Can ChatGPT give youmortgage advice?

It can give you information, and it is good at it: what an offset account is, how a repayment is worked out, what LMI means, what to ask a lender. It cannot give you credit advice in the legal sense, because in Australia recommending a particular loan is credit assistance, which only a licensee or an authorised credit representative can provide. And it cannot see the things a recommendation depends on: today's rates, each lender's policy, and your actual file. So use it to understand, not to choose.

The arithmetic is. Ask it the repayment on $600,000 at 6.1% over 30 years and it will be right. The inputs usually are not. It does not know the rate any lender is charging today unless it searches, and even then it finds what someone published, not what the lender's system says. On 14 September 2026 the panel ran from 6.1% to 7.51%. An assistant working from a rate it remembers from training could be a point out, which on $600,000 is hundreds of dollars a month.

Roughly, and the range is the problem. On one person earning $100,000 with no debts, ten lenders' own models ranged from $526,550 to $581,992 on 14 September 2026. An assistant will give you one number, typically a rule of thumb like five times income. The real answer is which lender, and that needs each lender's servicing model, not a rule of thumb. The salary pages on this site are those models' output, and an assistant connected to Arthr can use them directly.

Connect it to Arthr's Model Context Protocol server. It is a free, open endpoint that gives an assistant tools for repayments, borrowing power, refinance savings and break-even, stamp duty, lender discharge steps, and a live serviceability check across the panel. Once connected, "how much can I borrow on $100,000" comes back with the lenders' own figures priced re-priced monthly and after each RBA decision, not a guess. The MCP page has the address and the setup steps for ChatGPT, Claude and Cursor.

For explaining, they are close. For numbers, whichever one is connected to a live source wins, and none of them are on their own. Gemini has Google Search, which finds published rates but not lender policy. ChatGPT and Claude can both connect to MCP servers, which is how they get real figures. The difference that matters is not the model, it is the data behind it.

For the decision, yes, and for the application, you have to. A broker at 1st Street Financial is licensed to recommend a loan and owes you a best-interests duty; an assistant is neither. The useful order is: use the assistant to understand the question, use Arthr's tools to get real numbers, then let a broker put their licence behind the choice.