Yes, AI can draft a suitability report that passes compliance review, but only when the draft is built from the client's own fact-find and the adviser's actual reasons, and a person checks every figure and claim. Drafts that pass are personal and traceable. Drafts that fail repeat generic text, misstate risk or costs, or invent reasons the adviser never gave.
The dividing line is simple to state: AI is the writer, not the reasoner. If the adviser's rationale exists before the draft (in notes, the fact-find, the research file), AI can turn it into clear, consistent prose far faster than a person typing. If the rationale doesn't exist yet and the AI is asked to supply it, the report will read smoothly and fail the first question a file checker asks: "Where does this reason come from?"
What a file checker looks for, section by section
Rules differ by market, but most suitability regimes expect a report to show that the recommendation fits the client's objectives, circumstances and risk profile, and to explain the costs, risks and disadvantages. Mapping that against what AI can do shows where the drafting help goes:
| Section | What the reviewer tests | AI can draft? | Human must supply |
|---|---|---|---|
| Client circumstances | Accurate, current, matches the fact-find | Yes, from the fact-find | Confirmation it's up to date |
| Objectives | In the client's terms, specific, prioritised | Yes, if the fact-find records them properly | Priorities and any conflicts between objectives |
| Attitude to risk and capacity for loss | Consistent with the risk questionnaire and the adviser's discussion | Partly: can summarise the results | How any mismatch was discussed and resolved |
| Recommendation | Clear, specific: product, amount, provider | Can word it | The recommendation itself |
| Why it's suitable | Each reason linked to this client's objectives and circumstances | Can write up reasons it's given | The reasons |
| Alternatives considered | Real alternatives, and why they weren't chosen | Can write them up | Which alternatives were considered and why rejected |
| Costs and charges | Correct figures, clearly explained, compared where relevant | Can lay out figures it's given | Every figure, checked against provider documents |
| Risks and disadvantages | Specific to this recommendation, not boilerplate | Yes, from the product and the client's situation | Checking they're specific, not generic |
| Next steps and ongoing service | Accurate to what was agreed | Yes | Confirmation of the agreed service |
Read down the last column. Everything there is either a judgement or a fact from a source document. That's the adviser's and paraplanner's work, and the report is only as good as it.
The inputs pack: what the AI needs before it writes a word
Most weak AI drafts are weak because of what went in. A filled-in inputs checklist for an illustrative case, a client consolidating two old workplace pensions into one plan:
INPUTS PACK - Case ref 2026-114
[x] Fact-find (dated 3 Sep 2026), including objectives in
client's words: "retire at 62", "one pot I can understand",
"not lose the guaranteed bit if there is one"
[x] Risk questionnaire result: 5 of 7 (adviser discussed;
client comfortable) + capacity for loss note
[x] Provider information: both existing plans (charges,
funds, any guarantees or protected benefits, exit fees)
[x] Research output: platform/plan comparison from the firm's
research tool, with the chosen and rejected options
[x] Adviser's reasoning note (8 bullet points, see below)
[x] Costs: the recommended plan's charges from the provider
illustration, and adviser fee as agreed
[ ] Existing plan B guarantee confirmation - STILL AWAITED
-> report must not be finalised until receivedThe unticked box is the most important line. If a pension might carry a guarantee and the provider hasn't confirmed it, no amount of good writing makes the report complete. A decent drafting routine makes the AI flag gaps like this rather than write around them.
The adviser's reasoning note: ten minutes that make the draft work
The single most useful input is a short note in the adviser's own words, written straight after the recommendation is settled. It doesn't need to be elegant. It needs to contain every reason the report will give. For the consolidation case, it might read:
- Client wants one plan she can follow before retiring at 62 (fact-find objective 2).
- Recommended plan's total charges are lower than both existing plans once policy fees are included.
- Fund range covers a risk-5 portfolio; current plan A's default fund is higher risk than her profile.
- Plan A has a small exit charge; the lower ongoing charge recovers it within about two years on these figures.
- Plan B: possible guaranteed element. Do not transfer until confirmed; if confirmed, recommend leaving plan B where it is.
- Considered consolidating into plan A instead: rejected because of its higher charges and narrow fund choice.
- Considered leaving both plans as they are: rejected because it doesn't meet objective 2.
- Client understood she'll be out of the market for a few days during the transfer and was comfortable with that.
Every line is a reason, a fact or a decision, and every line can be checked. With this in the inputs pack, the AI has nothing to invent: its job is to turn eight blunt bullet points into a clear, warm explanation for the client. Without it, the model will produce reasons anyway, drawn from what reports like this usually say, and that's exactly the generic text reviewers reject. If advisers resist writing the note, point out that it replaces a much longer conversation with the paraplanner, and that it is what they'd need to say to a file checker anyway.
