prioritize cleaning data, setting up strict permissions, and establishing governance rules before deploying your AI layer.
The Three Things AI Can't Tell You About Your Content
While AI excels at processing text, it lacks the human intuition needed to distinguish between authoritative, current information and outdated, personal notes in unorganized enterprise data. They are not equipped, on their own, to answer these three questions that matter enormously to a business:
1. Is this the current version, or a superseded one?
Most enterprise wikis have no reliable lifecycle status, for example, draft, in review, approved, retired, etc., attached to content in a way a machine can parse. Humans infer currency from context clues (who wrote it, when, are they still around, what channel it's in). AI has none of that instinct.
2. Is this authoritative, or someone's working notes?
A page in the "Engineering" space and a page in someone's personal space can look identical to an intelligent retrieval system. Without an explicit trust classification, both get pulled into an answer with equal confidence.
3. Who is accountable for this being correct?
Ownership is usually implicit—"Everyone knows Sarah maintains that page"—which means it's invisible to any system, human or AI, that wasn't in the room.
None of these are AI problems. This is a critical "knowledge governance" problem that AI has stopped letting us ignore, where feeding messy, unstructured data into AI systems creates confident, scalable misinformation. Fed into a retrieval-augmented AI system connected to your corporate data, wrong answers may be delivered with total confidence to a customer, an auditor, or a new hire—at scale, and fast. The core takeaway is that businesses must implement strict, explicit data management, ownership, and lifecycle tracking to prevent AI from amplifying operational risks.
What A Working Trust Layer Actually Looks Like
In practice, the fix isn't a sweeping company-wide content audit. It's a small number of machine-readable signals applied consistently, so that both people and AI can tell good content from stale content at a glance.
What's new in our organization is that we now have two audiences: our human team and our AI helpers. Information needs to be structured so both can read and act on it reliably.
The structure I've found holds up under real use has three parts:
1. A trust classification, applied at the page level, that distinguishes formal standards and policy from best-practice guidance from "useful but informal" notes. This is the single highest-leverage signal you can add, because it lets an AI system—or a search index, or a new employee—weight sources appropriately instead of treating a Slack-adjacent tip and a compliance policy as equally authoritative.
2. A lifecycle status, equally simple: draft, in review, final, retired. The point isn't process for its own sake—it's giving any downstream system, human or automated, a clean signal of whether this page should be trusted right now.
3. Clear content ownership, attached to the page itself rather than living in someone's head or an org chart three systems away. When something is wrong, someone needs to be findable in seconds, not after three Slack pings and a guess.
The technical detail that surprised me most in building this: AI tools reliably read page titles and body text, but they often can't see labels, app-level metadata, details in graphics (for example, org. charts and who reports to whom), or workflow status set through add-ons layered on top of the platform. If your trust signal lives somewhere the model can't see it, it doesn't exist as far as the AI is concerned—no matter how carefully your team applied it and the excellent graphics they created. To the AI, it is not there, full stop! That single finding should shape how any organization designs its governance layer: the signal must live in the content itself, not bolted on beside it.
This Is Compliance Work Now, Not Just Content Hygiene
There's a reason this is climbing the priority list faster than most L&D teams expected. As AI use transitions from casual internal experimentation to customer-facing, Human Resources (HR), sales, and finance, regulated-decision contexts, and basic everyday business applications, the bar changes.
Once AI outputs touch things with real consequences, e.g., a customer commitment or generating a report to the CFO, the question stops being "Does our AI sound helpful/plausible?" and becomes "Can we trace where that answer came from and prove it was a current and approved source when it was given?" That's an audit question. It's the kind of question a regulator, a customer contract, or an internal risk committee will eventually ask, and "the AI said so" is not an answer that, in a dispute, an audit, or a compliance review, holds water. You need to be able to show that the underlying source was correct, current, and approved.
Most organizations currently have no evidence trail for AI-generated answers whatsoever. They can't show which source document an answer was generated from, whether that source was the approved version, or who was accountable for it being correct at the time. That gap, between what AI can now do and what organizations can actually prove about what it did, is where the real exposure sits, and it's growing faster than most governance functions are moving.
Where To Start Monday Morning
You don't need a platform overhaul to begin addressing these points. These three moves, in order, will help:
1. Pick one high-traffic space and classify it first.
Don't try to boil the ocean. A pilot in your most-used knowledge area—onboarding, IT support, a single product line—will surface your real governance gaps faster than any policy document.
2. Make trust and lifecycle status visible in the content itself, not in a separate tracking system nobody checks.
If it's not visible where the AI (and your people) actually look, it isn't governance, it's paperwork. Remember, you have two audiences consuming your content now, not one.
3. Name an owner for every page that gets classified as authoritative.
If you can't name one, that's your answer about whether the page belongs in the authoritative tier at all.
Takeaway
So, here's my take on where competitive advantage in enterprise AI will actually come from. Organizations that extract lasting value from AI will not be those with the most advanced models. They will be the ones that can declare, with evidence, exactly what their AI knew, when it knew it, and who stood behind it. The real differentiator will be the ability to produce a defensible evidence trail that answers those questions with precision. In other words, the winners will be able to say, with evidence rather than assumption: "This is exactly what the system knew, when it knew it, and who stood behind that knowledge."
To reiterate, this is not a "content project" (i.e., cleaning up or organizing documents). It is a governance capability, the organizational muscle that creates, maintains, and surfaces provenance, version control, and accountability around the knowledge AI uses. My observation is that most organizations are currently treating this as a reactive exercise: they build the governance layer after AI systems are already in production and problems appear.
The winners in the AI age already understand that capability needs to be designed and installed before widespread deployment if organizations want to capture real value rather than manage accumulating risk.
The advantage will belong to those who can prove what their AI knew, and on whose authority. Most organizations are still building that proof after deployment, once the damage is already done.
I spent the last two years building exactly this governance layer from the ground up—inside a multi-country, multi-acquisition organization, underneath a live AI-connected Confluence environment. The lesson I'd pass on: it's not a technology decision. It's a design decision, and it must be made before the AI goes live, not after.
