AI hallucinations in suitability reports occur when artificial intelligence produces incorrect, invented or unsupported information that appears credible. For financial advisers and paraplanners, the risk extends beyond obvious factual errors. AI can misstate pension values, misinterpret client objectives, introduce unsupported reasons for recommendations or alter important disclosures. The safest approach is to distinguish between information that should be extracted from verified documents, figures that must be calculated, wording that should remain fixed and explanations that AI can help generate. Human review remains essential throughout.
A suitability report is not simply a document that needs to read well. It needs to explain the advice accurately and demonstrate why the recommendation is appropriate for the client.
That makes AI-generated errors particularly concerning. A paragraph can sound professional, refer to the correct investment product and contain convincing technical language while still misrepresenting the client’s circumstances.
For UK financial advice firms, the objective should not be to eliminate AI from report writing. It should be to use AI selectively, with appropriate controls over the information it handles and the statements it produces.
1. How AI Errors Can Look Polished, Personalised and Convincing
AI hallucinations are not always obvious inventions. In financial advice documentation, they can involve small inaccuracies, unsupported assumptions or subtle changes in meaning. A generated paragraph may correctly identify the client’s circumstances but introduce a reason for the recommendation that the adviser never provided. Because the surrounding language appears professional, the error can be difficult to detect during a routine review.
Consider an example involving a client who is 62 and planning to retire within 3 years.
Their fact find records a pension valued at £425,000, annual expenditure of approximately £32,000 and a preference for maintaining flexibility around retirement income.
The adviser has recommended reviewing the client’s existing pension arrangements and considering an appropriate retirement income strategy.
A general-purpose AI tool might produce the following explanation:
You have sufficient pension savings to support your planned retirement expenditure, and the recommended strategy will provide the flexibility and security you need throughout retirement.
At first glance, the paragraph sounds reassuring.
However, it makes several claims that may not be supported by the information supplied.
The client’s pension value and estimated expenditure alone do not establish that their retirement income is sustainable. The explanation also assumes that the proposed strategy provides sufficient security, without accounting for investment risk, inflation, tax, life expectancy or other relevant financial circumstances.
The AI has produced a confident conclusion where further analysis may be necessary.
Why hallucinations happen
Large language models generate language based on statistical patterns and the context they receive. They are not inherently designed to distinguish between information that has been verified and information that is merely plausible.
When a prompt contains incomplete details, an AI model may attempt to produce a coherent answer rather than explicitly identifying what is missing.
This creates several potential problems:
- Invented Financial Details: Figures, Product Features Or Client Circumstances That Were Never Provided.
- Unsupported Conclusions: Statements About Suitability, Affordability Or Risk That Lack Evidence.
- Misinterpreted Source Information: Details That Are Taken Out Of Context Or Attributed To The Wrong Document.
- Outdated Technical Information: Explanations Based On Rules Or Product Information That May No Longer Apply.
- Altered Regulatory Wording: Disclosures Or Qualifications That Have Been Rewritten Without Approval.
- Missing Qualifications: Important Risks Or Exceptions That Are Omitted From An Otherwise Convincing Explanation.
Importantly, not every AI error is technically a hallucination. Some arise from incorrect extraction, faulty calculations or unsuitable instructions.
For a suitability report reviewer, the distinction may be less important than the outcome: the final document contains information that cannot be properly supported.
2. Anatomy of a Paragraph: 5 Sentences, 5 Different Types of Control
Not every sentence in a suitability report should be handled in the same way. Some statements should come directly from the client record, others depend on verified calculations, and certain disclosures should use approved wording. AI can help explain an adviser-approved recommendation, but it should not independently invent the reasoning. Treating every sentence as freely generated text introduces unnecessary risk.
To understand why this matters, consider a fictional client who has received a recommendation to consolidate selected pension arrangements.
A draft suitability report might contain the following paragraph:
You currently hold pension arrangements with a combined value of £285,000. The recommended arrangement has an annual charge of 0.75%, equivalent to approximately £2,137.50 based on that value. Consolidation will make your retirement savings easier to manage. The proposed investments remain subject to market fluctuations, and their value can fall as well as rise. Based on your circumstances and objectives, we believe the recommendation is appropriate for you.
