79% of legal professionals now use AI in their work, up from 19% in 2023, per Clio's Legal Trends Report.Source ↗
Law firm AI adoption nearly tripled in one year — from 11% in 2023 to 30% in 2024 — and ChatGPT is the most-used tool at 52%, per the ABA Legal Technology Survey.Source ↗
Document review (77%), legal research (74%), and document summarization (74%) are the top generative AI use cases among legal professionals, per Thomson Reuters.Source ↗
A public tracker has documented roughly 1,490 court decisions worldwide — more than 1,000 in the US — where a party relied on AI-hallucinated citations or evidence, with sanctions escalating from $5,000 in 2023 to $15,000 per attorney and bar suspensions by 2026.Source ↗
writingChatGPTClaude

First-draft demand letters from your case facts

A demand letter is formulaic in structure but has to be precise in its facts, and drafting one from scratch eats an hour or two of non-billable-feeling time in a flat-fee or contingency matter. Drafting correspondence is one of the most common AI tasks in legal practice — but the draft is only safe if the AI is confined to your facts and kept away from inventing legal authority.

Prompt
You are an experienced civil litigation attorney drafting a formal pre-suit demand letter. Draft the letter using ONLY the facts I provide below. Do not invent facts, dates, names, or dollar amounts. Do not cite or paraphrase any case law, statute, or regulation — where legal authority would strengthen a point, insert the placeholder [CITE — attorney to add authority] instead.

Matter type: {{matter_type}}
Facts (parties, timeline, what happened, prior attempts to resolve): {{case_facts}}
Damages and amount demanded: {{demand_amount}}
Response deadline: {{response_deadline}}

Structure:
1. Re: line identifying the matter
2. Introduction stating representation
3. Factual background in chronological order, using only my facts
4. Liability position framed as our contention, not as settled law
5. Itemized damages
6. Clear demand with the deadline and how to respond
7. Consequences paragraph stating intent to pursue available civil remedies — no threats beyond that
8. Reservation of rights

Tone: firm, professional, no bluster. Keep it under 700 words. If any fact needed for a section is missing from what I gave you, insert [VERIFY: description] rather than guessing, and list all [VERIFY] and [CITE] items at the end.

Fill in your details and the prompt updates live — then copy.

analysisClaudeChatGPT

First-pass contract review that flags risk clause by clause

Reviewing counterparty paper under deadline pressure is where things get missed — the one-sided indemnity, the auto-renewal, the limitation of liability that quietly isn't mutual. Document review is the single most common generative AI use case among legal professionals at 77%, and a structured first pass from AI means your own read starts from a risk map instead of page one.

Prompt
You are a senior transactional attorney performing a first-pass issue review of a contract on behalf of the {{client_role}}. My client's priorities and concerns: {{key_concerns}}.

Contract text:
{{contract_text}}

Produce:
1. A one-paragraph plain-English deal summary (parties, term, money, what each side must do).
2. A risk table with columns: Section reference | Issue | Why it matters for my client | Suggested redline language. For every issue, quote the exact clause language you are relying on. If you cannot quote it from the text I provided, write [NOT FOUND] instead of assuming.
3. A missing-provisions check: indemnification caps, limitation of liability, termination rights, IP ownership, confidentiality, data protection, insurance, assignment, dispute resolution — mark each Present / Absent / One-sided, with the section reference.
4. The top 5 negotiation priorities, ranked, with one sentence of rationale each.

Constraints: do not state what any statute or case law requires — flag legal questions as [ATTORNEY: legal question] instead. Analyze only the text provided. Frame everything as issues for attorney review, not conclusions.

Fill in your details and the prompt updates live — then copy.

communicationChatGPTCopilot

Plain-English client updates that go out the same day

Failure to communicate is a perennial top source of bar complaints, and the reason updates slip is that translating a procedural event into language a client understands takes real effort at the end of a long day. AI is good at exactly this translation step — you supply what happened, it drafts the explanation, you verify and send.

Prompt
You are a legal writing assistant helping an attorney update a client who has no legal background. Draft a short update email.

What happened: {{case_development}}
Client context (who they are, what they care about, anxiety level): {{client_context}}
What happens next and what we need from the client: {{next_steps}}

Requirements:
- Under 300 words, 8th-grade reading level, short paragraphs.
- If a legal term is unavoidable, define it in a plain-language phrase in the same sentence.
- Neutral about outcome: no predictions, no guarantees, nothing like "we're going to win" — describe what the development means procedurally, not what result it promises.
- Use ONLY the facts I provided. Do not add procedural details, dates, or explanations of the law I did not give you. If something a client would obviously ask is missing, put a "Questions for attorney" list at the end instead of answering it yourself.
- End with: what happens next, what we need from them (or "nothing right now"), and an invitation to reply with questions.
- Subject line under 8 words.

