ChatGPT for legal documents: Use cases, tips & prompts for legal teams

Legal work has always been document-heavy. Contracts, NDAs, engagement letters, settlement agreements, SOWs — a significant portion of a legal team's time goes not into legal judgment, but into drafting, formatting, and reviewing documents that follow largely predictable patterns.

ChatGPT has changed what that work looks like. Legal teams at law firms and in-house departments are using it to produce first drafts faster, review language more efficiently, and spend more time on the work that actually requires legal expertise. This is a practical guide to ChatGPT for lawyers and legal teams of all sizes, covering the common use cases, the prompts that produce solid results, and the important limitations to keep in mind.

Can lawyers actually use ChatGPT? A quick note on ethics and accuracy

Before exploring legal ChatGPT use cases, here are the answers to the two questions that come up most often:

Is it ethical to use ChatGPT for legal work?

In most jurisdictions, using AI tools for legal work is permitted, but the lawyer remains responsible for the output. The American Bar Association and most state bars have issued guidance making clear that AI-assisted drafting is acceptable as long as the attorney reviews the work, understands it, and takes responsibility for it. Several bars have also issued guidance on disclosure obligations when using AI. Check your jurisdiction's current guidance before relying on AI-generated content in client-facing work.

Is ChatGPT accurate for legal content?

ChatGPT is a language model, not a legal database. It can produce plausible-sounding legal language that contains errors, outdated references, or jurisdiction-specific mistakes. It should be used as a drafting assistant and a starting point rather than as a source of legal authority. Always review and verify AI-generated legal content before use.

With those caveats established, here's how legal teams are actually using it.

7 practical use cases for ChatGPT in legal document work

1. Drafting first versions of standard agreements

The most common use case is also the most straightforward. For agreements that follow a consistent structure — NDAs, consulting agreements, service contracts, employment agreements — ChatGPT can produce a usable first draft in seconds.

The key is specificity. A vague prompt produces a vague draft, and a specific prompt produces something much closer to what you actually need.

Try this prompt: Draft a mutual non-disclosure agreement for a potential technology partnership between two US-based companies. The agreement should cover a 3-year confidentiality period, exclude information already in the public domain, and include a provision requiring written notice before any disclosure is required by law. Governing law should be Delaware.

What to do with the output: Treat it as a starting point. Review every clause against your standard language and jurisdiction requirements. The value is not that the draft is perfect, it’s that you’re editing rather than writing from scratch, which is significantly faster.

2. Reviewing and summarizing contract language

Reviewing a 40-page commercial agreement before a client call becomes a different task when you can ask ChatGPT to surface the most important provisions first.

Try these prompts: 

• I'm going to paste a commercial services agreement. Please summarize: (1) the key payment terms, (2) any limitation of liability provisions, (3) the termination rights for each party, and (4) any unusual or non-standard clauses I should flag. Here is the agreement: [paste text]

• In the indemnification clause I just pasted, what obligations does the vendor have if a third-party IP claim is made? Flag any language that could create unlimited liability for either party.

What to do with the output: Use the summary to orient yourself before reading the full document. Verify any flagged provisions directly in the contract text rather than relying solely on the AI's characterisation of them.

3. Drafting client-facing communications

Legal writing and client communication are different skills, and ChatGPT is particularly useful for translating complex legal concepts into plain language or for drafting professional communications quickly.

Try these prompts: 

• Draft a client email explaining that we've received a request to extend the response deadline by 14 days. We are inclined to agree as a professional courtesy but want to set the expectation that we will not agree to further extensions. The tone should be professional and firm but not adversarial.

• Write a short engagement letter covering scope, fees ($X per hour, billed monthly), payment terms (net 30), and termination rights (either party, 30 days written notice). The client is a small business owner retaining us for general commercial matters.

What to do with the output: Review the tone carefully before sending. Read your agreement as your client would to check that it sounds like you. Adjust anything that feels impersonal or off-brand for your firm. 

4. Creating and refining contract clauses

Rather than drafting a full agreement, you can use ChatGPT to draft or improve specific clauses. This is particularly useful when you are redlining an agreement and need alternative language quickly.

Try these prompts:

• Draft three alternative versions of a limitation of liability clause for a SaaS services agreement: one that favors the vendor, one that favors the customer, and one balanced version. Each should cap liability at 12 months of fees paid.

• Rewrite this indemnification clause to narrow the scope to direct damages only and exclude consequential damages. Keep the core obligation intact: [paste clause]

• I need a force majeure clause that includes pandemics, cyberattacks, and supply chain disruption. It should excuse performance for the duration of the event but require the affected party to give written notice within 5 business days.

What to do with the output: Compare the generated versions against your preferred standard language. For high-stakes provisions like limitation of liability and indemnification, treat the output as a reference rather than final language, and make sure the clause integrates correctly with the rest of the agreement before accepting it.

5. Generating template libraries and playbooks

For legal teams that create the same document types repeatedly, ChatGPT can help build template libraries and internal playbooks — standard clause banks, fallback positions, and guidance documents for non-legal teams who need to draft agreements.

