Introduction: Why Generative AI Fits Tax Particularly Well
Most people meet generative AI through consumer tools that write emails or summarise webpages. In tax work, the same underlying capability applies to more consequential tasks: assembling written evidence, extracting meaning from complex texts, and drafting explanations that must stand up to scrutiny.
This does not mean the technology replaces technical judgement. It means the first draft becomes faster, the search for prior knowledge becomes easier, and the effort involved in turning technical work into usable written output is reduced. That shift matters in a profession where deadlines are fixed and workload is often seasonal, yet expectations around quality and traceability keep rising.
Where Generative AI Fits in Everyday Tax Work
Readers new to tax often assume the hardest part is the calculation. In practice, much of the time pressure comes from the surrounding work: understanding what has happened, identifying what matters, documenting the reasoning, and producing a clear, reviewable record.
A simple way to see where generative AI fits is to follow a typical tax workflow:
Intake and triage: summarising source documents, emails, meeting notes, and transaction summaries so you can see what you are dealing with.
Research: extracting key passages from guidance you provide and drafting the questions that need answering.
Drafting: producing first drafts of memos, risk notes, narrative documentation, and stakeholder explanations.
Review: checking for internal consistency, rewriting into plain English, and tightening structure before a human reviewer signs off.
Knowledge reuse: finding and reusing internal precedents, templates, and prior wording so the team stays consistent.
This is where the technology feels real on a Tuesday afternoon. It reduces the time spent getting from scattered inputs to a coherent first draft, then helps you improve clarity and consistency before formal review.
How Tax and Accounting Organisations are Adopting Generative AI today
The clearest evidence on adoption in tax still comes from practitioner surveys rather than controlled academic studies. A Thomson Reuters Institute survey of tax and accounting firm professionals (fielded in January and February 2024) found that many firms are still early in adoption, with some respondents reporting experimentation with public tools, but relatively few describing systematic rollout.
In that survey, 10 percent of respondents said their firms were using generative AI at an organisation wide level, while a further 40 percent said their firms were planning or considering its use. This framing is useful because it shows high intent but uneven maturity/application.3
Where firms are using, or planning to use, generative AI, the most cited use cases were accounting and bookkeeping (84 percent), tax research (84 percent), tax return preparation (69 percent), and tax advisory services (67 percent). Document review and correspondence drafting were also commonly cited (both 62 percent).
Where Generative AI Creates Value in Tax Work
Generative AI’s contribution in tax is often misunderstood because people look for it in the wrong places. It is not primarily about automating a tax computation. Most tax functions already have software for that. The impact is more visible in the human time parts of the cycle.
Document summarisation and triage is one of the clearest examples. Tax teams routinely deal with long materials: internal policies, contract terms, board papers, consultation documents, technical updates, and enquiries from colleagues. Generative AI can condense these into a structured overview and highlight the sections that appear most relevant. Even when a professional rereads the source, the first pass saves time by reducing search effort and organising the content.
Drafting is the second major area. Tax work produces large volumes of written output, including internal memos, tax risk notes, audit responses, narrative documentation, and stakeholder communications. Generative AI can draft a coherent first version quickly, then assist with rewriting for clarity and tone, particularly when a technical conclusion needs to be explained to a non-specialist audience. The professional remains responsible for accuracy and for evidencing claims, but the ‘blank page’ problem becomes far smaller.
Knowledge retrieval is a third area, especially for larger organisations. Many tax teams have significant internal knowledge locked in past memos, templates, and prior advice. When generative AI is connected to approved internal repositories, it can provide a conversational way to find relevant precedents and extract a usable summary. This is often where beginners see the fastest benefit because they can locate prior examples and agreed wording in minutes, rather than searching folders or chasing colleagues for what the team did last time.
Finally, there is value in standardisation. Tax work often needs consistent language: consistent descriptions of the business, consistent treatment of recurring issues, consistent internal policies. Generative AI can help enforce that consistency by rewriting documents to align with agreed internal wording, which can reduce friction in audits and reviews.
