Artificial intelligence can read a lease, identify a restriction and compare wording against a lender requirement in seconds. That does not mean it has understood the transaction, advised the client or decided whether the property will remain readily saleable and mortgageable.
The attraction is obvious. Conveyancing firms face capacity pressures, fixed fees, demanding clients and constant pressure to complete transactions more quickly. A system that can review incoming documents and highlight apparent risks offers the prospect of faster work and larger caseloads.
But the central question is not whether artificial intelligence can find words in documents. It is whether a regulated law firm can safely allow the system’s output to become the legal analysis on which a client and lender rely.
That is a much more difficult question.
Reading a document is not the same as understanding a transaction
A conveyancing file is not simply a collection of documents. It is a developing body of knowledge assembled from the title, contract, searches, replies to enquiries, mortgage offer, lender instructions, valuation, survey, correspondence and conversations with the client.
The significance of one document may depend on information found nowhere within it.
A lease restriction may matter differently to a client intending to let the property. A right of way may take on greater importance after the client mentions plans for an extension. A service charge provision may require closer examination when the survey identifies major external works. A restrictive covenant that appears familiar may become critical when the client explains how the property will be used.
An automated document review may accurately extract the relevant clause and still fail to identify the real risk. The failure is not necessarily one of reading. It is a failure of context.
There is also a danger that the system reviews the property for a notional or neutral buyer. There is no truly neutral client in conveyancing. The same legal provision may be tolerable for one buyer, unacceptable to another and decisive for a buyer with a particular intended use, financial position or timetable. Advice that is technically accurate for an abstract purchaser may therefore be inadequate for the client who is actually buying the property.
The knowledge held in conversation
Lawyers acquire material information through dialogue. Clients disclose their intentions unevenly and sometimes indirectly. A comment during a telephone call, an answer given in an email or a question raised after reviewing a plan may change the advice that should be given.
A system reviewing the title and lease may not know that the buyer:
- intends to work from the property
- plans to keep several vehicles there
- expects to extend or alter the building
- is purchasing for an elderly relative
- will depend on short term letting income
- expects to sell again within a few years
- has received information from the agent that conflicts with the legal documents
Unless all relevant information is brought together, the system is reviewing an incomplete version of the transaction. A lawyer who accepts its output without performing that synthesis risks doing the same.
This is why the Legal Ombudsman’s recent guidance on residential conveyancing places importance on the consumer’s attributes, needs and circumstances, the nature of the transaction and clear communication. The required service is not satisfied merely because the correct documents were processed.
The individual lender trap
There is another important limitation when technology reviews a problem through the requirements of the lender currently involved, just as there is when it reviews the property for a generic buyer without understanding the actual client.
A positive answer to the question “will this lender accept the property?” is not necessarily a positive answer to the question “is this property readily mortgageable and marketable?”
Lenders do not adopt identical policies. One lender may accept a short lease, unusual construction, escalating ground rent, absent freeholder, estate rentcharge or title defect that causes concern to several others. Requirements also change over time.
If a system tests the problem only against the current lender’s instruction, it may correctly conclude that no report is required. That conclusion could still be inadequate advice to the buyer. The buyer may later discover that a substantial part of the lending market takes a different view, restricting the pool of future purchasers or affecting value, remortgage options and saleability.
The conveyancer acts for the borrower as well as the lender. Compliance with one lender’s handbook answer is therefore not the outer limit of the retainer. The lawyer must still consider what the issue means for the client and the property in the wider market.
The two contextual questions must be considered together. What does the issue mean to this particular client, and how might lenders and future buyers view it? A system that answers only one of those questions has not completed the legal analysis.
This is also why historical and comparative lender information can be important. The relevant professional question is sometimes not merely what the present lender says today, but whether lender attitudes differ and whether those attitudes have been moving in a direction that could affect a future transaction.
Mortgageability and marketability are connected but different
Mortgageability asks whether a lender is willing to take the property as security on stated terms. Marketability asks a broader question about the ability to sell the property, the likely pool of buyers and the obstacles a future buyer may face.
The two overlap because many buyers require mortgage finance. A defect that is acceptable to the present lender but unacceptable to a meaningful group of other lenders can reduce the future market. The present mortgage offer does not make that wider risk disappear.
A competent report on title must therefore do more than announce that the lender is satisfied. It should explain the problem in terms the particular buyer can understand, address reasonably foreseeable practical consequences and enable the buyer to make an informed decision.
This contextual obligation is difficult to automate because it is not a simple exercise in matching a clause to a rule. It requires judgment about degree, future uncertainty, the client’s plans and the interaction between legal acceptability and commercial reality.
Who is actually doing the legal work?
There is an important distinction between a tool that assists a lawyer and a system whose conclusion is adopted as the firm’s conclusion.
If the technology extracts dates, organises documents or flags clauses for review, the lawyer remains visibly responsible for the analysis. If the system decides what is significant, determines whether enquiries are required and generates the substance of the advice, the firm must confront a harder question: who has actually performed the legal work for which the client is paying?
Describing the process as human supervised does not answer that question. Supervision must be real. A person cannot meaningfully supervise an analysis they lack the time, information or technical competence to reproduce and challenge.
The SRA warning notice on the misuse of AI is explicit that solicitors and firms remain accountable for the work and outputs produced with AI. Competence, effective supervision, confidentiality and proper governance continue to apply. The technology has no separate professional responsibility to the client.
The automation bias problem
The greatest risk may not be that a system makes an obviously absurd error. It may be that it produces a clear, plausible and professionally worded answer that discourages further thought.
