Skip to main content
openai tutorialOpenAI tutorial

Tutorial 06: Legal Workflows with ChatGPT and Custom GPTs

Master legal workflows with Custom GPTs for contract review, NDA triage, legal research planning, clause comparison, and compliance checking.

What You'll Learn

This tutorial walks you through achieving legal workflows with ChatGPT and Custom GPTs. OpenAI does not offer a dedicated Legal Plugin like Claude; instead, you use Custom GPTs with custom instructions and structured prompts to achieve similar outcomes: contract review, NDA triage, and other legal tasks.

Learning Objectives

By the end of this tutorial, you will:

  • Understand how to replicate legal plugin workflows with Custom GPTs
  • Master contract review, NDA triage, and briefing workflows
  • Configure your organization's playbook in a Custom GPT
  • Understand how Custom GPTs compare to dedicated legal tools (e.g., Harvey, Legora)

Custom GPTs can provide legal-focused workflows similar to a dedicated Legal Plugin:

  • Pre-built workflows for common legal tasks via custom instructions
  • Configurable playbook integration
  • Risk-flagging with color-coded assessments (RED/YELLOW/GREEN)
  • Redline suggestions with specific language
  • Structured prompts for consistent output

How It Compares to Competitors

FeatureChatGPT Custom GPTsClaude Legal PluginHarveyLegora
Contract ReviewCustom GPT + prompts/review-contractWorkflowsTabular Review
NDA TriageCustom GPT + prompts/triage-ndaYesYes
Custom PlaybookFull controlFull controlVariesVaries
Risk RatingsRED/YELLOW/GREENRED/YELLOW/GREENYesYes
Redline SuggestionsYesYesYesYes
Bulk ProcessingManual/conversationVaries by productEnterpriseEnterprise
Pricing ModelPublic plans (verify current)Public plansEnterpriseEnterprise

Harvey and Legora features are from vendor materials; verify capabilities and current offerings with each vendor.

Requirements

What You Need

  • ChatGPT plan that supports Custom GPTs (e.g., Plus, Pro, Business, Enterprise—verify current plans at OpenAI)
  • Ability to create Custom GPTs
  • Optional: Playbook configuration in GPT instructions

Pricing and plan names are set by OpenAI and may change. Verify current pricing and feature availability on the official OpenAI site.


Part 2: Workflow Reference

Contract Review - Clause-by-Clause Analysis

Purpose: Comprehensive contract analysis against your playbook

How to Use: Create a Custom GPT with your playbook in the instructions. Then upload a contract and use:

Review this contract against our playbook.
Provide a clause-by-clause analysis with risk ratings (GREEN/YELLOW/RED).
Flag any deviations from playbook positions and recommend negotiation strategy.

What It Does:

  1. Identifies contract type and parties
  2. Analyzes each material clause
  3. Compares to your configured playbook positions
  4. Assigns risk ratings (GREEN/YELLOW/RED)
  5. Provides specific redline suggestions
  6. Generates negotiation priority list

Example Output Structure:

## CONTRACT ANALYSIS: Software License Agreement
**Contract Type**: SaaS Subscription Agreement
**Our Role**: Customer/Licensee
**Counterparty**: Acme Software Inc.

### CLAUSE ANALYSIS

#### 1. LICENSE GRANT (Section 2.1)
GREEN - Acceptable
- Grants non-exclusive, worldwide license
- Includes affiliate usage rights
- Standard use restrictions

#### 2. LIMITATION OF LIABILITY (Section 8)
RED - Must Negotiate
Current: "Vendor's liability shall not exceed $10,000"
Issue: Cap is far below contract value ($150,000 ACV)
Playbook Position: Minimum 12 months fees ($150,000)

**Suggested Redline**:
"Vendor's liability shall not exceed ~~$10,000~~ the greater
of (a) fees paid in the twelve (12) months preceding the
claim or (b) $150,000."

#### 3. INDEMNIFICATION (Section 7)
YELLOW - Should Negotiate
Current: IP indemnity excludes "modifications by Customer"
Issue: Overly broad exclusion could limit protection
Playbook Position: Narrow exclusions to material modifications

**Suggested Redline**:
Add: "provided such modification materially alters the
functionality of the Service in a manner not contemplated
by the Documentation"

### NEGOTIATION PRIORITIES
1. RED Liability Cap (critical gap)
2. RED Data Breach Indemnity (missing entirely)
3. YELLOW Indemnity Exclusions (overbroad)
4. YELLOW Termination Notice (60 vs 30 days)
5. GREEN Auto-renewal period (acceptable at 30 days)

### CUMULATIVE RISK SCORE: 7 (HIGH)
Recommend: Partner review before signing

NDA Triage - Rapid Pre-Screening

Purpose: Quickly categorize incoming NDAs for appropriate handling

How to Use: Create a Custom GPT for NDA triage. Upload the NDA and prompt:

Triage this NDA. Categorize as:
- GREEN - STANDARD APPROVAL: Matches our standard or better
- YELLOW - COUNSEL REVIEW: Minor deviations
- RED - FULL REVIEW: Significant issues

For each category, list specific concerns and recommended action.

