AI for Australian Financial Planners: Can It Draft a Statement of Advice?

Yes. AI can draft parts of a Statement of Advice (SOA) from verified client facts and planner-approved recommendations. The planner still owns the scope, inquiries, product investigation, best interests assessment, advice, fee and conflict disclosures, and final approval. Check the current RG 175 and licensee procedures before using any generated wording or issuing the document.

The tempting shortcut is to upload a fact find and ask for a complete SOA. That reverses the work: the system fills gaps with plausible assumptions, then presents them as advice. A safer hand-off starts only after the planner has documented the client's relevant circumstances, the options considered, the reason for the recommendation and the disclosures that apply.

Follow me on Instagram@sagnikteaches

Separate the advice decision from document assembly

An SOA is the written explanation of personal advice to a retail client, not a form the planner completes after an algorithm has selected a product. ASIC says it should help the client understand and decide whether to rely on the advice. Its guidance calls for the advice and reasoning, certain remuneration and benefits, relevant conflicts, and significant consequences of switching products to be set out clearly, concisely and effectively. The details depend on the actual advice. A fluent document can still fail if the underlying recommendation was not investigated.

Connect on LinkedInSagnik Bhattacharya

AI is useful at the assembly stage: putting approved facts into a consistent section order, turning a planner's reasoning note into plain language, checking a draft for missing headings, and comparing a proposed disclosure paragraph with the approved source. It can prepare a question list from contradictory notes. It should not determine the scope solely from a meeting transcript, decide that a client can tolerate a loss, choose a fund from a marketing sheet, or state that a switch is in the client's interests because the new product has a lower headline fee.

Subscribe on YouTube@sagnikteaches

Write a hand-off sheet with two columns. Planner decision includes the client's objectives, material needs, scope, alternatives investigated, recommendation, reasoning and any limitation or conflict. Drafting support includes ordering, summarising, plain-language explanation and cross-reference checks. If the input for a drafting section has no completed planner decision, the output should be a review question rather than a guess. The adviser CRM and AI feature comparison can help keep document automation distinct from advice.

ASIC's guide to digital advice, RG 255, says the Corporations Act is technology neutral: the obligations for traditional and digital advice are the same. That does not mean an AI-produced SOA gets a lower review standard. The planner and licensee must still meet their obligations and show how they did so. A document generator may speed up the wording; it does not inherit the duty to make reasonable inquiries or place the client's interests first.

What the planner signs off before AI writes prose

Use the current RG 175 and the licensee's approved process as the authority. ASIC's advice conduct and disclosure guidance explains the best interests duty, appropriate advice, incomplete-information warnings, SOA content and record keeping. The checkpoints below translate those topics into a drafting workflow. A compliance lead should adjust them for authorisations, client type and advice scope.

Check the document's name and required content as well. In March 2025 the government released draft legislation that would replace the SOA with a “client advice record” as part of its Delivering Better Financial Outcomes reforms. ASIC's guidance read for this tutorial in November 2026 still describes the SOA obligation, so confirm with your licensee whether any replacement has commenced before rebuilding templates. The planner-owned checkpoints above apply to either document.

A single green tick beside “fact find complete” is too blunt. The planner should identify information that is missing, stale or internally inconsistent and decide whether further inquiry is needed. If advice is based on incomplete or inaccurate relevant information after reasonable inquiries, ASIC describes a warning obligation. AI can flag the missing field; it cannot decide that a statutory warning makes a poorly investigated recommendation acceptable. Stop the drafting hand-off where the absence of information changes the advice.

For example, a meeting note says the client wants to retire in six years, while an older fact find says ten. An AI summary that picks the more recent statement without confirmation may be wrong: the note could refer to a partner. The planner should confirm the objective, record the answer and reassess the strategy. Only then should the drafter update the SOA. A highlighted conflict in the notes is more valuable than a smooth paragraph that hides it.

