A client rarely arrives with a legal question. They arrive with 150 emails, 100 WhatsApp messages, several competing accounts of events and a conviction that one document proves everything. The solicitor’s first task is not to write. It is to understand: what happened, what remains uncertain and which facts could affect the legal outcome.
Yet much of the profession’s discussion about artificial intelligence begins at the other end. We ask whether it can draft a contract, produce a witness statement or summarise authorities more quickly. These visible uses may encourage firms to automate output before improving the thinking that should precede it.
The more promising application is less theatrical. AI can assist with organising an unstructured file: identifying documents, constructing a provisional chronology, comparing accounts, locating gaps and testing whether assertions are supported by the record. That creates the possibility of something broader than faster internal working: legal diagnosis offered to the client as a defined service before implementation begins.
The illusion of generative speed
Drafting is rarely the true bottleneck in a difficult matter. Once the relevant facts, legal issues and objective are clear, an experienced solicitor can often draft efficiently. Time is lost earlier: reading duplicated correspondence, reconciling dates, identifying which version of a document is operative and discovering that an apparently central allegation has no supporting evidence.
Generative tools can make weak analysis look finished. In producing a coherent narrative, they can smooth over the very gaps and contradictions that matter most. A fluent document may conceal an incomplete chronology, an untested assumption or an authority that does not exist. Polished language can remove the natural friction that would otherwise cause a lawyer to stop and check.
That concern now has an unmistakable regulatory context. In its warning notice on misuse of AI, published on 17 August 2026, the Solicitors Regulation Authority highlights false material, fabricated citations and disclosure of confidential information. Responsibility remains with the solicitor and firm, regardless of the tool used.
The courts have made the same point more sharply. In R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank, the Divisional Court addressed false authorities placed before the court and stressed lawyers’ duties to verify submissions. AI is not a colleague to whom responsibility can be transferred, still less a defence when inaccurate material is filed.
What governed diagnostic workflows can do
Legal diagnosis is not a single automated act. The natural sequence is evidence, diagnosis, strategy and only then drafting. The file must first be made intelligible, with the factual record separated from assertion and missing information identified. Only after that can the solicitor characterise the legal problem, evaluate risk and recommend a route forward.
Within a governed diagnostic workflow, AI-assisted tools may create a document inventory, extract dates, group correspondence, compare drafts and flag apparent inconsistencies for human review. They can also expose negative space: the bank statement referred to but absent, the unexplained six-month interval, or the decision for which no attendance note has been found.
The client-facing deliverable is not merely a faster draft. It may comprise a source-linked chronology, a record distinguishing what is established from what remains uncertain, material information gaps, the issues requiring legal judgment and a bounded set of strategic options. It retains value even if the client decides to negotiate, mediate, instruct another firm or take no further action.
The workflow remains an investigative aid, not a statement of fact. Every material proposition must be traced to its source. An inconsistency may result from different terminology rather than dishonesty. Silence in the documents may be legally important, or it may simply reflect an incomplete upload. A chronology assembled by a tool remains provisional until a competent person has checked it.
The diagnostic threshold
There is a line that technology should not be permitted to cross without professional determination. A tool can locate a reference to a payment; it cannot decide, by itself, whether the payment was authorised, whether capacity was in issue, whether an explanation is credible or whether the point is proportionate to pursue. Those questions require legal knowledge, context and judgment.
The same distinction applies to relevance. A family may have argued for 20 years, but only a small part of that history may bear on the validity of a transaction. Conversely, a minor document may alter the analysis. Legal diagnosis is not automated classification; it is the solicitor-led process of deciding which verified facts matter and why. Client objectives, including speed, privacy, cost and continuing relationships, also shape that judgment. That legal judgment is also subject to regulation.
Governance before productivity
A diagnostic model is only defensible if governance comes before convenience. Firms should know where information is processed, whether prompts or documents are retained, who can access them and whether they may be used for training. Confidential material should not enter a public tool merely because it is convenient. Data protection, privilege and contractual safeguards must be addressed at system level.
The workflow also needs an audit trail identifying the source material, dataset limitations, tasks performed with technological assistance and the human checks applied. A conclusion that cannot be justified by the source record and the solicitor’s reasoning is not ready to become legal advice.
Supervision should be calibrated to risk. Extracting dates from routine correspondence differs from summarising medical evidence or assessing inconsistencies in a witness account. Higher-risk tasks require tighter controls and source-level verification. Firms should define permitted uses rather than merely telling staff to ‘use AI carefully’.
A different economic proposition
There is a persistent pricing paradox in legal practice. Clients most need advice when their affairs are at their most disorganised, yet that is when a firm has greatest difficulty defining scope and offering a predictable price. An open-ended hourly retainer passes much of that uncertainty to the client; a fixed fee agreed before the file is understood transfers it to the firm.
A stand-alone diagnostic assessment offers a possible middle course. Stage one is a bounded review of an agreed document set, producing a factual map and solicitor-led options. Stage two is implementation: correspondence, negotiation, proceedings, restructuring or referral under a separate scope. The client buys clarity before committing to substantial expenditure, while the firm prices later work from a better-informed baseline.
Not every file can be reduced to a fixed fee. Document volume, poor disclosure and emerging evidence remain real constraints. The perimeter must therefore be explicit, with assumptions and excluded material recorded. Diagnosis improves the basis for pricing; it does not abolish uncertainty.
In developing a diagnostic approach for stalled estate administrations, I have found that the apparent dispute is often wider than the blockage. Hostile correspondence may reduce to a missing valuation, uncertainty about authority, an unanswered accounts request or an outstanding procedural step. Separating factual diagnosis from adversarial positioning can prevent the correspondence itself from becoming the dispute. Legal work often expands because the file has never been reduced to a verified problem statement.
A product category, not a technology feature
Most firms sell legal labour or an eventual legal output. A diagnostic service sells an earlier result: what happened, what matters, what is missing and what should happen next. It can be scoped and valued independently of later representation.
AI has not invented legal diagnosis. Good solicitors have always imposed order on confused facts before advising. Governed technology may reduce the mechanical burden and make a discrete diagnostic stage commercially viable where an open-ended retainer might otherwise follow.
Firms should test the proposition. A controlled pilot should use an approved system, a defined category of work and a written verification process. It should measure accuracy, missed issues, professional time and client understanding—not merely drafting speed. Training must cover plausible error, confidentiality and human reasoning.
The part that remains human
Diagnosis remains valuable precisely because it ends in a professional decision. The solicitor must determine what is established, what is missing, what is legally significant and what course is proportionate for this client. AI can help organise the evidence from which that decision is made. It cannot own the conclusion.
The more significant distinction may not be between firms that use AI and firms that do not. It may be between using technology principally to accelerate legal output and using it earlier to improve how matters are understood, scoped and priced.
The opportunity is not faster prose. It is better legal orientation. Before asking what AI can draft, firms should ask whether it can help their lawyers understand the problem that deserves to be addressed—and help their clients purchase clarity about it.
Justin Patten is principal solicitor at Human Law www.human-law.co.uk
.jpg&w=256&q=75)
.png&w=3840&q=75)
Copyright & permissions