The paragraph contains 5 sentences, but they perform 5 different functions.
| Sentence | Information type | Appropriate control |
| Combined pension value of £285,000 | Client-specific factual information | Extract from verified pension documents and check the valuation date |
| Annual charge of 0.75% and £2,137.50 illustration | Financial calculation | Confirm the charging basis and calculate using an approved method |
| Consolidation will simplify retirement savings | Recommendation-specific explanation | Use the adviser’s recorded rationale and confirm the claim is supported |
| Investments can fall as well as rise | Risk disclosure | Use relevant, approved wording and ensure the explanation is appropriate |
| Recommendation is appropriate for the client | Professional suitability conclusion | Require adviser-supported reasoning and appropriate review |
The figures and statements in this example are illustrative, not a client recommendation.
The calculation of £2,137.50 represents 0.75% of £285,000 for 1 year, assuming the entire amount is subject to that annual rate. An actual charge assessment may be more complicated, particularly where different charging tiers, underlying fund costs or other fees apply.
That distinction demonstrates why simply instructing an AI model to write the paragraph is not enough.
Why the difference matters
The pension value should not be creatively rewritten. It needs to match the relevant source information.
The charge should not be estimated by a language model when an established calculation can provide a reliable result.
The statement about consolidation should come from the adviser’s recommendation rather than a generic assumption that combining pensions is always beneficial.
The risk disclosure should meet the firm’s approved requirements and reflect the specific investments involved.
Finally, the suitability conclusion should be supported by the client’s circumstances and the professional assessment.
Each element needs a different form of control.
A well-designed AI reporting workflow should recognise these distinctions rather than treating every part of the report as interchangeable text.
3. What to Extract, What to Calculate, What to Fix and What to Generate
A more reliable approach to AI suitability reports is to divide the content into 4 categories: extracted information, calculated information, fixed wording and generated explanations. This allows you to introduce AI where language generation is genuinely useful while preserving greater control over figures, factual details and mandatory statements. The result should be a report that is easier to review and support with evidence.
1. Extract verified client information
Client-specific facts should come from reliable documents rather than AI assumptions.
These sources might include fact finds, pension statements, meeting notes, investment valuations and product documentation.
For example, if a client has 3 pension arrangements, your report may need to identify their providers, values, charging structures and relevant features.
An AI-assisted system can help locate and organise this information.
However, extraction should not be treated as automatic verification.
A pension statement could be outdated, a meeting transcript might contain a transcription error, or the same figure could appear differently across documents.
Where information conflicts, the discrepancy should be identified and resolved before the report is finalised.
For important data, your review process should make it possible to answer a straightforward question: where did this figure come from?
2. Calculate financial figures using reliable methods
AI language models should not be relied upon as the sole mechanism for financial calculations.
They may produce correct arithmetic, but they can also make mistakes or apply incorrect assumptions.
Suitability reports can contain calculations involving charges, withdrawals, investment allocations, income requirements and other financial considerations.
These should use appropriate formulas, calculation engines or verified external outputs.
Imagine a client has £300,000 invested and an adviser wants to illustrate the effect of an annual charge of 1%.
The straightforward annual charge calculation is £3,000, assuming a constant £300,000 value and a flat 1% charge over the period.
The arithmetic is simple, but a real illustration may require additional assumptions about how and when charges are applied, investment performance and whether other fees are included.
AI can help explain the result in accessible language. The underlying calculation and assumptions should be independently reliable.
3. Keep approved wording fixed
Some sections of a suitability report should remain consistent across the firm.
These may include particular risk warnings, regulatory disclosures, standard explanations and approved descriptions of the firm’s services.
Where exact wording is required, allowing a generative AI model to rewrite it introduces unnecessary uncertainty.
For example, a model asked to make a disclosure more concise might remove an important qualification.
The revised paragraph could sound clearer while providing a less complete explanation of the relevant risk.
A more controlled process uses approved content blocks or template sections that are protected from unauthorised changes.
This also makes it easier to update wording consistently when your firm changes its reporting standards.
4. Generate explanations where professional judgement has already been established
AI can add value when helping an adviser or paraplanner explain information that has already been assessed.
For example, the adviser may have recorded that a proposed investment allocation reflects the client’s agreed attitude to risk, investment timeframe and need for access to capital.
An AI-assisted drafting tool could help turn that reasoning into a clear explanation.
However, the software should not independently decide that the recommendation is suitable.
The distinction is important.
Generating language to communicate a professional conclusion is different from generating the conclusion itself.
Purpose-built suitability report software can help firms organise these activities within structured templates rather than relying entirely on free-form prompts.
