Legal AI governance: who owns the answer sent to a client?

The brief

In LexisNexis’s July 2026 UK survey, 18% of respondents reported no owner for AI governance. The public report describes a sample of 543 people across legal settings, including non-lawyers. This is not evidence that 18% of all UK law firms lack an owner. Read the report and method.

The useful question for a legal team is specific: who is responsible for the answer that reaches the client? Our proposed review follows one AI-assisted communication from its underlying sources through review, approval and any later correction. Name a person who can resolve an uncertain point before release. A broad policy can set direction, but the working record should show who accepted responsibility for this particular answer.

The full record

The sample spans private practice, in-house teams, the Bar, the public sector and academia; 6% of respondents were non-lawyers. LexisNexis is a commercial legal-information provider. Its public report provides a useful respondent-level signal, not a census or independently verified audit of firms. Source: LexisNexis AI governance report.

Our analysis: assign the decision, not just the tool

A team can nominate someone to approve software while leaving responsibility for individual outputs unclear. These are different decisions. Tool approval asks whether a system may be used for a purpose. Communication approval asks whether a particular answer is accurate, appropriate and ready to leave the team.

Before reviewing an AI-assisted client letter, identify the responsible reviewer and the issues that require specialist judgement. Give the reviewer the underlying material, not only the generated draft. Record which parts were checked, which were changed and which remain uncertain. An approval should have a defined scope.

Use a question the reviewer can answer

A request to “check the AI” is too vague for a useful handover. A more workable request identifies the proposed claim, the source relied upon and the point requiring confirmation. For example, ask whether the cited document supports the stated date, rather than whether the whole letter looks plausible.

If an answer cannot be verified, give the unresolved point an owner and decide what can safely be communicated in the meantime. Do not allow a deadline to turn an unverified sentence into a confirmed statement. Record a qualified answer as qualified, including what information would change it.

Plan the correction route before release

Our proposed exercise uses a fictional or appropriately controlled communication. Introduce a source correction after the draft has been approved. Ask the team to identify which statement changed, who must review it and which recipients need a corrected version. Preserve the original and the correction with their actual dates.

The exercise has a narrow result: whether the team can locate ownership and manage a change in the tested example. It does not measure the accuracy of every AI output or establish the quality of the entire governance programme. Publish those limits alongside any later benchmark.

For formal communication, the practical unit is the attributable answer: its evidence, reviewer, recipient and correction path. That gives an ownership policy something concrete to govern.

Read the public LexisNexis report on 19 October 2026. Fieldwork was in July; the report’s public methodology describes 543 UK respondents, including non-lawyers. We did not obtain respondent-level data, full weighting details or an independent firm audit. Findings are attributed to the commercial publisher; proposed exercises are original analysis.

1. LexisNexis: AI governance report, July 2026 survey