ChatGPT for Legal Work: Prompts and Tips

Before: The Weak Prompt

A recent bar association survey found that most attorneys using ChatGPT rely on a single, near-identical prompt for almost every task -- something like "review this and tell me if there are any issues." That single finding explains a lot about why so many lawyers report being underwhelmed by chatgpt prompts for lawyers they've seen online: the prompt itself is doing almost no work.

Review this contract and tell me if there are any issues.

What this does: technically produces a response, but "issues" is undefined -- so the model has to guess whether you mean liability exposure, ambiguous language, missing clauses, or something else entirely, and it will spread its attention thin across all of them rather than going deep on what actually matters for your specific concern.

Why It Fails

Legal review isn't one task, it's several distinct tasks wearing the same trench coat: checking for missing standard clauses, flagging ambiguous language, verifying internal consistency, and assessing liability exposure are all different reading tasks that benefit from different instructions. A generic "find issues" prompt asks the model to do all four at a shallow level instead of one at a thorough level.

⚠️ Common mistake: Asking for a general contract review without specifying which type of issue you're most concerned about. This produces a scattershot response that touches lightly on several categories instead of a thorough pass on the one that actually matters for this specific document and situation.

After: The Improved Prompt

Review this commercial lease for ambiguous language specifically -- 
clauses where the meaning could reasonably be interpreted two 
different ways. For each one you find, quote the specific clause, 
explain the two possible interpretations, and note which party each 
interpretation would favor.

What this does: narrowing the task to ambiguity specifically, and requiring the two-interpretation breakdown for each flagged clause, produces a genuinely useful analysis instead of a vague "this seems fine" or an unfocused list of unrelated observations.

⚡ Pro tip: Run separate passes for separate concerns -- one prompt for ambiguity, one for missing standard clauses, one for liability exposure -- rather than one prompt trying to do all three. Each individual pass will be more thorough than one prompt doing all three shallowly.

Breaking Down Each Element

Specifying "ambiguous language" as the exact category narrows the model's attention to a well-defined linguistic problem rather than an open-ended sense of "issues." Requiring a direct quote for each flagged clause makes the output verifiable -- you can check the quote against the actual document instead of trusting a paraphrased summary. And asking which party each interpretation favors adds the practical context that turns a linguistics exercise into something actually useful for negotiation.

A paralegal at a small firm uses a similar structure for a different but related task: checking a contract against a standard clause checklist to catch omissions before a partner's review of the final draft.

Compare this service agreement against this list of standard clauses 
our firm typically includes: [list]. Flag any that are missing 
entirely, and any that are present but use noticeably different or 
weaker language than our standard template.

What this does: comparing against a named, specific checklist rather than asking the model to intuit what "standard" clauses should exist catches both outright omissions and the subtler problem of a clause being present but watered down.

⚠️ Common mistake: Asking ChatGPT to identify "missing standard clauses" without providing your firm's actual standard list. Without a reference point, the model will draw on general contract conventions, which may not match your firm's specific practice or the particular type of agreement at hand.

Variations for Different Contexts

Client-facing legal writing needs a different kind of specificity than internal review -- the goal shifts from thoroughness to clarity for a non-lawyer audience. An estate planning attorney uses ChatGPT to draft explanatory letters that accompany a will or trust document:

Write a letter to a client explaining the key provisions of their new 
trust document in plain language. Client is not familiar with legal 
terminology. Focus on: who the trustee is and what they do, how 
assets will be distributed, and when the trust becomes active. Avoid 
legal jargon or explain it immediately if you must use a term.

What this does: naming the three specific provisions to explain, rather than "summarize the trust," keeps the letter focused on what a client actually needs to understand rather than an exhaustive walk-through of every clause, most of which won't matter to their day-to-day understanding.

⚡ Pro tip: For client-facing legal communication, always specify which 2-3 provisions matter most for that specific client's situation rather than asking for a full summary. A focused explanation gets read and understood; a comprehensive one gets skimmed.

A litigation associate uses ChatGPT to prepare deposition question outlines, where the value is in organizing scattered case facts into a logical, useful sequence rather than generating novel legal strategy from scratch:

Here are the key facts and prior statements from this witness: 
[paste]. Organize a deposition question outline that starts with 
background/foundation questions, then moves to questions that probe 
inconsistencies between their prior statements and the documented 
timeline.

What this does: structuring the outline around foundation-then-inconsistencies, rather than a flat list of questions, mirrors how an actual deposition strategy is built and surfaces the inconsistencies that matter most in a logical sequence rather than scattered throughout.

⚠️ Common mistake: Treating ChatGPT's legal research or case-law summaries as verified and citation-ready without independent confirmation. Always verify any case citation or legal claim against a proper legal research database before it goes anywhere near a filing -- this is one of the highest-stakes accuracy requirements in any professional use of AI, and it's not optional.

Internal Memos and Case Summaries

Beyond client-facing work, a substantial amount of legal writing is internal -- memos to partners, case summaries for a team file, research summaries for a colleague picking up a matter. A corporate associate at a mid-size firm uses ChatGPT to turn scattered research notes into a structured memo before a partner review:

Turn these research notes into a memo structure: issue presented, 
brief answer, discussion (organized by the three sub-issues in my 
notes), and conclusion. Keep my actual analysis and citations exactly 
as written -- just organize them into this structure, don't add new 
legal analysis.

What this does: the explicit instruction to organize rather than add new analysis keeps the model in a formatting role rather than a substantive legal reasoning role, which matters because the associate's own analysis is what the partner is actually reviewing -- not an AI's independent take on the law.

⚠️ Common mistake: Letting ChatGPT add its own legal analysis or conclusions when the actual task is organizing existing analysis into a cleaner structure. Being explicit about "organize, don't add new analysis" prevents the memo from drifting into unverified legal claims that weren't part of the associate's own research.

A family law attorney uses a similar structure for case file summaries, useful both for internal handoffs and for refreshing her own memory before a scheduled hearing on a case that's been open for many months already:

Summarize this case file into: key dates, current status of each 
motion, and open items requiring action before the next hearing on 
[date]. Base this only on the documents provided, and flag if any 
date or status seems unclear from the file.

What this does: the instruction to flag unclear items rather than fill gaps confidently prevents the summary from presenting an uncertain detail (a filing date that's ambiguous in the file, for example) as settled fact, which matters a great deal when the summary is being used to prepare for an actual hearing.

⚡ Pro tip: For any case summary or memo drawn from source documents, always instruct ChatGPT to flag anything unclear in the source material rather than resolve ambiguity on its own. An attorney catching a flagged ambiguity in five seconds is far better than an attorney discovering a wrong assumption in a hearing, in front of a judge, with no time left to fix it.

Save and Reuse This

The pattern across every prompt here: define exactly which legal concern you're checking for, provide your own reference standards where they exist (clause checklists, firm templates), and treat anything client-facing as a clarity exercise rather than a completeness exercise. Legal work rewards precision in the prompt as much as it rewards precision in the document itself.

If you're running contract reviews, client letters, or deposition prep regularly, it's worth saving these prompt structures with your firm's specific checklists and tone built in, rather than reconstructing them from scratch on every matter. PromptABCD works well for this kind of reusable legal workflow template, letting you swap in new case details without rebuilding the underlying structure each time a new matter comes across your desk.