AI: Its Uses & Limitations

Everyone is talking about AI and how it might one day remove the need for professionals.

In reality, AI is a very useful tool – particularly where tasks are systematic and good quality information is readily available. For example, a search about what to do on a holiday in Florida can generate a credible list of suggestions and activities.

However, when the subject matter becomes less clear, AI often struggles to produce the correct answer. In some cases, it even invents information and provides meaningless or fake justifications for it.

We experienced this first-hand in a recent dispute with Companies House. A standard Google search offered no help, so we turned to an AI tool. It sided with Companies House. However, after consulting a human expert, we were advised that our interpretation was correct. We fed this back to Companies House – and our submission was accepted.

There have also been several court cases where individuals have used AI to help generate a defence, only to find that it wasn’t up to the job – and they lost.

A recent tax tribunal case, Zzaman v The Commissioners for His Majesty’s Revenue and Customs [2025] UKFTT 539 (TC), highlights the dangers of relying too heavily on AI.

While Mr Zzaman ultimately lost his appeal, the judge made some important observations about AI-generated submissions:

“Mr Zzaman’s statement of case, written with the assistance of AI, did not provide grounds for allowing his appeal. Although some case citations were inaccurate, AI did not generate fictitious cases (as happened in Felicity Harber v HMRC [2023] UKFTT 1007 (TC)). However, the cases cited did not support the propositions being advanced with recognised legal authority.

This highlights the dangers of relying on AI without human checks. These tools may not have access to the right sources, may misinterpret the question, or may miss relevant material. The result can be a plausible-sounding but incorrect answer – or worse, fake authorities or irrelevant references.

There is no reliable way to prevent this, but risks can be reduced by:

• Using clear prompts
• Asking for citations from specific paragraphs
• Verifying that the tool has live internet access
• Instructing it not to guess when unsure
• Requesting details about any limitations in its response

Without these checks, there is a real risk that AI-generated material will raise false expectations and waste everyone’s time – including the court’s.”

For a closer look at how to frame questions and check what comes back, we’ve set out our approach to getting the correct answer from AI.

So while AI has a role, particularly in research, it is not a replacement for expert advice – especially when it comes to tax planning or litigation.

At Wilds, we have many years of experience advising clients on complex tax matters. We’re happy to work with AI where it adds value – but more importantly, we know when to challenge it, when to refine its suggestions, and when to rely on our professional expertise instead.

We look forward to continuing to support our clients with top quality, bespoke advice – combining the best of both technology and human insight.