A $240,000 switch: the arithmetic is the easy part

Consider an illustrative client with $240,000 in an existing investment account. A proposed account has an estimated annual platform fee of 0.50%, compared with 0.30% for the existing account, on the same assumed balance. That is $1,200 against $720, a $480 annual difference. Assume a one-off $300 exit or transaction cost. The first-year fee difference and one-off cost total $780. These invented figures demonstrate review; they do not describe a product or support a recommendation.

AI can calculate that arithmetic and draft: “On the assumed $240,000 balance, the proposed platform fee is approximately $480 higher each year; a further $300 one-off cost may apply.” The planner must verify the fee bases, whether tiers or caps apply, what the account actually holds, and whether the estimates compare the same services. An AI model may multiply correctly while comparing a platform fee on one side with a total product cost on the other. That is a false comparison with correct arithmetic.

The planner must then investigate the client's objectives and relevant options. Perhaps the proposed account offers a feature the client values; perhaps the current account already provides it. Perhaps an existing benefit would be lost, or a tax or insurance consequence matters. No generic prompt can determine these from the $480 difference. The SOA needs to explain the chosen strategy and the significant consequences of a switch, including lost benefits where relevant. If the investigation does not support switching, the planner should not force the draft towards the product initially discussed.

Here is a useful correction exercise. An illustrative AI sentence says: “Switching will save you money and improve performance.” The figures above show a higher platform fee, not a saving, and no performance evidence has been supplied. The review note should say: “Delete both claims. State the verified fee difference. Planner to document whether any non-fee benefit supports the proposed strategy.” The final advice may be to retain the existing account, to switch for a justified reason, or to defer. That decision belongs to the planner and must be evidenced.

The comparison should have a source date. A fee schedule copied from a prior client file can be stale or apply to another balance tier. Keep the current product disclosure material or other approved source, the calculation, and the planner's interpretation together. Do not ask AI to browse for a fee and quietly treat the first result as definitive. For a live recommendation, the licensee's research and product approval rules determine the sources and review route.

Draft from an evidence pack with missing-field flags

Give the system a compact evidence pack rather than a raw folder dump. It should contain only fields approved for that environment: client reference, advice scope, verified circumstances, planner-selected strategy and product, alternatives considered, reasons, fee comparison, switching effects, conflicts and approved disclosure wording. Each item needs a source or a reviewer. An unknown value is marked unknown. Do not replace it with a typical value to make the prose look complete.

Draft the explanatory sections of an SOA from the approved evidence below.
Do not select a strategy or product. Do not infer missing client facts,
fees, benefits, risks, dates or legal conclusions. Preserve the planner's
decision and the licensee's approved disclosure wording. For every claim
that lacks a source, write REVIEW REQUIRED outside the client draft.

Advice scope: [planner-approved]
Relevant client circumstances: [verified items with dates]
Planner recommendation: [approved decision and rationale]
Alternatives considered: [planner's assessment]
Fees and switching effects: [checked calculations and sources]
Conflicts and remuneration: [approved disclosure]
Unresolved questions: [list]

Return: client-facing draft, then a separate review register with each
claim, its evidence source and any missing or conflicting input.

In an illustrative trial, the planner's evidence pack says a client wants to reduce ongoing fees but gives no confirmed current fee schedule. A safe output does not claim a saving. Its review register says: “Current account fees unverified; comparison cannot be completed.” A bad output assumes that the proposed product's advertised fee is lower and writes a persuasive switch rationale. The reviewer should reject the draft and collect the missing source, not patch the sentence with a disclaimer.

Run a second trial with a deliberate conflict: one note records an advice fee of $2,000, another approved engagement record says $2,400. The model should surface the conflict, not average the figures or choose the later date automatically. The planner or authorised colleague resolves the fee and documents why. Only then is it suitable for the SOA. A system that repeatedly hides conflicts is unsuitable for this stage even if its prose is easy to read.

