AI provides answers in a matter of seconds, but we rarely ask where the answer comes from. Here, we look at why source criticism is still needed in the workplace and what is required to ensure that an organisation’s knowledge can be trusted.
Not so long ago, an employee would search for answers across several sources. An intranet, a couple of colleagues, perhaps a file. They compared the information. They assessed which source seemed most up to date, and they questioned what they found, often without realising that they were, in fact, practising source criticism.
Today, the same question is put to an AI chatbot. The answer arrives in three seconds, phrased with absolute certainty. And it is precisely that certainty that is the problem — a self-assured answer is rarely perceived as a suggestion to be scrutinised. It is perceived as the definitive answer.
Critical evaluation of sources has not disappeared. It has shifted — from the person making the enquiry to the AI providing the answer. But who is scrutinising the sources now?
The easy answer is to blame the model. But in the vast majority of cases, the AI isn’t the problem. The problem is what it’s based on: old PDFs that were never archived, three versions of the same procedure in three different systems, an intranet page that nobody was responsible for after the reorganisation.
An AI model is only as good as the knowledge it has access to. If that knowledge is scattered, contradictory or out of date, so too will the answer be — merely formulated more convincingly than before.
This is no longer a matter for the IT department alone. It is a matter of quality, traceability and accountability — and is therefore increasingly landing on the desks of business development managers, quality managers or COOs.
Three situations that highlight the problem
The quality incident. Someone acted on an AI-generated response that turned out to be based on an outdated procedure. In hindsight, no one can say who approved the original information, or when it was last validated.
The Copilot roll-out. The organisation has invested in an AI assistant linked to internal documents. But no one has actually checked which documents it is retrieving from — or whether they are up to date.
The key person leaving. The knowledge held in the mind of a single employee is not documented in writing, and the AI, of course, cannot find it either. Instead, it makes guesses.
A common objection is that the organisation already has a quality management system, a document management platform or procedures for updating. It’s reasonable to think that this is enough.
But a document management system simply stores files — it does not ensure that what is actually displayed, whether to a human or an AI agent, is the approved and most recently validated version. And procedures are of little help if three similar documents are stored in three different systems and nobody knows which one is the correct one anymore.
The difference lies not in storing more knowledge, but in quality-assuring it to the greatest possible extent, regardless of whether it is a human or an AI agent who is to utilise it.
Moving from scattered knowledge to quality-assured knowledge involves three interlinked elements:
Version control — when a procedure is changed, everyone — both people and AI tools — should automatically see the new version, not an old copy that happens to have been left behind.
Real-time traceability — AI agents must be able to retrieve knowledge directly from the verified source via open interfaces such as MCP, rather than relying on outdated training data. Furthermore, every response must be traceable back to the original source.
Access control — the right content must reach the right users. If an employee does not have access to a guide, they must not be given answers to the content via an AI assistant either. Access rights accompany the knowledge every step of the way, regardless of where it is displayed.
That is the difference between storing knowledge and managing it — which version is valid, where the answer comes from, and who is allowed to see what.
AI has not made source criticism unnecessary. It has made it invisible — the answer looks too good to be questioned. But any organisation that can answer the question “who has verified this, and when?” has a significant head start, regardless of whether the question comes from an employee or an AI agent.
That is precisely the question we are helping over 300 organisations to answer with confidence. And we’d be happy to show you more in practice what this could look like for you.
Book a demo and we’ll show you how to give both people and AI agents one verified source of truth.