No one can give you as much as they can copy for you from Claude or ChatGPT

In recent months, I have seen numerous online articles and posts claiming substantial savings in legal fees through the use of AI.

In practice, however, my observations are quite different. Unreviewed outputs generated by these systems can create significant legal and commercial exposure, particularly in commercial contracts. Independent tests conducted on different legal documents within this environment indicate that AI systems may amend and adjust documents by reference to their own opaque standards. Even dozens of carefully formulated prompts may fail to prevent outcomes that could be seriously detrimental to the client.

Perhaps more concerning is the way AI-generated legal work is sometimes validated. Users who are not sufficiently aware of the limitations of these systems may ask the same AI system whether its own revisions and recommendations are correct. This is inherently unreliable as a quality-control mechanism. AI generators may describe or rate their own work as “10 out of 10”, while the legal product delivered in a particular case may be inadequately tailored to the client's circumstances and, in some cases, potentially harmful.

There is a further issue. Examples of erroneous AI-generated legal outputs are sometimes described by the AI itself as being of a “top law firm” standard, occasionally with references to specific, well-known law firms. Such descriptions may have no proper basis. They can create a misleading association between an AI-generated document and the trademarks, reputation, professional standards and know-how of third parties who had no involvement in producing or reviewing it.

The real question is therefore not simply how much AI can reduce legal fees. It is how much an apparently inexpensive AI-generated legal product may ultimately cost if nobody competent reviews what the system has produced.