In the third of his weekly series, Ed Molyneux explains why provenance – knowing where data came from and how conclusions were reached – is the difference between AI that helps your practice and AI that threatens it.
Last week I wrote about where AI excels and where it fails in conveyancing. The response I got most – from conveyancers, insurers and regulators – was a version of the same question: “So how do I know I can trust what it tells me?”
That question is the right one. And the answer is provenance.
What provenance actually means
Provenance in this context means three things. First, data lineage: where did this information come from? Was it extracted from an official search result, entered by the seller on a TA6 form, pulled from the title register, or generated by an AI model?
Second, reasoning transparency: what logic led from that data to this conclusion? If a system flags a short lease as a critical risk, can you trace the reasoning back through the specific lease term, the specific threshold that triggered it, and the specific regulatory or lending context that makes it material?
Third, evidence completeness: what information was available, what was missing, and does the conclusion account for gaps?
Without all three, you have a black box. And black boxes are not insurable.
The professional liability problem
This is not theoretical. Your professional indemnity insurer needs to understand how you reached a conclusion. “I reviewed the title register and identified…” is defensible. “The AI told me there were no restrictive covenants” is not.
The Law Society’s guidance on AI use in legal practice makes this explicit: the solicitor remains responsible for the advice, regardless of what tools were used to generate it. Which means any AI system you rely on needs to produce outputs you can stand behind professionally – outputs where you can point to the source data, follow the reasoning, and identify any gaps.
Most AI tools in legal practice today cannot do this. They generate plausible text from statistical patterns. They cannot tell you which specific clause in which specific document triggered a particular conclusion, because that is not how they work. They predict what a good answer looks like. They do not construct one from verified evidence.
Why this matters beyond your firm
The Department for Business and Trade published a significant report earlier this year on Smart Data governance – the framework for how data should be shared across regulated sectors of the UK economy. The report identifies provenance as the critical enabler of trusted data sharing, and makes a point that I think conveyancers should pay attention to.
The report distinguishes three levels of interoperability. Data formats – getting systems to understand each other’s structures – are largely solvable by the market. AI and modern software tools can translate between formats. Identifiers – correlating data about the same entity across systems – benefit from central coordination. But trust and provenance, the report argues, require government infrastructure. Only a central authority can provide the cross-sector trust that allows data from one context to be relied upon in another.
Property transactions are a perfect example. A single conveyancing matter draws on data from land registration, local authority searches, environmental agencies, energy performance assessments, lender requirements, and seller disclosures. Each comes from a different source with different reliability. Some is verified by official bodies, some is self-declared, some is inferred. Without provenance, an AI agent processing all of this has no basis for weighting one source over another. It treats a seller’s assertion about boundary ownership the same as a Land Registry title plan.
What good provenance looks like
A provenance-aware system works differently. Every piece of data carries metadata showing its source, when it was provided, and under what verification framework. When the system analyses risk, it constructs its conclusions from that verified evidence and records the chain. The output is not ‘there is a short lease risk’. The output is ‘the lease term is 125 years from 1969, giving 68 years remaining. This falls below the 80-year threshold that most lenders require. Source: title register, verified by HMLR, retrieved 14 March 2026’.
When the answer is uncertain, a provenance-aware system says so. It distinguishes between ‘we have evidence and it indicates a risk’ and ‘we do not have sufficient evidence to assess this’. That distinction – which AI on its own routinely fails to make – is the difference between a system that helps you and one that exposes you.
And when information changes, the provenance chain updates. A seller provides additional documentation, a search result arrives, and the system re-evaluates against the new evidence and shows you what changed, what triggered the change, and whether any previous conclusions need revisiting.
The shift that is coming
The DBT Smart Data report recommends that government establish the trust infrastructure that all sectors build upon. In property, that means verified, machine-readable data with clear provenance flowing between participants – not PDFs emailed between firms and manually re-keyed into case management systems.
This is already happening. The Property Data Trust Framework, and live implementations of those standards like National Property Transaction Network, are already carrying thousands of transactions’ worth of provenanced data between participants. Deterministic risk engines can evaluate that data against hundreds of checks and produce conclusions with full audit trails. The infrastructure for provenance-backed intelligence exists today.
What changes for your practice is not that AI replaces your judgment. It is that AI – properly built, with genuine provenance – gives you a verified foundation to exercise that judgment from. Not a second opinion from a black box. A transparent evidence base you can interrogate, challenge, and rely upon.
Next week: why property needs a shared system of record, and what that means for how data flows between participants in the chain.
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.