Triage Categories:

CategoryDescriptionAction
GREEN - STANDARD APPROVALMatches our standard or betterParalegal can execute
YELLOW - COUNSEL REVIEWMinor deviationsAttorney quick review
RED - FULL REVIEWSignificant issuesFull legal analysis

Example Output:

## NDA TRIAGE RESULTS

**Document**: Acme Corp Mutual NDA (v2.1)
**Type**: Mutual Non-Disclosure Agreement
**Direction**: Two-way protection

### TRIAGE RESULT: COUNSEL REVIEW

**Reason**: Contains 2 non-standard provisions requiring attorney review

### FLAGGED ITEMS:

1. **Non-Solicit Clause (Section 6)**
   - Not typically included in standard NDA
   - 24-month restriction on hiring
   - May need business approval

2. **Jurisdiction (Section 9)**
   - Specifies Texas courts exclusive jurisdiction
   - Our standard: Delaware or mutual agreement
   - Acceptable but note for negotiation if sensitive deal

### STANDARD ITEMS (No Issues):
- Definition of Confidential Information
- Mutual obligations
- 3-year term
- Standard exclusions
- Return/destruction requirement

### RECOMMENDATION:
Route to associate for 15-minute review of flagged items.
If non-solicit is acceptable to business, can proceed.

Operational Impact

Triage workflows can reduce review time materially, depending on document quality, playbook maturity, and team process.


Vendor Check - Agreement Status

Purpose: Check status of agreements with known vendors

How to Use: If you have document access, create a Custom GPT that can reference your contract database. Otherwise, use a prompt with manual input:

I need a vendor status check for [vendor name].

Available information:
- [List active agreements, expiration dates]
- [Renewal dates]
- [Historical notes]

Provide: Active agreements, upcoming renewals, recommended actions.

Brief Generation - Contextual Briefings

Purpose: Create briefing documents for various needs

Prompt for Incident Response:

Create an incident response brief for [data-breach/ litigation hold/ etc.].

Include:
- Immediate actions (0-24 hours)
- Notification requirements by jurisdiction
- Template resources
- Key contacts

Prompt for Daily Brief:

Create a daily brief on my pending matters. Include:
- Matters requiring attention today
- Upcoming deadlines
- Status updates needed

Purpose: Generate standard responses for common requests

Prompt for DSAR:

Draft a Data Subject Access Request response template.
Include: verification statement, data categories table, rights notice,
retention caveats, supervisory authority contact. Mark internal notes
separately.

Prompt for Litigation Hold:

Draft a litigation hold notice for [situation].
Include: preservation scope, custodian list, document types,
compliance requirements, acknowledgment section.

Part 3: Configuring Your Playbook in a Custom GPT

Configuration via Custom Instructions

Create a Custom GPT and add your playbook to the instructions. Example structure:

You are a legal contract review assistant for [Firm Name].

## PLAYBOOK POSITIONS

### Liability Cap
- Standard: 12 months fees
- Minimum: contract value
- Carve-outs: indemnification, data breach, IP infringement, confidentiality, gross negligence, willful misconduct

### Indemnification
- Required from vendor: IP infringement, data breach, gross negligence
- Required mutual: third party claims from breach
- Unacceptable: customer indemnifies for vendor negligence

### Data Rights
- Ownership: customer owns all customer data
- Vendor usage: service delivery only
- Prohibited uses: AI training, analytics, marketing
- Deletion timeline: 30 days
- Breach notification: 72 hours

### Termination
- Preferred: annual with 30 days TFC
- Acceptable: 60-90 days notice
- Unacceptable: no TFC, >90 day notice

## RISK THRESHOLDS
RED: uncapped customer liability, no vendor indemnity, data used for training
YELLOW: liability cap below 12 months, limited indemnity carve-outs

## OUTPUT FORMAT
Always use RED/YELLOW/GREEN ratings. Provide specific redline suggestions for RED and YELLOW items.

Setting Up Your Custom GPT

Step 1: Create a new Custom GPT in ChatGPT Step 2: Name it (e.g., "Contract Review - [Firm Name]") Step 3: Paste your playbook into the instructions Step 4: Enable file uploads for contract documents Step 5: Test with a sample contract Step 6: Refine instructions based on output

Customize for Your Needs

The playbook drives all risk assessments and redline suggestions. Generic playbooks will produce generic results. Invest time customizing positions for your organization's risk tolerance and standard negotiation positions.


