Articles13 November 2026
Why AI gives your business wrong answers, and how to fix it
An AI assistant can only be as right as the documents it reads. Most disappointing first attempts trace back to the state of the data, not the AI.
A common first experience of AI at work goes like this. Someone connects an assistant to the company’s documents and asks it a question they already know the answer to. The reply is fluent, confident and wrong. The conclusion is that AI is not ready.
Usually the AI is fine. The trouble is what it was given to read.
What “grounding” means
A general AI model knows a great deal about the world and nothing about your business. Grounding is the practice of giving it your own information to answer from: your price list, your policies, your project records. A grounded answer is drawn from those sources and can point to them.
Grounding is what turns a clever general tool into one that is useful for your work. It also means the tool inherits every flaw in the information it is given.
The three usual problems
Duplicates. If a shared drive holds four versions of the price list, the assistant has no way to know which is current. It may quote any of them.
Out-of-date material. Old policies, expired contracts and abandoned drafts all look like valid sources. People know to ignore them. Software does not.
Missing knowledge. A great deal of what a business knows is in inboxes and in people’s heads. If it is not written down anywhere the assistant can reach, it does not exist.
What getting ready looks like
None of this needs advanced technology. It needs housekeeping, done in a sensible order.
- Pick one area where good answers would save real time, such as customer queries or internal policies.
- Find the definitive version of each document in that area, and archive the rest.
- Write down the important things that exist only in people’s heads.
- Decide who is allowed to see what, before the assistant can see everything.
- Name someone responsible for keeping that area current.
Then connect the assistant to that one clean area and test it with questions you know the answers to.
The useful side effect
Work done to prepare for AI pays off even if you never switch the AI on. Staff find things faster, new starters learn from documents that are correct, and the business depends less on any one person’s memory.