Use plain language without weakening the reasoning

AI can help turn a long technical note into language a client can follow. Keep the chain visible: client objective, relevant fact, option considered, reason for the advice, important cost or risk, and what the client must decide. “This product offers flexibility” is weak if the client cannot see which flexibility matters. “You said you may need to draw on these funds within three years; this option allows the withdrawal pattern described in the product terms” is more useful, provided the planner has verified the need and the product feature.

An illustrative before-and-after shows the boundary. Before: “The recommended portfolio is suitable for your moderate risk profile and will help you meet your goals.” After planner review: “You told us that keeping an accessible reserve is more important than seeking the highest possible return. We have recommended holding $40,000 outside the proposed investment allocation while reviewing the remaining balance against your agreed timeframe. The portfolio can fall in value; we discussed the effect of a loss on your planned withdrawal.” The $40,000 is an invented example. The real amount and reasoning must come from the advice file.

Do not let simplification remove qualifications. If a recommendation depends on the client retaining an emergency reserve, a shortened summary should keep that condition. If an option involves loss of insurance cover, the consequence must not disappear because the model was told to be concise. Ask a reviewer to compare the client version with the planner's reasoning note and supporting documents, paragraph by paragraph. Read the final document from the client's perspective: can they see what they are being asked to do and why?

This is distinct from general ChatGPT use in a financial advice practice. The issue here is the SOA's evidential chain. A prompt can improve clarity, but the file must still show the planner's inquiries, research and judgement. If a generated paragraph contains a claim that cannot be traced to those materials, remove or substantiate it before issue.

Fees, benefits and conflicts need controlled wording

ASIC's guidance identifies certain remuneration, benefits and conflicts as SOA content. Set up a maintained disclosure source approved by the licensee. The drafting assistant may place that text in the right section and flag blank fields, but it should not paraphrase away an amount, relationship or qualification. Check that the disclosure relates to the actual recommendation, current arrangements and recipient. A standard block copied into every SOA can be as misleading as a missing block if the facts differ.

Suppose a planner's advice fee is an illustrative $2,400 and an ongoing service arrangement is $1,800 a year. A draft that says “Our fee is $2,400” may leave the reader unaware of the continuing cost. A draft that combines them as “$4,200” may make a one-off fee look annual or conceal the ongoing term. The reviewer should check the amounts, periods, services, consent and approved wording separately. If a benefit is contingent, describe the condition from the verified source rather than present it as certain.

A conflict example: a proposed investment is issued by a related entity. The AI may produce a generic line that the firm “may have commercial relationships”. That does not identify the relevant relationship or help the client understand its possible influence. The planner and licensee must determine the actual disclosure and whether the advice remains appropriate and gives priority to the client's interests. The drafting system can check that the approved conflict field is populated; it cannot clear the conflict.

If a tool vendor claims to generate a compliant SOA automatically, ask to see how it handles a missing conflict, a changed fee schedule and a recommendation that is withdrawn after review. Test its permissions, version history and export. The advice software vendor questions cover product evaluation; the SOA acceptance test should focus on whether a specific client's document is traceable and correct.

Keep the draft, the reasons and the approval together

ASIC says personal-advice records must be kept accessible for at least seven years. The final SOA is only one part of the evidence. Retain the fact find, inquiries, source documents, comparisons, assumptions, file notes, conflict analysis, calculations, draft corrections, approval and delivery record under the licensee's retention policy. A chat transcript may show wording changes but not the full professional judgement. The planner's own reasoning note should exist independently of the model's generated explanation.

An illustrative audit trail might read: “Client objective confirmed in meeting note dated [date]; account balance checked against statement dated [date]; fee comparison recalculated by planner; switching consequence added after review; final SOA approved by [role] on [date].” The square brackets are deliberate placeholders for a real file, not information to send to a client. A reviewer should be able to trace every material number and statement to a source without reconstructing the drafting conversation.

Record rejected AI suggestions as well as accepted changes when they reveal a risk in the workflow. If the model recommends a product outside the approved advice scope, the log should show that the suggestion was rejected and why. If it repeatedly invents a retirement age from an old note, change the input structure. For routine safe drafting, retain the controlled output and review evidence required by the licensee; do not create a sprawling archive of irrelevant prompts that makes the real decision harder to find.