Fill in your details and the prompt updates live — then copy.

planningClaudeChatGPT

Deposition outlines built from your case file

A good deposition outline means synthesizing pleadings, documents, and prior testimony into an examination plan — hours of prep for every hour on the record. Attorneys report using AI to organize deposition topics and even simulate witness Q&A; the AI can't know your case, but it can turn your own summary of it into a structured outline faster than a blank page can.

Prompt
You are a litigation support assistant helping an attorney prepare a deposition outline. Work ONLY from the information I provide — do not invent facts, documents, or testimony. Where a line of questioning depends on a document or fact I have not given you, insert [NEED DOC: description] instead of assuming it exists.

Case background: {{case_summary}}
Witness and their role: {{witness_role}}
Key documents and known facts (my summaries): {{key_documents}}
What I need from this deposition: {{deposition_goals}}

Produce a working outline:
1. Foundation module: background, role, and knowledge-establishing questions.
2. Topic modules in logical order. For each: the goal in one line; open-ended questions moving from broad to specific, one fact per question, no compound questions; the document to introduce with an [Ex. __] placeholder; follow-up branches for the two most likely answers.
3. Admissions to lock in, phrased as short leading questions for the end of each module.
4. A closing checklist: exhibits to authenticate, privilege landmines to avoid, and open [NEED DOC] items.

No legal conclusions or objections coaching — this is a draft outline for attorney revision.

Fill in your details and the prompt updates live — then copy.

automationClaudeGemini

Turning document dumps into a verified case chronology

Personal injury, employment, and insurance files arrive as hundreds of pages of medical records, emails, and reports, and someone has to turn them into a dated chronology before anyone can think about the case. Document summarization is a top-three legal AI use case at 74% — and running it as a repeatable batch process, with page-cite verification built in, is where the real hours come back.

Prompt
You are a litigation assistant building a case chronology from documents I will paste in batches (medical records, emails, incident reports — anonymized). Process each batch the same way so results can be merged.

Case issues to flag entries against: {{case_issues}}

For the batch below, extract EVERY dated event into a table with columns:
Date (YYYY-MM-DD) | Source document and page/paragraph | Event (25 words max) | People involved | Issue flag (from my list, or none) | Key language (short verbatim quote)

Rules:
- Include only events explicitly stated in the text. Never infer a date, fill a gap, or resolve an ambiguity — if a date is unclear or missing, use [UNDATED] and note why.
- Where two documents conflict, record both entries and add a "CONFLICT" note describing the discrepancy; do not pick a side.
- Quote key language exactly as written, including errors.
- After the table, list: date range covered, documents processed, and any pages that appear to be missing from a sequence.

Document batch:
{{document_text}}

Fill in your details and the prompt updates live — then copy.

Common questions from lawyers

Is it against bar rules for a lawyer to use AI at all?

No. ABA Formal Opinion 512 (2024) and guidance from state bars including California and Florida permit generative AI, but they map your existing duties onto it: competence with the tool's limits, confidentiality of client information, candor to the tribunal, supervision of staff who use it, and reasonable fees. What gets lawyers in trouble is not using AI — it's filing its unverified output.

Can I paste client documents into ChatGPT?

Not into a consumer version that may retain or train on your inputs. California's bar guidance says inputting client information into such tools can itself breach the duty of confidentiality, and Opinion 512 requires informed client consent before feeding confidential information to tools without adequate safeguards. Use an enterprise or zero-retention deployment, or de-identify the material first.

How do I avoid the fake-citation sanctions I keep reading about?

Adopt one non-negotiable rule: no citation goes into any filing unless you have pulled and read the authority in Westlaw or Lexis yourself. A public tracker counts roughly 1,490 court decisions involving hallucinated AI material, with penalties now reaching $15,000 per attorney and bar suspensions. Asking the chatbot whether its cases are real does not count — that's what the Mata v. Avianca lawyers did.

Do I have to tell clients or the court that I used AI?

Sometimes. Opinion 512 requires client consent when you input their confidential information into AI tools, and Florida's Opinion 24-1 requires disclosure where AI use affects billing. A growing number of judges have standing orders requiring parties to disclose or certify AI use in filings — check each court's orders before you file.

Related professions