The question tax professionals asked two years ago was "should we be using AI?" In 2026, the question is "which AI tools are actually worth using, and how do we use them without creating new risks?"
This guide answers both questions practically — without hype, and without dismissing the genuine capability that now exists.
What AI is actually good at in a tax context
Before reviewing specific tools, it helps to be clear about what large language models (LLMs) — the technology underlying most AI tools — are and are not good at.
LLMs are good at:
- Summarising large volumes of text quickly and accurately
- Drafting structured documents from source material
- Identifying patterns in unstructured text
- Explaining complex topics in plain language
- Research assistance — finding relevant cases, guidance, and commentary
LLMs are not good at:
- Performing arithmetic reliably without tool assistance
- Applying law to specific facts with the precision of legal advice
- Knowing what has changed after their training cutoff
- Being consistently right on edge cases in complex regulatory regimes
The implication for tax work: AI is an excellent research and drafting assistant. It is not a replacement for professional judgment on specific positions.
The tools worth knowing about
General-purpose LLMs
Claude (Anthropic) and ChatGPT (OpenAI) are the most widely used general-purpose AI assistants. Both are capable of summarising HMRC guidance, drafting correspondence, explaining IFRIC 23, and assisting with research. For tax professionals, the key difference is:
- Claude is generally regarded as more precise on technical content and less prone to confident confabulation
- ChatGPT has a larger ecosystem of integrations and plugins, including tools that can access live data
For most day-to-day research and drafting tasks, either is capable. The more important variable is the quality of the prompt you provide.
Practical use cases for tax professionals:
- "Summarise the key requirements of HMRC's UTT notification regime, citing the relevant legislation"
- "Draft an internal memo explaining the IFRIC 23 treatment of [described position] to a non-technical stakeholder"
- "Identify the key differences between the UK and OECD approaches to transfer pricing documentation"
In each case, treat the output as a first draft that requires professional review — not a final product.
Specialist tax AI tools
The market for specialist tax AI tools is developing rapidly. The most credible category is document review and extraction — tools that can ingest large volumes of tax documents (contracts, workpapers, ERP outputs) and identify relevant clauses, positions, or data points.
These tools are particularly useful for:
- Due diligence review of large contract populations
- Identifying potentially uncertain tax positions in workpapers
- Extracting structured data from PDFs and unstructured financial documents
The quality of these tools varies significantly. The key questions to ask are: (1) how are hallucinations identified and managed? (2) what human review is built into the workflow? (3) can the tool cite its sources?
Microsoft Copilot in Excel and Word
For many tax professionals, the most practically useful AI integration is already in tools they use daily. Microsoft Copilot in Excel can assist with formula construction, data analysis, and summarising large datasets. In Word, it can draft and restructure documents from prompts.
For teams whose primary workflow is Excel-based, this is worth exploring — the barrier to adoption is low and the productivity gains on routine tasks (reformatting data, drafting standard correspondence) can be significant.
The risks you need to manage
Confidentiality
The most significant practical risk for most tax professionals is data confidentiality. Inputting client data or commercially sensitive information into a public AI tool — even for analysis — can breach confidentiality obligations and data protection requirements.
Rule of thumb: never input specific client names, financial figures, or identifiable information into a general-purpose AI tool unless you have verified that the tool does not use your inputs for model training and have appropriate data processing agreements in place.
Enterprise versions of major AI tools (Claude Enterprise, ChatGPT Teams) typically offer stronger data protections. Even then, check the terms.
Hallucination
AI tools sometimes produce plausible-sounding but incorrect information — a phenomenon called hallucination. This is particularly dangerous in a tax context, where an incorrect citation to a tribunal case or an inaccurate statement of legislation could have real consequences.
Mitigation: always verify specific legal/legislative claims against primary sources. Use AI for research assistance and drafting; verify with authoritative sources before relying on the output.
Over-reliance and skill atrophy
Less discussed but worth naming: if AI tools make routine research and drafting easy, there is a risk that tax professionals become less capable at the underlying skills. This matters both for quality control (you need to be able to identify when the AI is wrong) and for long-term professional development.
The right posture is "AI as a capable junior colleague who sometimes makes things up" — useful for speeding up work, but always requiring supervision.
SAO implications of AI tool use
Under the January 2026 AI governance standards, companies need to document how AI tools are used in tax compliance processes. This has practical SAO implications: if an AI tool was used in preparing a disclosure and the disclosure turns out to be wrong, the governance trail needs to show that appropriate human review was applied.
Document: (1) which AI tools are used, (2) for what tasks, (3) what human review is applied to AI-assisted outputs, and (4) how errors would be identified. This documentation does not need to be complex — a brief policy statement covering these points is sufficient for most organisations.
The honest summary
AI tools are genuinely useful for tax professionals in 2026. The best of them can materially speed up research, drafting, and document review. They require thoughtful use — clear prompts, appropriate data handling, and consistent human review of outputs.
The tax professionals getting the most out of these tools are not treating them as oracles. They're treating them as capable assistants that need direction, oversight, and the occasional correction.
Tax Haus publishes practical analysis on AI and UK corporate tax. Subscribe to the briefing or explore our services.
Topics