Glossary
This page defines every term the guide links to, in plain English and for education rather than advice. Any number a term depends on, such as a limit, rate, or age, lives on This year’s figures rather than here.
Advice
A recommendation made for your particular situation by a person who takes responsibility for it, usually under a licence or a professional duty. An answer from an AI assistant is not advice, however personal it sounds, because nobody stands behind it.
Anonymise
To remove or replace the details in a question or document that would identify you or your accounts, keeping only what the answer turns on. Names become roles, account numbers are dropped, and exact figures become round ones where exactness does not matter.
Assumption
A fact a projection or recommendation treats as settled without saying so: a rate of return, a span of years, a situation that does not change. Changing one assumption and seeing whether the conclusion survives is the quickest test of how much the conclusion is worth.
Decision note
A short written record, kept with the relevant paperwork, of a money decision that an AI assistant informed: the date, the question, what the tool said, the source it was checked against, and what you decided. It is what lets you show what you relied on and review how it turned out.
Hallucination
A confident, fluent statement from an AI assistant that is not true: a rule that does not exist, a figure that was never published, a source nobody wrote. It reads the same as a correct answer and can only be caught by checking outside the tool.
Language model
The system behind an AI assistant: software trained on a very large body of text to predict what word plausibly comes next. It composes replies rather than looking them up, which is why it is fluent, confident, and sometimes wrong.
Leading question
A question whose wording signals the answer the asker expects, such as “why is this a bad idea”. An AI assistant tends to answer a leading question in the direction it points, so the reply reflects the question more than the facts.
Personal data
Any detail that identifies you or could be used to get into your accounts: your name, address, date of birth, identity and account numbers, credentials, and documents that carry them. The tool rarely needs any of it to answer a money question.
Primary source
The original place a fact comes from rather than a summary of it: the statute, the agency’s own page or form, the lender’s own agreement, or the research paper. For a money rule, it is the document a court or regulator would point to, not an article or an answer about it.
Prompt
The text you type into an AI assistant: the question, any facts you include, and any instructions about how to answer. The reply is shaped by the prompt more than by anything else, which is why the fourth step of the loop is about writing it.
Secondary source
An account of a fact rather than its origin: a news article, a blog post, a summary, or an answer from an AI assistant. It can point you to the primary source and is useful for that; it does not replace it, because it can be out of date or wrong in ways the original is not.
Training cutoff
The date up to which the text used to train a language model was collected. Rules, rates and products that changed after it are absent from the model or present in their old form, unless the tool has been given a way to look things up.