Ammonite’s Planbot, for example, allows financial advice firms to use their existing document templates, source files, formatting and approved wording to support report production.
Its documents remain editable, allowing advisers and paraplanners to check and refine the output before final approval.
4. Keeping Mandatory Wording and Approved Disclosures Out of AI’s Reach
Mandatory wording and firm-approved disclosures should not be freely rewritten by AI. Where specific content is required, your reporting process should preserve the approved version and apply it in the correct context. This reduces the risk of important qualifications being omitted or unintentionally changed. The wording should still be reviewed to ensure it remains relevant and appropriate for the client.
This matters because financial advice documents often contain language that has been carefully considered by the firm’s compliance team.
It may have been reviewed to ensure that the explanation is accurate, balanced and consistent with regulatory requirements.
When an AI model rewrites this information, it can introduce subtle changes.
For example, consider the following illustrative risk warning:
The value of your investments can fall as well as rise, and you may get back less than you invest.
An AI assistant might shorten this to:
Your investments may fluctuate in value over time.
Although the revised statement is not necessarily false, it removes the explicit explanation that the client could receive less than the amount invested.
For a particular recommendation, that difference could be material.
The example does not mean every firm must use identical wording. It demonstrates why approved language should not be altered without considering the effect on its meaning.
What does the FCA require suitability reports to explain?
The relevant FCA requirements depend on the investment, service and type of recommendation.
For firms subject to COBS 9.4, a suitability report must, among other things, explain the client’s demands and needs, why the recommended transaction is suitable and any possible disadvantages.
The required level of detail should also reflect the complexity of the transaction and the type of client.
These obligations are set out in the FCA Handbook’s suitability report rules.
Other suitability requirements can apply, including those covering MiFID business, insurance-based investment products and pension transfers.
Your firm therefore needs to ensure its reporting templates reflect the actual advice being given.
A generic AI explanation of investment risk may be useful as a drafting starting point, but it cannot establish which disclosures are required for a particular recommendation.
How to manage approved wording centrally
A practical approach is to establish a controlled library of approved content within your reporting process.
This might include:
- Standard Risk Warnings: Approved Language For Relevant Investment And Product Risks.
- Fee And Charge Explanations: Consistent Descriptions Of The Firm’s Charging Arrangements.
- Product Disclosures: Wording That Reflects Relevant Product Features And Limitations.
- Service Descriptions: Approved Explanations Of Ongoing Advice And Review Arrangements.
- Regulatory Statements: Required Content That Must Be Included In The Appropriate Circumstances.
- Client Communication Standards: Consistent Terminology And Explanations Across Reports.
The firm should also determine who can amend this material and how revisions are approved.
This is especially useful when several paraplanners work with the same templates.
Rather than relying on each employee to maintain the latest wording, approved changes can be introduced through a central process.
The objective is consistency without losing the ability to tailor explanations to individual clients.
5. Building Checks That Help You Evidence Where Every Figure Came From
A suitability report should allow important financial figures and factual statements to be traced back to reliable supporting information. This means recording the source, relevant date, calculation method and any assumptions used. AI can assist with organising and presenting information, but your firm needs checks that identify missing evidence, conflicting values and unsupported conclusions before the document reaches the client.
A polished report is only part of the evidence of suitable advice.
The underlying file should also support the information and reasoning used to produce it.
For example, suppose your report states that a client’s existing pension has an annual charge of 1.15%.
A reviewer should be able to establish whether that figure came from a current provider statement, charging schedule or another reliable source.
They should also understand what the percentage includes.
Does it represent the product charge, investment charge, adviser charge or a combination?
Without that information, even an accurately copied percentage could be misleading.
A practical verification process for AI-assisted suitability reports
You can introduce a straightforward series of checks before reports are approved.
Check the original source. Establish whether significant financial details can be matched to the relevant supporting documents.
Check the date. Confirm that valuations, tax information and product details are sufficiently current for the recommendation.
Check the calculation. Review formulas, inputs, assumptions and any externally produced illustrations.
Check the recommendation. Ensure that the explanation reflects the adviser’s recorded reasoning rather than a generic justification produced by AI.
Check the disclosures. Confirm that required risks, limitations, charges and other information are included and appropriately explained.
Check the final document. Make sure subsequent edits have not introduced inconsistencies between different sections.
These checks should form part of the firm’s normal review and sign-off process.
They need not require every sentence to have a visible reference in the client-facing report. The important point is that material claims can be supported by the underlying advice record.
