How it produces an answer
Tools such as Claude, ChatGPT, Microsoft Copilot and Gemini are built on a large language model. The model learned patterns from a very large amount of text, and it writes by predicting what text should come next. It does not look facts up the way a search engine does unless the tool is set up to search, and it does not know your documents unless you give them to it.
In plain terms
It writes what sounds right. Usually that is also what is right, but not always, and it sounds equally sure either way.
What it does well
- A first draft of a letter, clause, policy or summary, so you start from something instead of a blank page.
- Pulling dates, parties, amounts and renewal terms out of a long document into a table.
- Comparing a document against a list of positions you give it.
- Explaining a term or clause in plain language, which you then check.
Where it goes wrong
- It can invent things. A clause number that does not exist, a case that was never decided, a quote that is close but not exact. This is called a hallucination.
- It can leave things out. A summary can read well and still miss the one exception that matters.
- Its knowledge stops at a date. The model learned from material up to a cutoff, so it may not know recent law or events.
- It only sees what you give it. If the contract refers to terms on a website, the tool does not have them unless you provide them.
What this means for you
Use it for the first pass and check the result before anyone relies on it. The rest of this section shows you how: which tool and plan to use, what it can do on its own, the words you will hear, how to ask, what is safe to share and how to check.