A prompt that ties every reason back to the fact-find
General assistants on a business plan (ChatGPT Business, Claude Team, Microsoft 365 Copilot) can do this with a tight prompt. The key instruction is that every reason must cite where it came from:
You are drafting the "Why this is suitable" section of a
suitability report. Use ONLY the inputs pasted below.
Rules:
- Write one short paragraph per reason in the adviser's note.
- After each paragraph, add a source tag in square brackets:
[FF objective 2], [ATR result], [Research p3], [Adviser note 4].
- Do not add reasons that are not in the adviser's note.
- Link each reason to a specific client objective, in the
client's own words where possible.
- If a reason depends on information that is missing or
unconfirmed, write [MISSING: what is needed] instead.
- Plain English, second person ("you"), no marketing language.
Adviser's note: [paste]
Fact-find extract: [paste]
Risk profile: [paste]
Research summary: [paste]An illustrative piece of the output:
You told us you want "one pot I can understand" before you
retire at 62. Bringing your two plans together into a single
plan means one annual statement, one set of investments to
review and one provider to deal with. [FF objective 2]
[Adviser note 1]
The recommended plan's total yearly charge is 0.45%, lower
than your current plan A (0.95%) and plan B (0.80%), so more
of your money stays invested. [Research p3] [Adviser note 3]
Your plan B may include a guaranteed element. We will not
transfer it until the provider confirms whether it does, and
what you would give up. [MISSING: plan B guarantee confirmation]This is close to usable, and the tags make the reviewer's job fast: they can check each claim against its source. Two fixes still needed. The charge comparison mixes a total charge for the new plan with what may be only the fund charges for the old ones; the paraplanner checks the provider documents and finds plan A's 0.95% excludes a separate policy fee, which makes the comparison look closer than it is. The sentence gets corrected figures and a like-for-like basis. And the tags are removed before the client version is produced, but kept in the file copy for the reviewer. That second copy is worth its weight at file review.
The payback claim in note 4 deserves the same treatment, and the sum belongs in a spreadsheet rather than the model. Say plan A holds $48,000, its exit charge is $400, and the like-for-like charges, once the policy fee is included, are 1.05% a year against the new plan's 0.45%. The yearly saving is 0.6% × $48,000 = $288, so the exit charge is recovered in about 17 months ($400 ÷ $288 = 1.4 years). That supports "within about two years", but only at today's pot size and charges, and the report should say so rather than present a payback date as certain. Do the sum yourself and paste the result in as an input; if the model ever does it, check it.
Joint cases need one extra rule. Take an illustrative couple investing a shared sum, where one partner's questionnaire scored 3 of 7 and the other's 6. Given both results and no instruction, a draft will often average them into "you are both balanced investors", a sentence neither questionnaire supports. Add a line to the prompt: "If two clients have different risk results, state each result separately, then explain the recommendation using only the adviser's note on how the difference was discussed and resolved." If the note says nothing about it, the draft comes back with a [MISSING] tag. That is the right outcome, because it is the first question a reviewer would ask.
Five ways AI-drafted reports get sent back
- Invented or inflated objectives. The fact-find says "retire at 62"; the draft says "you want to retire early and enjoy an active lifestyle with travel". The AI filled a gap with a plausible story. Spot it by checking every objective sentence against the fact-find wording.
- Generic risk warnings. "Investments can go down as well as up" and nothing else. Reviewers want the risks of this recommendation for this client: here, losing a possible guarantee, exit charges, being out of the market during a transfer. Spot it by asking whether each risk paragraph would make sense in another client's report. If it would, it's boilerplate.
- Wrong or mismatched figures. Charges on different bases, a fee rounded, a transfer value from an out-of-date statement. Language models are unreliable with numbers; why AI gets maths wrong explains why. Every figure is checked against a source document, every time.
- Risk profile drift. The questionnaire says 5 of 7, the draft says "you are a cautious investor", and the recommended fund is adventurous. The AI took "cautious" from a stray line in the notes. Spot it by reading the risk section and the fund choice side by side.
- Alternatives that were never considered. The draft says "we also considered leaving your plans where they are and an annuity purchase". The annuity was never discussed. If a reviewer asks for the research on it, there isn't any. Only alternatives in the research file go in.
The second failure is the one worth rewriting by hand for the first few reports, so the prompt can learn from your versions. A before and after for the consolidation case:
Before (generic): The value of investments can fall as well as rise, and you may get back less than you invested. Past performance is not a guide to future returns.
After (specific): Moving plan A means paying its $400 exit charge, which the lower charges on the new plan should recover in under two years at today's pot size. Your money will be out of the market for a few days during the transfer, so any rise in that window would be missed. We are not moving plan B until its provider confirms whether it carries a guarantee, because transferring would give that guarantee up.