Part 4: Workflow Examples

Workflow 1: End-to-End Contract Review

1. Receive contract from business team

2. Open your Contract Review Custom GPT

3. Upload contract

4. Prompt: "Review against our playbook. Provide clause-by-clause analysis with risk ratings."

5. Review output:
   - Check risk ratings
   - Review suggested redlines
   - Verify against your judgment

6. Generate response or manual drafting based on suggestions

7. Document in matter management

Workflow 2: NDA Volume Processing

Day starts: 15 NDAs in queue

1. Open NDA Triage Custom GPT

2. For each NDA:
   - Upload NDA
   - Prompt: "Triage this NDA. GREEN/YELLOW/RED with specific concerns."

3. Sort results:
   GREEN STANDARD (6): Send to paralegal for execution
   YELLOW COUNSEL (7): Quick 10-min reviews
   RED FULL REVIEW (2): Schedule detailed analysis

4. Process YELLOW queue: Quick review flagged items

5. Process RED queue: Full contract review GPT for detailed analysis

Total time: ~2 hours vs. 6+ hours manually

Workflow 3: Incident Response

1. Incident reported (potential data breach)

2. Open ChatGPT, prompt: "Create incident response brief for data breach.
   Include: immediate actions, notification requirements by jurisdiction,
   template resources, key contacts."

3. Review notification requirements

4. Prepare communications using templated response prompts

5. Document privileged analysis in separate conversation

6. Generate status updates as needed

Part 5: Best Practices

Do's

  • Do configure your playbook in Custom GPT instructions before heavy use
  • Do verify all citations and legal conclusions
  • Do use separate Custom GPTs or conversations for matter separation
  • Do review RED items carefully before accepting suggestions
  • Do customize templates for your jurisdiction

Don'ts

  • Don't rely solely on ChatGPT output for legal decisions
  • Don't skip verification of case citations
  • Don't use generic playbook for specialized matters
  • Don't share ChatGPT output with clients without review
  • Don't assume ChatGPT catches everything

Professional Judgment Required

Custom GPTs are powerful efficiency tools, but they are not a replacement for attorney judgment. Always verify critical recommendations, especially RED-flagged items and suggested redlines.

Quality Control Checklist

For every GPT-assisted review:

  • Verified contract type identification is correct
  • Confirmed our role (customer/vendor) is correct
  • Checked each RED item manually
  • Verified suggested language is appropriate
  • Confirmed jurisdiction-specific requirements
  • Reviewed for issues the GPT may have missed
  • Applied professional judgment to recommendations

Custom GPT Guardrail Checklist

Before deploying a legal Custom GPT to your team:

  • Playbook positions reflect current firm standards
  • Instructions explicitly state "verify all citations" and "do not provide legal advice without qualification"
  • File upload scope is limited to intended document types
  • Separate GPTs or conversations used for matter isolation
  • No client names or matter identifiers in GPT instructions
  • Sharing set to Organization only (not Public) for firm GPTs
  • Version or date noted in instructions for change tracking

Part 6: GPT Actions (Optional Integration)

When to Use Actions

GPT Actions let Custom GPTs call external APIs. Use them when you need:

  • Live data from your contract management system
  • Integration with legal research tools
  • Document retrieval from SharePoint, iManage, or NetDocuments
CapabilityClaude Legal PluginChatGPT Custom GPT
Slash commandsBuilt-inSimulated via instructions
MCP-style connectorsNativeVia Actions (API integration) or MCP apps (plan-dependent)
Playbook file~/.claude/legal-playbook.jsonIn GPT instructions or knowledge file
Vendor check/vendor-check + MCPActions + your API

If you have an internal contract API, create a GPT Action that your Custom GPT can invoke. Otherwise, use manual input or separate integrations. See OpenAI Actions documentation for setup. MCP app support varies by plan—check OpenAI Help Center for current options.


Part 7: Troubleshooting

Common Issues

IssueCauseFix
Wrong risk ratingsGeneric or missing playbookAdd explicit positions to instructions; include RED/YELLOW thresholds
Inconsistent output formatVague instructionsAdd "OUTPUT FORMAT" section with required structure
GPT ignores playbookPlaybook buried in long instructionsPut playbook at top; use clear headers
Wrong party roleNot specifiedAdd "Our Role: [customer/vendor]" to prompt or instructions
Missing redlinesModel skips suggestionsAdd "For RED and YELLOW, provide specific alternative language"
Token limit exceededLong contract + long playbookSplit review by section; use summarization first

Prompt Refinement Tips

  • Start prompts with context: "We are the customer. Review against our SaaS playbook."
  • Request structured output: "Use RED/YELLOW/GREEN. Provide a negotiation priority list."
  • Add verification reminder: "Note that all citations require Westlaw/Lexis verification."

Do This Now

  • Create a Custom GPT with your playbook in the instructions
  • Run a contract review on a sample contract and review the risk ratings
  • Run NDA triage on an NDA and confirm the result matches your judgment
  • Add one workflow to your personal checklist for next week

Workflow Quick Reference

TaskCustom GPT SetupPrompt
Contract ReviewPlaybook in instructions"Review against playbook. Clause-by-clause with RED/YELLOW/GREEN."
NDA TriageTriage criteria in instructions"Triage this NDA. GREEN/YELLOW/RED with concerns."
Incident BriefN/A"Create incident response brief for [type]."
DSAR ResponseN/A"Draft DSAR response template."
Litigation HoldN/A"Draft litigation hold notice for [situation]."


Next Steps

Continue to Tutorial 07: MCP Integrations for Legal Work for connecting to external legal data sources.

Previous: Tutorial 05: Building Custom Negotiation Playbooks

Sources

Additional Reading