What Happens When You Ask an AI a Legal Question
A plain-English look inside large language models: what they are really doing when you ask about Kenyan law, and why it changes how you should use them.
When you type a legal question into an AI chatbot, it is not searching a library of Kenyan statutes and judgments and reading you the answer. It is doing something stranger, and understanding it changes how much you should trust what comes back.
Here is the short version: a large language model predicts the next most likely word, over and over, based on patterns it absorbed from enormous amounts of text during training. That is the whole trick. It is astonishingly good at sounding right — because sounding right is precisely what it was trained to do.
Why that matters for legal work
Two consequences follow directly, and both matter in practice.
First, the model has no live connection to the law unless someone gives it one. Ask a bare chatbot about a section of the Employment Act and it answers from a blurry memory of text it saw in training — which may be outdated, incomplete, or quietly wrong. It is not reading the Act. It is remembering the general shape of things people have written about the Act.
Second, confidence is not knowledge. The model produces fluent, authoritative prose whether or not the underlying claim is true. There is no little voice inside it saying “I’m not sure about this one.” That is why a citation that looks perfect can point to a case that does not exist.
The fluency is real. The reliability is not automatic. Those are two different things, and the gap between them is where advocates get burned.
So what makes it useful anyway?
Prediction-from-patterns is genuinely powerful for the right jobs: restructuring a rambling set of facts, drafting a first version of a letter, explaining an unfamiliar concept in plain terms, or spotting the issues in a scenario. These are tasks where fluency is the value and you remain the one who verifies.
The failure mode — and the fix — is what we cover in AI Risks & Verification. The one-line version: never let the model be your source of authority. Let it draft; you verify against the primary source.
That is also the entire design philosophy behind JuriBase. Instead of asking a model to remember Kenyan law, we connect it to the actual case law and legislation and make it show you where every answer comes from — so you can click through and confirm before you file.
Frequently asked questions
Does an AI chatbot look up the law when I ask a question?
Not by default. A plain language model answers from patterns learned in training, not from a live search of statutes or case law. It only consults primary sources if it has been connected to them — which is exactly what a grounded tool like JuriBase does.
Can I trust an AI's answer on a point of Kenyan law?
Treat it as a well-read but unverified junior: useful for a first draft or orientation, never as authority on its own. Always confirm the actual statute or judgment before you rely on it.
Alvin Otanga
AI engineer and researcher building grounded, verifiable legal AI for Kenyan advocates.
Alvin is the founder of JuriBase, where he builds AI that is grounded in Kenyan case law and legislation — so advocates get answers they can actually verify and cite. He writes here to teach lawyers how AI really works, from the inside.