What happens when source documents disagree?
Conflicting information is a particularly useful test of an AI reporting workflow.
Imagine that a fact find records a pension value of £210,000 while a provider statement shows £198,000.
The difference could arise because the records were prepared on different dates, contributions have changed the value or the original information was incorrect.
A general-purpose AI tool may select one figure without making the discrepancy clear.
A controlled process should instead flag the conflicting values and require the appropriate person to establish which figure is relevant.
The same principle applies to discrepancies in income, expenditure, risk assessments and stated client objectives.
When information is missing or contradictory, the correct response is not to make the document sound complete.
It is to resolve the issue.
6. Why Human Review Remains Essential When Using AI for Paraplanning
AI can help prepare suitability reports, but it cannot remove the firm’s responsibility for the advice and documentation provided to clients. Human review is necessary to establish that the information is accurate, the recommendation is properly supported and the report communicates the relevant benefits and disadvantages. The level of review should reflect the complexity of the advice and the risks involved.
The role of a paraplanner involves considerably more than drafting.
An experienced paraplanner may identify inconsistencies in a fact find, challenge assumptions in a recommendation or recognise when further information is required.
Those activities depend on technical knowledge and professional judgement.
AI may help highlight potential issues, but it should not be assumed to recognise every problem.
For example, a generated report could correctly describe a client’s risk tolerance while failing to explain why the proposed investment is appropriate for their capacity for loss.
The paraplanner reviewing that document needs to recognise the distinction.
Similarly, a report might accurately reproduce the charges associated with a recommended product but fail to explain the implications of replacing an existing arrangement.
Neither problem is resolved simply by producing more polished language.
The value of paraplanning software should therefore be measured partly by how effectively it supports the professional review process.
Reducing repetitive document preparation can give paraplanners more time to focus on the circumstances that require closer examination.
AI should support the review process, not replace it
One practical approach is to separate document preparation from document approval.
Software can help collect information, populate appropriate templates, structure explanations and apply formatting.
The paraplanner or adviser can then examine the content, verify important information and identify any sections that require changes.
The final approval should follow the firm’s established procedures.
This approach recognises that AI can improve efficiency without becoming the authority responsible for the recommendation.
7. Can AI Ever Produce an Accurate Suitability Report?
Yes. AI-assisted software can contribute to an accurate suitability report when it operates within a suitable process using reliable information, appropriate templates, controlled calculations and professional review. However, no AI-generated report should be assumed accurate simply because the language is convincing. Accuracy needs to be established through checks against the client file, supporting calculations and the adviser’s recommendation.
The distinction is between AI-assisted document production and unrestricted AI generation.
With unrestricted generation, the model may be expected to interpret information, decide what matters and produce the full explanation with limited oversight.
With a controlled process, the system is used for defined tasks while important information and professional decisions remain subject to verification.
That is a more appropriate starting point for financial advice documentation.
8. A Better Way to Use AI in Suitability Report Writing
Hallucinations are a genuine consideration when introducing generative AI into financial advice documentation.
However, the solution is not necessarily to avoid the technology.
The more practical response is to be selective about what AI is permitted to do.
Client facts should come from reliable records. Calculations should use appropriate methods. Approved disclosures should remain controlled. Generated explanations should reflect decisions already made by qualified professionals.
A reporting process built around those distinctions can help reduce the opportunity for unsupported information to enter the final document.
It also gives advisers and paraplanners a clearer basis for reviewing the work.
Rather than trying to identify every possible hallucination in a completely generated report, your firm can establish controls at the points where different types of information enter the document.
This is particularly important as firms increase their use of AI across suitability reports, annual reviews and other client communications.
9. Explore More Controlled Suitability Report Production With Ammonite AI
If your firm wants to introduce AI without giving up control over report wording, supporting information and professional review, purpose-built document software may offer a more practical approach than relying entirely on general-purpose chatbots.
Ammonite AI has developed Planbot to help financial advisers and paraplanners produce structured documents using their existing templates and supporting information.
The platform can work with PDFs, Word documents, spreadsheets, CSV files and meeting transcripts, while allowing your team to review and edit reports before they are finalised.
Planbot is designed to reduce repetitive document preparation without replacing the professional judgement required to assess suitability and approve recommendations.
Book a Planbot demo to explore how a structured, AI-supported reporting process could help your team prepare suitability reports more efficiently while retaining control over the finished documents.