A single realistic example makes the capability concrete without turning this into a guide. If you have a technical note and you want a clean, reviewable first draft, you might ask: “Rewrite this into a two page tax memo with headings, assumptions, open questions, and a short plain English summary for a finance director.” The output is not the answer. It is a structured draft that reduces writing time and makes review easier.
What Changes for the Tax Professional
The tax professional’s role does not become less important in a generative AI world. If anything, certain parts of the role become more prominent: verification, judgement, explanation, and governance.
One reason is that generative AI shifts effort away from routine drafting and towards review and quality assurance. Evidence from adjacent finance work supports this general pattern, indicating that professionals using AI enabled software can reallocate time away from routine entry towards higher value tasks such as business communication and quality assurance. The details vary by context, but the underlying mechanism is relevant: when writing and structuring becomes faster, the relative importance of checking and explaining increases.
A second reason is that generative AI can make technical work look more certain than it is. Tax often involves interpretation and risk judgement. A fluent draft can create false confidence, especially under deadline pressure, which raises the premium on professional scepticism and disciplined review.
A third reason is that tax professionals become more central to governance questions. Teams need to define what can be entered into a tool, what must never be entered, what outputs need review, and what documentation should record that AI contributed to a work product. This is less about bureaucracy and more about maintaining defensibility.
Risks That Matter in Tax, and Why They are Different
Most generative AI risk discussions start with the idea of hallucinations, where a model produces plausible text that is wrong. In tax, the issue is not just that the output can be wrong. It is that the output can sound correct, which can undermine review discipline.
Research from Harvard Business School highlights a related phenomenon in a different domain: narrative explanations can increase alignment with AI recommendations, rather than strengthening human oversight, particularly under cognitive load. The specific setting studied is not tax, but the implication is directly relevant to tax review under time pressure: narrative confidence can weaken scepticism.
Confidentiality is the second major risk. Tax data can include commercially sensitive figures, transaction details, and legally privileged communications. The risk is not only data leakage. It is also uncertainty about data retention and whether inputs are used to train models. This is why many organisations restrict use of public tools for anything beyond non confidential drafting and prefer enterprise settings with clear data controls.
Auditability is the third risk. Tax conclusions need evidence trails. If generative AI contributes to drafting a memo or an explanation, teams may need a policy on what must be retained to show how the conclusion was formed and validated. The more an output influences an external communication, the stronger the audit trail expectation tends to be.
These risks are manageable, but they are not optional considerations in tax. They are part of maintaining professional standards.
Minimum Controls for a Tax Team Using Generative AI
The risk discussion becomes more useful when translated into operating rules. A minimal playbook that many teams can adopt without heavy process change looks like this:
- Approved tools only, in controlled settings. Make it explicit which tools are permitted, and which are not.
- No client confidential or legally privileged inputs unless explicitly permitted and protected. If you cannot explain where the data goes and how it is handled, do not enter it.
- Cite sources, do not accept uncited claims. Treat any factual or technical statement as a draft until verified against authoritative material.
- Human review is mandatory for anything client facing, filing related, or position setting. Draft support is fine but sign off remains human.
- Record keeping is deliberate. Keep a simple note of where generative AI was used for material outputs, and what checking was performed, so the work stays defensible.
This is not a full governance framework. It is a baseline that reduces avoidable risk while still allowing productivity gains.
UK Context: Digital Compliance Pressure and Why it Reinforces Generative AI Adoption
Even though this article is focused on generative AI rather than analytics, the UK compliance environment matters because it shapes how tax teams invest in technology and process discipline.
HMRC publishes estimates of the UK tax gap, defined as the difference between the amount of tax expected to be paid and what is actually paid. The latest published estimate for the 2023 to 2024 tax year is 5.3 percent, or £46.8 billion, and HMRC reports it collected 94.7 percent of all tax due. These figures are one reason UK tax administration emphasises compliance effectiveness and operational efficiency.
