Artificial intelligence tools can support a wide range of accountancy activities, including drafting, summarisation, document review, research and client communication. However, no single AI model is suitable for every assignment, and outputs must be assessed against the reliability of the underlying sources, the sensitivity of the information involved and the level of professional judgement required.
Key development
Large language models generate responses by identifying patterns in training data and predicting text one element at a time. They do not operate as authoritative databases and may produce inaccurate, incomplete or outdated information. Accountancy firms should therefore treat AI as an assistive tool rather than as a substitute for professional review, verified technical sources or responsible judgement.
Implications for accountancy professions
Financial reporting
AI can assist with preparing first drafts of accounting papers, summarising reporting requirements, reorganising technical material and converting complex explanations into clearer language for management or clients.
Its output should not be treated as evidence that a particular accounting treatment is correct. Errors may arise from:
- outdated reporting requirements;
- incorrect assumptions about the applicable reporting framework;
- incomplete analysis of facts and circumstances;
- failure to identify relevant exceptions or disclosure requirements; and
- apparently confident conclusions that are not supported by authoritative guidance.
Technical accounting conclusions should continue to be supported by the applicable standards, interpretations, regulator publications and the firm’s approved technical resources.
Tax
AI may be useful for preliminary research, organising client information, drafting tax correspondence and identifying questions for further investigation. It can also help professionals explain tax concepts in language appropriate for non-specialist clients.
Tax work presents particular risks because rules may change frequently and vary by jurisdiction. AI-generated tax content should therefore be checked for:
- the correct country and tax regime;
- the relevant basis period or year of assessment;
- current legislation, administrative guidance and filing requirements;
- commencement and transitional provisions; and
- whether the response reflects the client’s complete facts.
AI should not be used as the sole basis for tax advice, computations, filing positions or representations to a tax authority.
Audit and assurance
AI can support audit planning and administration by summarising documents, preparing draft requests, organising issues and helping teams identify possible areas for follow-up.
However, AI-generated content does not constitute audit evidence. Audit teams remain responsible for determining whether evidence is sufficient and appropriate, evaluating management judgements and exercising professional scepticism.
There is also a risk that AI-generated summaries may omit qualifications, contradictory evidence or information contained in appendices and supporting documents. Teams should review the underlying source material rather than relying only on an AI-produced summary.
Compliance and regulatory work
AI can help monitor developments, prepare preliminary compliance checklists and summarise publicly available announcements. Models with access to recent web information may be useful for identifying new regulatory publications or industry developments.
Information obtained from the open web must still be evaluated for reliability, relevance and currency. Firms should distinguish between:
- official guidance;
- consultation documents;
- professional commentary;
- news reporting; and
- unverified online material.
A current web result should not automatically be treated as an authoritative compliance requirement.
Client communication and advisory services
General-purpose AI tools can assist with drafting emails, reports, presentation material and plain-language explanations. They may also support brainstorming and the early stages of advisory work.
The draft should be reviewed to ensure that it:
- accurately reflects the agreed scope of work;
- does not overstate the certainty of the advice;
- is consistent with the client’s facts;
- clearly distinguishes assumptions from conclusions;
- uses appropriate terminology; and
- does not contain invented sources, figures or regulatory references.
Selecting the appropriate AI tool
Different AI models have different strengths. Some are designed for speed and general drafting, while others are better suited to long-document analysis, structured reasoning or research involving recent developments.
A practical selection approach is to consider:
Nature of the task. Routine drafting may require a different tool from a complex accounting analysis or review of a lengthy agreement.
Required source base. Work involving technical conclusions should use authoritative accounting, tax or regulatory sources rather than relying solely on a general model.
Need for current information. Research into recent tax announcements, filing changes or regulatory developments may require current-source access.
Level of risk. Higher-risk matters require stronger controls, documented review and greater reliance on approved technical systems.
Confidentiality. The tool must be appropriate for the sensitivity of the client and firm information being processed.
Managing the Risk Equation
AI use in professional accountancy work creates several recurring risks:
- Fabricated references: The system may produce standards, regulatory publications, tax provisions or decisions that do not exist.
- Wrong date or jurisdiction: The answer may apply an obsolete rule or refer to a different country.
- Incomplete analysis: The response may overlook an exception, disclosure, filing condition or material fact.
- Overconfidence: The wording may appear authoritative even when the conclusion is uncertain.
- Confidentiality exposure: Client records, commercially sensitive information and internal workpapers may be disclosed if entered into an unsuitable platform.
- Automation bias: Staff may accept an efficient or polished response without adequately checking it.
These risks reflect the broader limitations of AI-generated content and reinforce the need for verification and professional judgement.
Practical issues
Information governance
Firms should establish clear rules governing what information may be entered into AI tools. Client-identifiable data, payroll information, tax records, bank details, forecasts and audit work-papers should not be uploaded unless the system has been approved for that purpose.
Approved tools
Staff may have access to multiple public and enterprise AI products. Firms should identify which tools are authorised, what types of work they may be used for and which functions require additional approval.
Verification procedures
Review procedures should address both the AI-generated output and the sources supporting it. For technical matters, staff should record the authoritative materials used to confirm the conclusion.
Staff capability
Professionals need sufficient knowledge to identify when an AI answer is incomplete or implausible. Training should cover prompt design, source evaluation, confidentiality, documentation and appropriate review.
Documentation
Where AI contributes materially to an accounting memorandum, tax analysis, audit process or advisory report, the firm may need to document:
- the purpose for which AI was used;
- the information supplied to the system;
- the sources used to verify the output;
- amendments made by the professional; and
- the final reviewer’s conclusion.
Client expectations
Engagement teams should avoid creating the impression that AI eliminates the need for professional work. Clients should understand that AI may improve efficiency, but responsibility for advice, conclusions and deliverables remains with the firm.
Recommended workflow
A controlled approach may involve two stages:
- Use general AI for preliminary work, such as brainstorming, summarising, identifying possible issues or preparing a first draft.
- Validate the result using authoritative sources, firm methodology and professional review before it is used in financial reporting, tax filings, audit documentation or client advice.
The original training material similarly distinguished between broader AI-assisted work and research grounded in verified professional sources.
Action points
Firms should consider:
- issuing a formal AI-use policy;
- maintaining a list of approved AI tools;
- defining prohibited and restricted data;
- requiring source verification for technical conclusions;
- incorporating AI review into quality-control procedures;
- training staff on model limitations and confidentiality risks;
- documenting significant AI use in professional work; and
- monitoring changes in AI functionality, regulation and professional standards.
Conclusion
AI can improve efficiency across drafting, research, document review and client communication, but its value depends on selecting the right tool and applying appropriate controls.
Professional accountants should use AI to support their work, not to replace authoritative sources, technical competence or professional judgement.