ChatGPT for mortgage brokers: what is safe, what to keep out, and 10 prompts

What a broker can safely use ChatGPT for

The safe uses share one feature: the prompt contains nothing that identifies a client. You describe the situation in general terms and add the names and figures yourself afterwards.

What needs care, and what should not go in

• Identifiable client financial data in a personal account. Names, addresses, dates of birth, payslips, bank statements, tax documents, credit reports and identity documents. A free or personal plan gives your business no agreement about confidentiality, and on those plans your chats may be used for model training unless you change a setting. Details are in the next section.
• Lender policy from memory. A general assistant does not have your lender panel's current policy. It will still answer, and the answer can be out of date or simply wrong. Only ask it about policy text you have pasted in, and confirm anything that decides a deal with the lender.
• Serviceability. It does not have lender calculators, assessment rates or living expense benchmarks. Use your serviceability tool.
• The recommendation. Which loan is in this client's best interests is your judgement. An assistant can help you word reasons you have already formed. It should not supply them.
• Whole call transcripts. A transcript is full of personal information even after you remove the name. Employer, suburb, income and family details together can identify a person.

ChatGPT settings and plans that matter for client data

This is what OpenAI's own pages said when we read them in October 2026.

• Personal plans can be used for training. OpenAI's page on how your data is used to improve model performance says that when you use its services for individuals, such as ChatGPT, it may use your content to train its models, and that you can opt out. Once you opt out, new conversations are not used for training.
• The setting is called "Improve the model for everyone". OpenAI's Data Controls FAQ says that when it is off, your new conversations will not be used to train its models. Turning it off does not delete saved chats. The same page says that if you give feedback such as a thumbs up or down, the whole conversation attached to that feedback may be used for training.
• Temporary chat. OpenAI says a temporary chat stays out of your history and is not used to improve its models while it remains temporary, and that it may keep a copy for up to 30 days for safety purposes.
• Business plans are different by default. OpenAI's enterprise privacy page says it does not train its models on business data by default for ChatGPT Business, ChatGPT Enterprise and the API. For ChatGPT Business it says workspace admins can control how long data is retained and that deleted conversations are removed from its systems within 30 days unless longer retention is required by law or is reasonably necessary to protect its services or a third party, and that it will sign a Data Processing Addendum for ChatGPT Business, ChatGPT Enterprise and the API.

We could not open OpenAI's page on data residency, so we have not said where ChatGPT stores data. Ask OpenAI or check your plan before assuming anything is held in Australia.

Claude AI for mortgage brokers: the same questions

Claude, made by Anthropic, is used for the same jobs, and the same split between personal and business plans applies.

• Personal plans (Free, Pro and Max). Anthropic's privacy centre says it will use your chats to improve its models if you choose to allow it, controlled by a setting called Model Improvement in Privacy Settings. Its pricing page lists model training on these plans as opt-out. Open the setting and check which way it is switched.
• Retention on personal plans. Anthropic says that if you allow your chats to be used to improve Claude, it may retain that data in a de-identified format for up to 5 years in its training pipelines. A conversation you delete is removed from your history immediately and from its back-end systems within 30 days. Chats flagged as breaching its usage policy can be kept for up to 2 years.
• Incognito chats. Anthropic says incognito chats are not used to improve Claude, even if Model Improvement is on.
• Business plans. Anthropic says that by default it will not use inputs or outputs from its commercial products, such as Claude for Work and the Anthropic API, to train its models. Its pricing page lists the Team plan with no model training on your content by default.

The Australian rules that apply

Privacy Act and the Australian Privacy Principles

The Privacy Act 1988 applies to businesses with annual turnover above $3 million and to some smaller businesses listed in s 6D(4), for example one that discloses personal information about another person to anyone else for a benefit, service or advantage, unless that is done with the person's consent or as required or authorised by law (s 6D(7)). Whether a small brokerage is covered depends on its own facts, and many brokers also carry privacy obligations through their aggregator or licensee agreement. Ask your licensee.

Where the Act applies, the Office of the Australian Information Commissioner (OAIC) has published guidance on privacy and commercially available AI products. Its key points include:

• Privacy obligations apply to any personal information put into an AI system, and to the output where it contains personal information.
• Under Australian Privacy Principle (APP) 6, personal information put into an AI system may only be used or disclosed for the primary purpose it was collected for, unless you have consent or the secondary use would be reasonably expected.
• Businesses should update their privacy policies and notifications with clear information about their use of AI.
• As a matter of best practice, the OAIC recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools.

Two more principles matter. APP 11 requires reasonable steps to protect personal information you hold. APP 8 deals with sending it overseas, and s 16C generally makes you accountable if an overseas recipient handles it in a way that would breach the principles. OpenAI and Anthropic are both United States companies.

