In the second of his weekly series, Ed Molyneux gets specific about where AI excels and where it fails in conveyancing — and why understanding the difference is the key to using it safely.
Last week I introduced the “jagged frontier” of AI capability; the idea that these systems are superhuman at some tasks and bizarrely incompetent at others. This week, I want to get specific. Because if you’re going to use AI in your practice (and your competitors already are), you need to understand exactly where the frontier sits for the work you actually do.
What AI is genuinely good at
Let’s start with what works. Modern language models are remarkably capable at certain categories of task, and conveyancing involves a lot of them.
Reading and extraction: Give an AI model a 40-page lease and ask it to extract the ground rent schedule, the service charge provisions, and the forfeiture clauses. It will do this accurately, in seconds, across documents that would take a trainee twenty minutes to read carefully. It handles OCR’d scans, inconsistent formatting, and archaic legal drafting with surprising grace.
Summarisation: A title register, a set of search results, a local authority enquiry response — AI can produce clear, structured summaries that capture the material points. Not legal advice, but a reliable first pass that highlights what needs human attention.
Pattern matching at scale: Where AI really shines is processing volume. A human reviewing ten sets of title documents in a day is working hard. An AI can process a hundred in the time it takes you to make coffee. For firms handling high volumes of remortgages or new-build plots, this changes the economics fundamentally.
Translation: And I don’t mean languages. I mean translating between registers — taking a technical legal position and explaining it to a buyer in plain English, or taking a client’s anxious email and identifying the specific legal question buried in it. AI is genuinely excellent at bridging the gap between professional language and everyday understanding.
Where it goes wrong
Now the harder part. Because AI doesn’t fail the way humans fail. A tired trainee makes obvious mistakes — they skip a page, they misread a date. You can spot these in review. AI fails differently: confidently, fluently, and in ways that look completely plausible.
Hallucination: This is the well-known problem, but it’s worth being specific about what it means in practice. Ask an AI to review a title and identify restrictive covenants, and it might report a covenant that doesn’t exist — not because it’s making something up randomly, but because it’s pattern-matching against thousands of similar documents and generating what it expects to find. The output reads like a perfectly competent summary of a covenant that simply isn’t there.
False confidence on edge cases: AI models don’t say “I’m not sure.” They don’t flag that a scanned deed is partially illegible, or that a covenant’s scope is genuinely ambiguous. They pick the most probable interpretation and present it as fact. In conveyancing, where the edge cases are precisely where the risk lives, this is dangerous.
Missing what isn’t there: Perhaps the most subtle failure. AI is excellent at analysing what’s in front of it. It’s poor at recognising what’s missing. A human conveyancer reviewing a leasehold title knows to check whether the freeholder’s consent provisions have been complied with for previous alterations. They know to ask for the licence to alter even if nobody’s mentioned one. AI reads what it’s given. It doesn’t know what questions to ask about what it hasn’t been given.
No memory, no context: Each interaction starts fresh. The AI doesn’t remember that this particular freeholder has a history of unreasonable service charge demands, or that the management company changed hands last year and the new operator hasn’t registered. Your knowledge of ongoing matters, local quirks, and client history is invisible to it.
The capability question you should actually be asking
The question isn’t “should I use AI?” — that ship has sailed. It’s “how do I use it without creating professional liability?”
The answer isn’t to avoid AI, and it isn’t to trust it blindly. It’s to build systems where AI’s genuine strengths — speed, volume, extraction, translation — are combined with verification that ensures its outputs can be relied upon. Where every conclusion has a provenance trail showing what data it drew from, what reasoning steps led to the conclusion, and what legal context shaped the analysis. Where you can trace a risk flag back through the logic to the specific clause in the specific document that triggered it — and understand why it matters for this particular transaction.
That’s what I’ll cover next week: why provenance and explainability — knowing not just where an answer came from, but how it got there and what legal reasoning supports it — matter more in the age of AI than they ever have before.
About the author
Ed Molyneux is co-founder and CTO of Moverly, the property intelligence platform working with LMS and Connells Group to bring structured, verified data to property transactions. He is the architect 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.
