An analytical commentary published online addresses a theme of growing relevance for legal practice: as large language models – LLMs – become more widely adopted, the lawyer’s role in verifying and validating AI output does not diminish, but becomes more central and more demanding. The quality of legal services depends increasingly on the professional’s capacity to critically oversee everything produced by AI, whether generated within the firm or by the client.
Why professional validation is indispensable
Analysis of outputs produced by language models on complex legal matters reveals an approximate distribution across three categories: around 20% of generated content is correct, around 60% is understandable in its logic but not what a lawyer would professionally recommend, and the remaining 20% is incorrect or without legal foundation. This distribution makes the lawyer’s verification not an optional step but a necessary condition for AI output to translate into reliable legal advice. Without this oversight, the risk is not only inefficiency – it is providing the client with legally incorrect guidance.
The principle that emerges is clear: the output of a language model must always be reviewed by the person best positioned to confirm its accuracy. In legal matters, that person is the lawyer. Not because AI cannot be useful – it can be, in many specific contexts – but because no one else has the expertise to distinguish what is correct from what is plausible but wrong.
The new complexity in the client relationship
The widespread adoption of AI tools has also changed the way clients approach their lawyers. Increasingly, clients do not simply describe their problem but arrive with an analysis or strategy already generated by an AI tool, asking the professional to provide an opinion on that output. The result is often a significant increase in workload.
The issue is not the client’s use of AI, but its improper use: when the client asks AI to suggest a legal strategy – a task for which the client is not best positioned to assess its correctness – efficiency is lost and the risk of error increases.
What AI can do and what remains the lawyer's domain
Language models produce useful results in specific tasks: summarising large volumes of documents, analysing the tone of communications, answering factual questions within defined datasets, creating visual representations of data, and drafting sections of technical documents based on high-quality source material. In these activities, AI can contribute meaningfully to the firm’s efficiency, provided that the output is always subject to review by the competent professional.
The functional distinction is clear: clients can usefully employ AI to organise and summarise factual elements they know directly, such as reconstructing a timeline of events or mapping each party’s position. Legal assessment and strategy remain the exclusive domain of the lawyer, who is the only party capable of verifying their correctness.
