The draft Conveyancing Protocol mentions AI once. Ed Molyneux argues the clause that will really govern machine-written enquiries is the one that never uses the word, and that it is a test of data quality in disguise.
The Law Society’s draft Conveyancing Protocol, the first revision since 2019 and open for consultation until 14th January, mentions artificial intelligence exactly once. A new general obligation states that where AI tools are used, solicitors retain full professional responsibility for all outputs, and that AI-generated content must be reviewed and approved by a qualified solicitor before it is relied upon or sent to anyone. Sensible, and uncontroversial. But the clause that will actually bite on AI sits twenty-six pages later, and it does not use the word.
Step 15 governs additional enquiries, now defined as anything raised beyond the TA forms and the LPE1. The sanction there is not new: the 2019 protocol already said that indiscriminate use of standard additional enquiries may constitute a breach, and that the seller’s side need not deal with them. Seven years on, most seller’s conveyancers would say that sentence has been honoured mainly in the breach because there was no agreed way to tell a valid enquiry from an indiscriminate one.
The draft supplies one. Before raising an additional enquiry, the buyer’s conveyancer must consider whether it is specific to this property rather than generic; whether it is required to clarify an issue arising from the title, contract papers, replies or supporting documents; whether the information is already publicly available or obtainable from searches or the relevant authority’s online records; and whether it is being raised under an express instruction from the buyer or lender, which raising it confirms. The seller’s conveyancer is told in terms that inappropriate enquiries need not be answered, nor enquiries seeking opinion rather than fact.
You should read the two provisions together. Tools that generate enquiry lists from a contract bundle are already on the market and will be commonplace by the time this protocol is final. The draft does not forbid them. It makes the buyer’s conveyancer answerable for every question they produce, and gives the other side a four-limbed, protocol-backed reason to ignore any that fail. A firm that lets a tool raise forty enquiries, twenty-five of them generic and ten asking for planning documents already on the council’s portal, has not saved an afternoon. It has produced a protocol breach with its own name on it, and the reply will quite properly be a single line citing step 15.
So the headline has a two-part answer. The seller’s conveyancer answers the good enquiries and declines the rest, and the buyer’s conveyancer answers for all of them.
What makes an enquiry fail the test is not really a question about AI at all. Limb (c), do not ask for what is already available, is a statement about data. A generator produces generic questions for a specific reason: it could not see enough of the matter to ask a specific one. Give it the title, the lease, the searches, the TA6 and what the local authority already publishes, as one coherent set of facts rather than a pile of PDFs, and most of the generic questions evaporate, because the answers are in the material. Give it half a bundle and it will do what any trainee with half a bundle does, and ask about everything.
The draft knows this in its own way. It assumes the contract bundle will often be lodged with the marketing portal before a buyer is found, and otherwise due within five working days of instruction. It tells firms not to routinely provide or request planning or building regulations documents that sit on public registers. It wants register entries under six months old and replies refreshed after two. Each puts more of the facts in front of the buyer’s side earlier so that fewer questions need asking later; step 15 is the downstream enforcement of the same idea.
So the practical response is not to switch the tools off. Give anything that drafts enquiries the whole bundle, searches included, before it asks a single question; if it cannot take the whole bundle it should not be drafting enquiries. Run the review the AI clause requires against step 15 limb by limb, not as a general read-through. On the sale side, start declining in writing and citing the step, because a refusal never made teaches nobody anything. And ask the suppliers of your bundles, searches and forms how their data arrives: as documents to be read, or as facts that software can check against one another. The second form is what makes a machine-written enquiry specific, and it is what the protocol’s upstream provisions quietly assume.
The consultation asks twelve questions. I would add one. The draft tells a solicitor to approve AI output before relying on it, but says nothing about what an output must carry to be approvable: which document each conclusion rests on, and what was checked against what. Step 15 shows the Law Society can write that kind of test when it wants to. It is worth asking for the same discipline on the review itself.
About the author
Ed Molyneux is co-founder and CTO of Moverly and the original author of the Property Data Trust Framework (PDTF), the open standard for machine-readable property data now being adopted across the industry. Ed writes about AI, property data infrastructure, and the future of conveyancing.
The views expressed in this article are those of the author and not necessarily those of Today’s Conveyancer. This article is general information, not legal advice.