Repeated exposure to apparently reliable outputs can create automation bias. Fee earners begin by using the system as a second pair of eyes. Over time, the system becomes the first pair of eyes and the human review becomes confirmation rather than scrutiny.
Capacity pressure intensifies the danger. If the commercial purpose of the technology is to enable each lawyer to carry more matters, the available time for genuinely independent review may shrink. The firm can then appear to retain human oversight while its operating model makes that oversight increasingly superficial.
This reflects a concern raised in our earlier discussion of the conveyancing integration paradox. Connecting more systems can improve speed and visibility, but it can also create new dependencies and allow a mistake to move rapidly through the entire workflow.
Our earlier article on the accountability gap and explainability considered the same principle in a different regulatory setting. A decision that cannot be explained and defended by the responsible professional remains dangerous, however sophisticated the process that generated it.
Would the firm be insured if the AI gets it wrong?
A firm should not assume either that an AI related error is automatically uninsured or that its professional indemnity policy makes adoption of the technology safe.
Solicitors’ compulsory professional indemnity arrangements are designed to provide broad protection against civil liability arising from private legal practice. Whether a particular claim is covered will depend on the policy, the minimum terms, the nature of the loss and the surrounding facts.
Even where a client’s claim falls within the firm’s professional indemnity cover, several separate exposures may remain:
- the policy excess and future premium consequences
- regulatory action that is not removed by an insurance payment
- complaints and fee refunds
- lender panel consequences
- data protection and confidentiality liabilities
- cyber losses that may fall under a different policy
- contractual limitations or inadequate indemnities from the technology provider
- reputational damage and the operational cost of reviewing affected files
The prudent course is to discuss the proposed use with the firm’s broker and insurer before deployment. The firm should accurately describe what the system does, whether it handles client data, whether it generates recommendations, how outputs are checked and whether it will be used to increase individual caseloads.
A claim may be brought against the law firm because the client retained the firm, not the algorithm. Any right of recovery against a supplier may be limited by contract and may bear little relationship to the client’s loss.
What do the client care documents say?
Terms of engagement drafted before substantive AI use may not explain how the service is now delivered.
Firms should decide whether their client care documentation adequately addresses:
- the use of automated or AI assisted systems in providing the service
- the purposes for which client information may be processed
- the safeguards protecting confidentiality and legal professional privilege
- the extent and nature of human review
- responsibility for the final advice
- any material limitations affecting the work
Disclosure should not be used as an attempt to transfer professional responsibility to the client. A clause saying that AI may be used does not excuse inaccurate advice, inadequate supervision or a failure to understand the client’s circumstances.
The firm should also consider whether its description of the service and fees remains fair. If the client believes an experienced conveyancer will personally review the material, but the substantive analysis is generated automatically and checked only briefly, there may be a gap between the service presented and the service delivered.
Confidentiality is a design question
Property documents contain names, addresses, financial information, signatures and information about clients’ personal plans. Firms must understand where that information goes, who can access it, how long it is retained and whether it is used to train or improve any model.
These questions cannot be left to individual fee earners. The firm needs an approved system, documented procurement and due diligence, contractual safeguards, access controls, retention rules and a clear prohibition on staff using unapproved tools.
The fact that a system is convenient or integrated into an established workflow does not by itself answer the confidentiality, privilege or data protection questions.
A governance framework before the button is pressed
Before using AI for substantive conveyancing review, a firm should be able to answer the following questions:
- Purpose: Is the system extracting information, prioritising review or making legal judgments?
- Scope: Which documents, transaction types and risk areas may it review?
- Context: How are client conversations, objectives, surveys, valuations and other file information brought into the analysis?
- Lender divergence: Does the process consider only the current lender or the wider mortgageability and marketability risk?
- Verification: What must the lawyer check independently and where is that check recorded?
- Escalation: Which findings require review by a suitably experienced conveyancer?
- Competence: Have staff been trained to recognise both incorrect outputs and significant omissions?
- Auditability: Can the firm show what the system reviewed, what it produced and why the lawyer accepted or rejected it?
- Confidentiality: Are contractual and technical protections adequate for the information processed?
- Insurance: Have the broker and insurer been given an accurate account of the intended use?
- Client information: Do the terms of engagement fairly describe the service?
- Capacity: Is efficiency being used to improve quality or merely to increase the number of files carried by each lawyer?
Efficiency is valuable, but quality is the purpose
This is not an argument against artificial intelligence. Used properly, technology can identify information earlier, reduce repetitive work and give lawyers more time for advice and communication.
The danger arises when time saved by automation is immediately converted into additional caseload, leaving no professional dividend for the client. If every recovered minute is filled with another file, the capacity problem has not been solved. It has merely been concealed behind a faster process.
More than a decade ago, we warned about the race to be the cheapest conveyancer. The modern equivalent may be a race to process the largest number of transactions with the smallest amount of human attention.
Clients do not instruct solicitors simply to have documents scanned for unusual clauses. They pay for judgment, explanation, challenge and advice about the transaction as a whole.
The question every firm must answer
When an important issue is missed, it will not be enough to say that the system did not flag it.
The regulator, insurer, lender, ombudsman and client are likely to ask a more direct question: what did the responsible lawyer know, what should they have known and how did they reach the advice that was given?
AI may help the lawyer answer that question. It cannot be allowed to become the reason the lawyer never asked it.
This article provides general risk management commentary and is not legal or insurance advice. Firms should obtain advice appropriate to their proposed technology, regulatory arrangements, insurance programme and contractual documentation.