At the same time, the operational infrastructure around tax filing is changing. The joint online filing service for company accounts and Company Tax Returns is due to close on 31 March 2026, and government guidance directs companies towards commercial software routes thereafter. The relevance is simple: the compliance environment increases reliance on software workflows and documented processes, which is exactly where drafting support and standardisation have leverage.
How to Interpret Future Implications Without Hype
It is easy to overstate what generative AI changes. A more neutral interpretation, grounded in current evidence, is often more useful.
The most immediate impact is on time and attention. By reducing the effort required to produce first drafts and summaries, generative AI lowers the cost of getting work started. This can increase throughput and improve consistency if review discipline is maintained.
Over time, it also shifts how tax professionals spend their time. As drafting becomes faster and easier, greater value lies in judgment, verification, and clear communication. These activities become more central rather than less.
Governance, therefore, becomes a core capability. Organisations that are clear on permitted use, data boundaries, and review expectations tend to reduce risk and build confidence in adoption. Professional bodies have responded by publishing introductory guidance, reflecting the profession’s emphasis on responsible use rather than uncontrolled experimentation.
Additionally, concepts from responsible AI that are widely discussed in the technology sector are becoming increasingly relevant to tax. IBM’s framing of trustworthy AI, which includes explainability, fairness, robustness, transparency, safety, and security, aligns closely with what tax teams require from tools that influence compliance and advisory work.
Conclusion: Generative AI as a Drafting Engine Inside a High Consequence Function
Generative AI is already influencing tax work because it targets a central part of the job that has always been time consuming: turning complex material into usable written outputs. Evidence from tax and professional services surveys suggests that adoption is moving through a cautious phase where experimentation and planning are common, but organisation wide deployment remains limited. The use cases that dominate are tax research support, return preparation support, drafting, and knowledge retrieval, rather than autonomous decision making.
In the UK, this sits alongside a compliance environment that is increasingly software mediated, and efficiency driven. The most grounded way to understand the change is not that AI replaces tax expertise, but that it accelerates the drafting and structuring tasks around tax expertise. The responsibility for accuracy, judgement, and defensibility remains with the professional.
References
Deloitte (2025) Changing regulations, talent challenges, and AI are among top concerns in 2025 new Deloitte Tax Transformation Trends report
https://www.deloitte.com/global/en/about/press-room/top-concerns-in-2025-new-deloitte-tax-transformation-trends-report.html
Deloitte (2025) Tax Transformation Trends 2025
https://www.deloitte.com/global/en/services/tax/research/tax-transformation-trends.html
Harvard Business School (2024) Narrative AI and the Human AI Oversight Paradox in Evaluating Early-Stage Innovations (Working Paper No. 25 001)
https://www.hbs.edu/ris/Publication%20Files/25-001_8ebbe0cb-2a19-453c-9014-1e301e8dd2fb.pdf
HM Revenue and Customs (2025) Tax gap estimated at 5.3%
https://www.gov.uk/government/news/tax-gap-estimated-at-5.3
HM Revenue and Customs (2025) Measuring tax gaps
https://www.gov.uk/government/statistics/measuring-tax-gaps
ICAEW (n.d.) Generative AI guide
https://www.icaew.com/technical/technology/artificial-intelligence/generative-ai-guide
IBM (n.d.) What is trustworthy AI?
https://www.ibm.com/think/topics/trustworthy-ai
MIT Sloan Management Review (2025) How generative AI can make accountants more productive
https://mitsloan.mit.edu/ideas-made-to-matter/how-generative-ai-can-make-accountants-more-productive
Thomson Reuters (2024) 2024 Generative AI in Professional Services
https://www.thomsonreuters.com/en/reports/2024-generative-ai-in-professional-services
Thomson Reuters Institute (2024) Generative AI in Tax Firms 2024
https://www.thomsonreuters.com/en-us/posts/wp-content/uploads/sites/20/2024/07/Generative-AI-in-Tax-Firms-2024.pdf
UK Government (2025) Closure of the service to file your company accounts and tax return
https://www.gov.uk/guidance/closure-of-the-service-to-file-your-company-accounts-and-tax-return
