Personalisation tools can make legal AI outputs more consistent, but lawyers still need to separate standing instructions from memory and keep confidential matter details out of both.
Lawyers who work with generative AI often find themselves repeating the same instructions: write plainly, avoid speculation, ask when the brief is unclear. The point of personalisation settings in tools such as Microsoft Copilot, ChatGPT and Claude is to reduce that repetition by storing durable guidance about tone, process and context. Attorney at Work says this can make outputs more consistent and cut down on the amount of rewriting required, although it does not remove the need for careful prompting or human review.
The practical distinction is between standing instructions and memory. Standing instructions cover the rules you want the system to follow every time, such as whether to use headings, how to handle ambiguity and what level of detail to aim for. Memory, by contrast, is the background the tool keeps in mind across sessions, such as your role, practice area, jurisdiction or recurring audience. Microsoft says Copilot offers Custom Instructions and Saved Memories, while OpenAI and Anthropic similarly describe custom guidance and persistent context as part of their products’ personalisation features.
For legal work, the most useful settings are usually modest and precise. Attorney at Work suggests instructions that steer the AI towards plain English, professional tone and a willingness to challenge weak premises rather than simply echo them. It also recommends asking the system to separate decisions, open issues and action points when summarising meetings, and to provide a concise subject line and brief message when drafting email. Those are small controls, but they directly address the kind of routine work lawyers delegate most often.
The article also draws a clear line between useful context and information that should stay out of memory. Stable preferences, recurring workflow patterns and preferred terminology are good candidates for personalisation. Confidential client facts, temporary assignments, passwords and anything governed by firm policy are not. That distinction matters because memory is designed for convenience, not for storing sensitive matter-specific material. Firms that are formalising AI use are increasingly pairing user-level controls with wider governance measures, including admin settings, training and prompt libraries, according to Pylon Digital’s AI setup service and legal-sector platforms such as Sidebar AI.
For lawyers who are not sure where to begin, Attorney at Work recommends using the AI itself to help build the instructions. A good prompt is to ask the system to interview you one question at a time about role, audience, preferred format, risk tolerance and how it should behave when facts are missing. The resulting draft can then be trimmed for anything sensitive or too temporary before being tested on ordinary tasks. That approach turns personalisation into a practical exercise in workflow design rather than a vague attempt to make the model “sound like you”.
The wider lesson is that personalisation improves consistency, not judgment. A tool that remembers your practice area can still misunderstand the question; one that writes neatly can still omit a crucial fact or invent a source. For that reason, the article argues that the best standing instruction is not simply to match style, but to behave like a careful assistant: flag uncertainty, identify missing information, distinguish evidence from conclusion and leave final decisions to the lawyer.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