Try these prompts:

• Create a clause bank for a standard SaaS subscription agreement. For each of the following sections, provide a preferred version, an acceptable fallback, and a version we should not accept: (1) limitation of liability, (2) indemnification, (3) data processing and security, (4) termination for convenience.

• Write internal guidance for our sales team on what they can and cannot agree to in an NDA without legal review. Use plain language. Assume the reader has no legal background.

What to do with the output: Have a qualified attorney review any template or playbook before it is distributed or used. ChatGPT can produce a strong first structure, but the specific language in a clause bank needs to reflect your actual risk positions and will need to be tailored to your client base and jurisdiction.

6. Preparing for negotiations

Understanding what the other side might want — and why — helps you go into a negotiation better prepared. ChatGPT can help you think through common positions and identify likely pressure points before you get to the table.

Try these prompts:

• I'm about to negotiate a master services agreement with a large enterprise customer. They've sent back redlines that significantly expand the indemnification scope and remove the limitation of liability cap entirely. What are their likely reasons for these positions, and what are common middle-ground positions that both sides typically accept?

• What are the 5 most commonly disputed provisions in a commercial software license agreement, and what are the typical positions for licensor vs. licensee on each?

What to do with the output: Use it as preparation material, not as a negotiating script. The positions ChatGPT describes are general patterns, and your counterparty will have specific motivations that may differ. Treat the output as background context that helps you ask better questions and anticipate objections.

7. Drafting and sending agreements directly from ChatGPT

Once a legal document is drafted, getting it signed has traditionally meant switching tools — copying from ChatGPT into your document platform, setting up signature fields, adding recipients, and sending. The PandaDoc ChatGPT App removes that step.

With the PandaDoc app connected to ChatGPT, you can describe the agreement you need, have ChatGPT draft it, and then send it for signature — all without leaving the conversation. The draft becomes a formatted, signature-ready agreement in PandaDoc automatically, with fields placed and recipients added through the same chat.

Example workflow:

You: Create an NDA for a new vendor relationship — mutual confidentiality, 2-year term, Delaware governing law. Add legal@acme.com as the counterparty signer.

ChatGPT + PandaDoc: Here's your NDA. I've created it in PandaDoc with signature fields placed. Legal@acme.com has been added as the signer. Ready to send for signature?

For legal teams that send agreements at volume, removing the steps between "drafted" and "sent for signature" is a meaningful time saving.

What to do with the output: Review the generated agreement in the embedded PandaDoc widget before sending — check the key terms, confirm the parties are named correctly, and verify the signature fields are in the right places. Once you're satisfied, send directly from the conversation.

Get the PandaDoc ChatGPT App →

ChatGPT prompts for lawyers: tips for getting better results

Every ChatGPT prompt for lawyers works better when it includes critical context like the governing jurisdiction, the parties involved, and the specific output you need.

Be jurisdiction-specific

ChatGPT typically defaults to general US law unless told otherwise. Always specify the governing jurisdiction, and review the output for jurisdiction-specific requirements that may not be captured.

Define the parties and context

"Draft an NDA" produces a generic result. "Draft an NDA between a software company and a freelance developer for a 6-month engagement, California law" produces something much more usable and saves significant editing time.

Ask for multiple versions

For contested clauses, asking for two or three versions with different risk profiles is faster than iterating on a single draft. You can then compare and pick the starting point that best matches your position.

Give it context about your tone and style before you start 

You can paste a sample of your own previous work — a clause you like, an email that hit the right tone, an agreement you consider well-drafted — and ask ChatGPT to match that style. For example: "Here is an example of how I typically draft limitation of liability clauses. Use this as a reference for tone and style when drafting the clause I'm about to request." This is especially useful for client communications where your personal voice matters, and for firms that have a house style they want to preserve across documents.

Use it for structure, not final language

ChatGPT is excellent at producing a complete document structure quickly. The specific language, particularly in high-stakes provisions, should be reviewed carefully and often rewritten based on your firm's standards.

Never paste confidential information into a standard ChatGPT session 

If client confidentiality is a concern, use your firm's enterprise ChatGPT instance with appropriate data handling agreements in place, or use PandaDoc AI which keeps your agreement data within your PandaDoc environment.

Always review before sending

AI-generated legal language can be subtly wrong in ways that are not immediately obvious. Every AI-drafted document needs attorney review before it is used.

What ChatGPT can't do for legal teams

• It can't give legal advice. It can draft language, summarize provisions, and generate options. It cannot replace the legal judgment that comes from knowing your client, the facts, and the applicable law.

• It doesn't know your jurisdiction's current law. ChatGPT's training data has a cutoff date and may not reflect recent case law, regulatory changes, or jurisdiction-specific requirements.

• It can only act on the context you give it. Without additional input, ChatGPT doesn't know your client's risk tolerance, negotiating history, or preferred positions on contested provisions. However, this is exactly where structured guidance helps — if you provide a pricing policy, a clause bank, or a set of fallback positions (what PandaDoc calls a Playbook), ChatGPT can apply those standards consistently across every document. The more context you give it upfront, the more it reflects your client's actual interests rather than generic legal language.

• It can't verify accuracy. ChatGPT can produce confident, plausible-sounding language that is factually incorrect. Every output needs human review before it goes anywhere near a client.

Get the PandaDoc ChatGPT App →