NCCP Act: inquiries, the best interests duty and records

• Inquiries and verification. Before the preliminary assessment you must make reasonable inquiries about the client's requirements, objectives and financial situation and take reasonable steps to verify the financial situation (National Consumer Credit Protection Act 2009 (NCCP Act) s 117). An assistant can draft your question list. It cannot make the inquiries or verify anything.
• Best interests duty. The licensee must act in the best interests of the consumer in relation to the credit assistance (s 158LA). ASIC's Regulatory Guide 273 says it expects evidence of compliance to come predominantly from the broker's records (RG 273.21), and that those records generally include the options, the recommendation and the reasons (RG 273.165).
• Record keeping. A client can ask for a copy of the preliminary assessment within 7 years of the credit assistance quote (s 120). A chat history is not a client file. Anything you rely on has to be saved to the file in your own system. See NCCP record keeping requirements.
• Your licensee's rules. A licensee must take reasonable steps to ensure its representatives comply with the credit legislation (s 47(1)(e)) and must, unless it is regulated by APRA, have adequate risk management systems (s 47(1)(l)(ii)). If your licensee or aggregator has an AI policy, read it before you start. If there is none, ask.

None of these provisions mentions AI. They apply whatever tool drafts the words.

10 ChatGPT prompts for mortgage brokers that need no client data

Each prompt works in ChatGPT or Claude. Replace the square brackets with general details only. Add names and figures in your own email or CRM afterwards.

1. Explain a feature. "Explain how an offset account works on an Australian home loan to a first home buyer, in under 150 words, at a year 8 reading level. Give one worked example with round numbers and say when an offset is not worth an annual fee. Do not recommend a product."
2. Document request email. "Write a friendly email from an Australian mortgage broker asking a PAYG employee for the documents needed for a home loan application. Use a short checklist. Leave [client name] and [due date] as placeholders. Under 180 words."
3. Rate change update. "Draft a 120-word email to existing home loan clients after a Reserve Bank cash rate change of [amount, direction] on [date]. Explain what it may mean for variable repayments in general terms, invite them to book a review, and make no promises about savings."
4. Fixed rate expiry reminder. "Write three versions of a short message to a client whose fixed rate ends in about 90 days, inviting a review call. Plain and warm, no pressure, under 80 words each. Leave [client name] as a placeholder."
5. Summarise a policy document. "Here is a lender policy document: [paste the lender's text]. Summarise what it says about [topic, for example casual income] in dot points. Quote the exact sentence for each point. If the document does not cover something, say so and do not guess."
6. Fact find questions. "List 15 open questions an Australian mortgage broker could ask a couple refinancing their home to understand their requirements, objectives and plans for the next five years. Group them under goals, income, expenses and future changes."
7. Plain English check. "Rewrite the following explanation so a client with no finance background can follow it. Keep every fact the same and flag anything unclear: [paste your own wording with names and figures removed]."
8. Referrer email. "Write a short introduction email from a mortgage broker to a local accountant proposing a referral relationship. Australian English, under 130 words, one clear next step."
9. Social posts. "Give me five LinkedIn post ideas for an Australian mortgage broker this month, each with a one-sentence hook. No rate predictions, no guarantees, nothing that reads as personal advice."
10. Process checklist. "Create a checklist for a loan processor covering the steps from signed application to settlement for an Australian residential purchase. Mark the steps where the broker must make the decision personally."

A prompt for turning a de-identified call summary into a file note

If your licensee allows it, you can use a general assistant to structure a summary you wrote yourself. Write the summary from memory or your handwritten notes. Replace the name with "the client", round the figures, and leave out employer, address, dates of birth and account details.

This is the kind of result it gives. The client is fictional and the detail is deliberately general.

The limits of this method

• It is only as good as your summary. The assistant never heard the call. Anything you forgot is missing from the note, and nothing tells you it is missing.
• De-identifying takes time and loses detail. The client's own words and exact figures are what make a note persuasive. RG 273.165 expects records of relevant conversations, and a rounded, anonymised version is weaker until you put the detail back.
• It can still add things. Even with the instruction above, read every line. Delete any reason or fact you did not write.
• Nothing is locked. The output is text in a chat window, with no record of later changes until you save it to the file.
• It is written later. RG 273.169 says taking notes contemporaneously may help you capture your reasoning at each point. A note rebuilt from memory at the end of the day is further from that.

For the structure itself, see our mortgage broker file note template and how to write a compliant NCCP file note. The NCCP file note checklist is a quick test of a finished draft.

Where a purpose-built file note tool differs

CallNote is one tool of the second kind. It takes a transcript that already exists, from Aircall, Dialpad or Microsoft Teams automatically, or pasted, uploaded, forwarded by email or dictated as a voice memo, and writes the note in your own format. It never records calls and never stores call audio. You review the note and publish it, and it is then sealed and locked with append-only amendments. Data is stored in Sydney and is not used to train AI. The note is written by Anthropic's Claude models through its commercial API, so the transcript is processed outside Australia under business terms. It costs $149 a month Solo or $99 a seat a month for teams, with a 14-day free trial.

It is also narrower and dearer than a general assistant. It has no connection to Mercury Nexus, MyCRM, Salestrekker or other broker CRMs, so the note goes across by copy and paste or as a PDF. If you take few calls and write good notes the same day, you may not need it. We compare the options in AI note takers for mortgage brokers and list the wider field in the best AI tools for mortgage brokers. Our mortgage brokers page shows a broker call end to end.

To turn this into rules for your office, start from our AI note taking policy template.

Common questions