The Problem
Every ChatGPT conversation starts fresh. You repeatedly tell it you're a lawyer, which jurisdiction you work in, and what tone you need.
The Solution
Use Custom Instructions to set default behaviour for ALL conversations.
OpenAI's help centre could not be read when this guide was checked on 29 September 2026, so the guide names no plan and quotes no menu path or field label. Confirm those in OpenAI's help centre.
Before You Use Client Material
Custom Instructions set how ChatGPT answers. They do not change where your data goes, and an instruction about confidentiality does not make client information safe to enter. What you may put into ChatGPT depends on the plan your firm has, its agreement with OpenAI and your firm's policy. Keep client names and matter details out of Custom Instructions and out of your prompts, and use placeholders such as "the Client" and "the Buyer".
The example answers on this page are made up to show the format. No test results have been published for this template.
How to Set Up Custom Instructions
Access Custom Instructions
- Open ChatGPT
- Open its settings
- Find Custom Instructions among the personalisation settings
Two Things to Tell It
Part 1: Who you are (Context about you and your practice)
Part 2: How it should respond (How it should behave)
The labels on the screen may differ from these headings. Put each part in the field that matches it.
Template for Lawyers
Part 1: Who you are
I am a [seniority] [practice area] lawyer in [jurisdiction].
Practice details:
- Jurisdiction: [England & Wales / Scotland / Northern Ireland]
- Practice areas: [Corporate, Employment, Litigation, etc.]
- Firm type: [Magic Circle / High street / In-house / Boutique]
- Typical clients: [Startups / FTSE 100 / SMEs / Individuals]
Experience level: [NQ / 3 PQE / Senior Associate / Partner]
Example (fill in your details):
I am a senior associate employment lawyer in England & Wales.
Practice details:
- Jurisdiction: England & Wales (primarily)
- Practice areas: Employment law, HR advisory, tribunal work
- Firm type: Mid-size commercial firm
- Typical clients: SMEs and mid-market companies
Experience level: 6 years PQE
Part 2: How it should respond
Legal work requirements:
- Apply [jurisdiction] law unless I specify otherwise
- Always cite sources with full citations (case name, neutral citation, court, year)
- If you cannot verify a citation, mark it "UNVERIFIED - check independently"
- Never fabricate case law or statutory references
- Ask clarifying questions before answering if information is unclear
- Flag any assumptions you're making
Output format:
- Keep responses under [300/500] words unless I ask for detail
- Use bullet points for lists
- For legal analysis: Summary → Analysis → Recommendation
- Use [formal / professional-but-clear / client-friendly] tone
Critical rules:
- Never input client-identifiable information (I'll use placeholders)
- Flag when work requires human lawyer oversight
- If uncertain about anything, say so explicitly
- Prioritise accuracy over helpfulness
Example (customise to your needs):
Legal work requirements:
- Apply England & Wales law unless I specify otherwise
- Always cite sources: case law (full citation + court + year), statutes (Act name + year + section)
- Mark uncertain citations as "UNVERIFIED - please check"
- Never fabricate legal authorities
- Ask for clarification if my query is ambiguous
- Flag assumptions explicitly
Output format:
- Concise responses (max 400 words unless I ask for detail)
- Use bullet points and headings
- For client advice: Summary → Key risks → Recommendation → Next steps
- Professional but clear tone (avoid unnecessary jargon)
Critical rules:
- I will not provide client-identifiable data (client confidentiality and
data protection)
- Always flag when a matter needs specialist review
- Be explicit about uncertainty - don't guess
- Accuracy matters more than being comprehensive
What This Achieves
Once set, every new ChatGPT conversation will:
- Know you're a lawyer in your jurisdiction
- Apply the right law by default
- Use your preferred tone and format
- Follow your citation rules
You do not have to explain this in each conversation. Check that the instructions are being followed: an instruction is a request, and the answer can still depart from it.
Example Before vs After
Before Custom Instructions
Your prompt:
Can we dismiss this employee for poor performance?
ChatGPT response:
[Answers under the law of another country, which does not apply
to your client]
[Uses 800 words when you needed 200]
[No citations]
After Custom Instructions
Your prompt:
Can we dismiss this employee for poor performance?
ChatGPT response:
**Summary:** Possible, but you must follow a fair procedure to avoid unfair
dismissal claims.
**Key requirements (England & Wales):**
- Document performance issues
- Provide clear warnings
- Allow improvement opportunity
- Follow a fair procedure (Acas Code of Practice)
- Hold formal meeting before decision
**Risks if no process:**
- Unfair dismissal claim, if the employee has the qualifying service
- Compensation: check the current statutory limits
**Recommendation:** Before proceeding, confirm:
1. Length of service?
2. Contract terms?
3. Previous warnings documented?
**Authority:** Employment Rights Act 1996, ss.94 and 98; Acas Code of Practice
on disciplinary and grievance procedures
**Note:** This is general guidance - review specific facts with employment specialist.
Much better: Correct jurisdiction, concise format, cited sources, flagged limitations.
The example shows the format of a good answer. It is not a statement of the law: check every authority and figure in an AI answer against legislation.gov.uk and the Acas website before you rely on it.
Pro Tips
Customise by work type:
If you do multiple practice areas, choose your primary one or be general:
I handle corporate, commercial, and employment matters.
Default to England & Wales law.
Update as needed:
- Changed firms? Update firm type
- New jurisdiction? Update law to apply
- Different role? Update seniority
Use with the Memory feature: Custom Instructions hold what should apply to every conversation. For what ChatGPT carries over from individual conversations, and how to remove it, see ChatGPT Memory Management.
Test it: After setting up, start a new conversation and ask a legal question. Check if ChatGPT applies your instructions without you repeating them.
Common Mistakes
❌ Too prescriptive: Asking for 10 specific behaviours limits flexibility
❌ Too vague: "I'm a lawyer" (doesn't help much)
❌ Including client names: Never put client/matter details in Custom Instructions (they persist)
✅ Clear, general instructions about jurisdiction, tone, format, and citation standards
Quick Reference
Setup:
-
Open Custom Instructions in ChatGPT's settings
-
Part 1 (Who you are):
I am a [seniority] [practice area] lawyer in [jurisdiction].
Practice areas: [list]
Firm type: [type]
Typical clients: [description]
- Part 2 (How to respond):
Legal requirements:
- Apply [jurisdiction] law
- Cite sources fully
- Mark uncertain citations "UNVERIFIED"
- Ask questions if unclear
Output:
- Concise ([X] words max)
- [Preferred format]
- [Preferred tone]
Critical:
- Prioritise accuracy
- Flag assumptions
- Admit uncertainty
- Save → Test with a new conversation
Result: ChatGPT now knows who you are and how to help you. No more repeating yourself.