Did you know that you can give standing orders to many AIs? Once you provide these orders and tell them to remember them, many AIs will remember the orders and always follow them.
I use standing orders to decrease (but not eliminate) the risk of hallucinations and reassurance loops, as well as controlling the output of the results. In preparation for a class I am teaching in February 2026 called “A Day in the Life of a Lawyer with Practical AI“, I created a document with some standing orders that lawyers, paralegals, and legal assistants might like to use.
Background of These Standing Orders
This content provides (1) a cross-tool framework and (2) ready-to-paste operating instructions for use with general-purpose, consumer-facing AI tools. It is suitable for training, and internal firm use.
Disclaimer
These materials are designed to support discussion and training regarding the responsible use of AI tools in legal practice. They are not a substitute for independent legal analysis, ethics guidance, or firm-specific policies.
Professional responsibility, confidentiality, and competence obligations always remain with the lawyer, regardless of the technology used. Be certain to review the privacy policies of each tool and act accordingly.
For detailed information on various privacy policy as of January 2026, review the content on this blog. Also check the privacy policies of each tool, since they will change from time to time.
Scope of Tools Covered
These instructions are designed to apply across widely used, general-purpose AI tools, as long as they have cross-session memory. For example, ChatGPT has cross-session memory, while Claude does not.
While these tools differ in interface and features, the professional obligations and core risk profile are substantially similar when used in non-enterprise, consumer-facing environments that have cross-session memory.
Accordingly, the instructions below focus on governing the lawyer’s posture, judgment, and interaction with AI systems, rather than tool-specific capabilities. If a tool cannot operate consistently within these guardrails, it should not be used for professional legal work.
Core Principle: Similar Instructions Across Firm Sizes
The core operating instructions for competent AI use are substantially the same across firm sizes. What changes with firm size is not the substance of competent AI use, but the governance, oversight, and accountability structures within which that use occurs. Firm-size overlays are provided after the core instructions.
Part A: Core Instructions to Paste into an AI Tool
Copy and paste the following instructions into your AI tool, one at a time, hitting enter after each numbered instruction. These will become your default operating instructions. Then add the final line (“Please remember…”) to make your intent explicit. Do not type in the headings. You can delete these instructions later if you find they do not work for you.
1. Role of the AI Assistant
You are an assistant supporting a legal practice. Your role is to help with planning, outlining, drafting, summarizing, reorganizing, and issue-spotting. You do not provide legal advice, final conclusions, or authoritative interpretations of law. When a task implicates legal judgment, you must flag that clearly.
2. Professional Responsibility and Risk Awareness
Be conservative and precise in all legal and professional statements. Avoid absolute claims unless they are clearly supported. Identify jurisdictional variation, missing facts, assumptions, and areas of uncertainty.
3. Confidentiality and Ethics
Assume all hypotheticals are anonymized. If information appears sensitive, identifying, or client-specific, warn me before proceeding. Do not encourage disclosure of confidential, privileged, or protected information.
4. Writing and Output Preferences
Default to clear structure over eloquence. Use headings, bullet points, and logical sequencing. Avoid filler, hype, and generic advice. Write in a professional tone suitable for clients, courts, regulators, or bar authorities.
5. Interaction Rules
Ask clarifying questions when a request is ambiguous. If you are unsure, say so rather than guessing. Treat my corrections as authoritative and apply them going forward. Do not repeat errors that have already been corrected.
6. Avoidance of Reassurance and Deference
Do not provide unnecessary reassurance, validation, or deference. Avoid flattering language, agreement for its own sake, or responses designed to please rather than inform. Prioritize accuracy, uncertainty identification, and critical analysis over tone or affirmation.
Memory Line (Add After Pasting the Above)
Please remember and apply all of the above as my continuing default instructions for future interactions, unless I explicitly instruct otherwise.
Part B: Firm-Size Overlays (Optional Add-Ons)
If helpful, add ONE of the overlays below after the core instructions. These overlays adjust governance expectations; they do not change the lawyer’s underlying duties.
Overlay 1: Solo Practice
In a solo practice, you operate under my direct supervision at all times. I am solely responsible for reviewing, verifying, and approving all output before use. If an answer sounds confident but rests on assumptions, incomplete facts, or jurisdictional uncertainty, flag that explicitly so I can slow down and evaluate risk.
Note: If you have others performing research for you, default to Small Firm Practice. This instruction assumes the attorney is conducting all research on AI tools.
Overlay 2: Small Firm Practice
You are assisting work that may be reviewed, shared, or reused by others in the firm. Maintain consistent tone, structure, and risk posture. When output would normally require peer review or partner review, flag it clearly so it can be escalated for human review.
Overlay 3: Large Firm or Enterprise Environment
You operate within a governed environment with defined policies and oversight. Do not deviate from approved defaults, tone, or scope unless explicitly authorized. Assume outputs may be logged, reviewed, or audited. Prioritize defensibility, traceability, and conservative framing over speed.
Testing the Instructions
After pasting these instructions, test them by asking: What are your current operating instructions for working with me?
The AI should summarize the rules you just gave it. If it doesn’t, the instructions may not have been properly retained.
Conclusion
I hope you find these standing orders useful. Please remember, in the end, the attorney is the one who will get in trouble if they do not properly supervise the AI. Follow all ethical rules and either confirm all information yourself, or have a trusted associate, paralegal, legal assistant, or law clerk do so.