The separate financial adviser meeting-note tutorial addresses one possible input. A transcript can omit context, mishear a number or confuse the client and partner. Before it reaches the SOA evidence pack, a person must verify the material facts with the client record. The SOA approval step is too late to discover that the fact find was built from an unchecked summary.

Test the workflow on awkward advice files

Start with fictional or properly sanitised files spanning several situations: straightforward retention advice, a proposed switch with lost benefits, limited-scope advice, a fee conflict, an incomplete fact find and a client whose objective changed after the first meeting. For each, write the planner's reference answer before testing the system. Score the AI draft on factual fidelity, advice-scope fidelity, arithmetic, disclosure completeness, readable reasoning and quality of its missing-information flags.

A useful rejection rule is any unsupported recommendation, invented client circumstance, omitted material switching consequence, incorrect amount or altered approved disclosure. Count those as failures even if the document looks professional. If the system drafts a good summary but cannot handle fees, restrict it to summary wording. If it can spot contradictions but produces weak client prose, use it as a pre-draft checker. Define the permitted role from observed performance rather than a vendor feature list.

Measure the whole job: preparation of the evidence pack, AI drafting, planner correction, compliance review and filing. Faster typing is not a useful gain if the review becomes longer or less reliable. Recheck the process after a change to the licensee's template, product source, law or regulator guidance. Before issue, confirm that RG 175 and the licensee's current policy still support the document pathway chosen; do not assume an old SOA example is current authority.

If the client is receiving further advice, first decide whether an SOA is required under the current rules or whether a different record route is available. ASIC describes limited exceptions, with conditions concerning the earlier SOA, changes in circumstances and the basis of advice. This classification is a professional decision and should be documented before the model is asked to draft anything. A tool should never convert “we issued an SOA last year” into “an SOA is unnecessary now”.

Questions before issuing an AI-assisted SOA

Can a planner ask AI to choose the recommended product?

Treat that as a planner-owned judgement, not a drafting task. The planner needs to establish the client's relevant circumstances, investigate and assess reasonably relevant options, consider conflicts and explain why the advice is appropriate. AI may organise the evidence and flag missing comparisons, but a generated ranking cannot demonstrate that the professional reasoning was done or that the product information is current.

Is an SOA always required for further advice?

No. ASIC describes limited circumstances in which a record of advice may be used for further advice where the client's relevant circumstances and the basis of advice have not significantly changed, among other conditions. The planner and licensee should assess the current rules and the actual file before choosing the document. AI should never infer an exception merely because an earlier SOA exists.

What is the minimum record to retain from an AI-assisted draft?

Keep the final advice, fact find, scope, research, comparison work, fee and conflict disclosures, review notes and approval in the licensee's record system. ASIC says personal-advice records must be accessible for at least seven years. The precise set will vary with the advice, but a final SOA alone may not show how the planner reached the recommendation or checked the AI output.

Further reads

• AI Governance Checklist for Small Financial Advice Firms — Set firm-wide controls for advice drafting.
• AI Agents for Financial Advisers: What They Can Do Without Advising — Keep automated admin separate from advice.
• AI for Financial Planners Australia: Advice Records and Client Notes — Plan the supporting record and client notes.
• Advice-Specific AI or ChatGPT: Which Should an Advice Firm Use? — Compare specialised drafting with a general assistant.
• How Advisers Use AI to Prepare for Annual Client Reviews — Prepare review inputs without skipping judgement.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ASIC RG 175 AFS licensing: Financial product advisers—Conduct and disclosure (issued 21 November 2024); ASIC Limited AFS licensees: Advice conduct and disclosure obligations; ASIC RG 255 Providing digital financial product advice to retail clients, paragraph 255.86; Treasury ministers' media release of 21 March 2025 on the next tranche of advice reforms. Read 4 November 2026.