The standard warning can stay as a fixed line, but the reviewer is looking for the second paragraph. Once you have three or four rewritten risk sections like it, paste them into the prompt as examples of the detail you expect, and the drafts start arriving at that level.
A pre-submission check before the file goes to compliance
A short checklist the adviser or paraplanner runs on every AI-assisted report. Here it is filled in for the illustrative consolidation case:
| Check | Result |
|---|---|
| Every objective matches the fact-find wording | Yes, after removing "active lifestyle" |
| Every reason has a source tag, and the source says what the report says | Yes |
| Risk profile wording matches questionnaire and fund choice | Yes |
| Every figure checked against provider documents | Corrected plan A charge basis |
| Risks are specific to this recommendation | Added exit charge on plan A and guarantee point |
| Alternatives match the research file | Yes: "keep as is" and "consolidate into plan A" |
| No [MISSING] tags remain | No: plan B guarantee still awaited. Report held. |
| Client version has tags removed; file version keeps them | Yes |
The last-but-one row stops the report. That's the check working. The review itself should be structured rather than a straight read-through; setting up human review for AI work covers how to make it quick without making it shallow.
Some problems only show up across files. Suppose that in month two the compliance reviewer, sampling five reports, finds the sentence "This gives you confidence that your retirement plans remain on track" in four of them. Each report is fine on its own; together they look templated, which is the opposite of what a suitability report should be. The fix is a short list of phrases the prompt must never use, plus a monthly search across that month's reports for any sentence appearing in more than two. A paraplanner can run the search in a few minutes by pasting the month's "Why this is suitable" sections into one document and asking the assistant to list repeated sentences.
Illustrative numbers: a four-adviser firm's report turnaround
Consider a hypothetical firm of four advisers and one paraplanner producing about 30 new-advice reports a month. Before AI, a typical report takes the paraplanner around four hours from gathered inputs to draft, and compliance sends back roughly one in five for amendments, each costing another hour or so.
With the inputs pack and tagged prompt, suppose drafting falls to about one and a half hours including the paraplanner's checks. That's around 75 hours a month back across 30 reports, close to half the paraplanner's time. The number to watch is the send-back rate. If it rises, the drafts are faster but weaker, and the saving is being paid for in rework and risk. If it holds or falls (the tags tend to help, because reviewers can see where each claim comes from), the time saved is real.
Log why each report comes back, not only whether it does. An illustrative first month: 30 reports, 5 returned, 3 of them for charges shown on different bases, 1 for an alternative with no research behind it, and 1 for a risk section that left out the exit charge. Three returns with one cause is a pattern you can fix. A new line in the inputs pack ("charges for every plan on a total-cost basis, taken from provider documents") dealt with it, and month two had 2 returns, neither about charges.
These figures are an illustration. Measure your own: time five reports before and five after, and track the send-back rate for three months.
Advice-specific tools or a general assistant?
Advice-specific tools now draft suitability reports directly. Aveni Assist, AdvisoryAI and Automwrite, for example, all offer report drafting built around advice workflows, some with checks for missing reasoning or disclosures before sign-off. Their advantages are templates matched to advice work, integrations with meeting capture and CRMs, and audit trails designed for file review. A general assistant with a strong prompt costs less and is more flexible, but the structure, tagging and checks are yours to build and maintain.
Whichever you choose, the pass-or-fail factors are the same: inputs complete, reasons from the adviser, figures from source documents, and a named person who checked it. For the full comparison, including cost and data questions, see advice-specific AI versus ChatGPT. If paraplanning more broadly is on your list, paraplanning with AI sets out which parts of the job to automate and which to keep.
Further reads
- Can Financial Advisers Use ChatGPT? What Compliance Allows — The plan, data and record rules that sit underneath this.
- How Advisers Use AI to Prepare for Annual Client Reviews — Review reports: the same drafting approach, lower stakes.
- AI Note Takers for Financial Advisers: Compliant Meeting Records — Better meeting records mean better inputs for the report.
- AI Governance Checklist for Small Financial Advice Firms — The governance layer that AI report drafting sits inside.
- How to Check AI Content for Plain English and Readability — Make the finished report readable for the client.
- A Five-Minute Fact-Check Routine for AI Output Before It Goes Out — A five-minute fact-check routine for any AI draft.
- Best AI Tools for Mortgage Brokers at Each Stage of a Case — Eight stages of a mortgage case, the AI tool worth using at each, what it costs at list price, a real-looking example of it working, and when to skip it.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: vendor product pages for Aveni Assist, AdvisoryAI and Automwrite (September 2026). Suitability requirements differ by market; this is practical guidance, not regulatory advice. Check your regulator's rules and your compliance adviser's standards.