Garbage in gospel out

Every business says its data is a mess. It's become one of those accepted truths, like saying you're “bad at maths" or "not a morning person". But when AI enters the room, that shrug stops being funny. Suddenly, the mess matters.

AI doesn't clean your data. It magnifies it. Feed it bad information, and it will return bad insights only faster, with charts. It's like having a liar who's brilliant at Excel. The illusion of accuracy makes the problem worse, not better.

The truth is that most organisations don't need more data. They need better context.

As Forrester warns, GenAI doesn’t just draw from neat, structured tables. It mines across documents, logs, and prompt histories. Cleanliness in those unstructured layers: the drafts, duplicates, and forgotten files matter just as much. Those leftovers become AI's blind spots, or worse, its misdirections.

Teams need to understand what is current, duplicated, or decaying in shared drives. They need someone to ask, "Should we even keep this?" before the machine turns it into a pattern.

Both McKinsey and Forrester point out that data quality, not compute power, is what separates high-performing AI users from the rest. The firms that win don't have the biggest datasets; they have the cleanest habits. They've stopped treating data as an archive and started treating it as a living part of their work.

Cleaning data isn't glamorous. Nobody gets promoted for deleting things. But every hour spent untangling spreadsheets or tagging content properly pays off later, when the AI produces something that actually makes sense. You can't automate understanding until you understand what you're automating.

And you don’t have to rearchitect your entire stack to start. McKinsey suggests assembling small data-quality teams that incrementally tackle your messiest domains first.

So yes, "garbage in, garbage out" still applies. Always has. But AI has raised the stakes. The gospel bit, the trust, the belief that what you're seeing is true, that's new. Once a machine says something confidently enough, people tend to stop questioning, which makes human scepticism and the people who practice it more valuable than ever.

If you want your AI to be smart, start by teaching it good manners. Clean up after yourself, label things properly, and don't assume the system knows what you meant. The miracle isn't that AI simulates thinking. It's that AI can make us think harder about what we feed it.