AI for Lawyers: Documents, Client Work, and Content

You will read it in 24 minutes. You will close your first task today. In a week AI becomes part of your desk, and you will be surprised how many evenings it hands back.

Section 01What AI can do for a lawyer, and where its ceiling is

Start with an honest frame, so nothing disappoints you later. AI is not a legal database and not a second lawyer with a bar card. It carries no current code of statutes and no docket of cases inside it. What it does have is a strong command of language and structure: it knows what a lease or a demand letter usually looks like, and it assembles text to fit your task. Its real strengths grow out of that.

What AI does well:

• writes the first version of a document from your template and inputs;
• reads a long document from the other side and pulls out the risky clauses and contradictions;
• translates heavy legalese into plain English so a client actually understands it;
• structures the chaos: turns a tangled thread of messages into a timeline and a list of facts;
• generates alternative wordings when you are stuck on one phrase;
• takes on the content: posts, case studies, answers to common questions, a newsletter.

Now the ceiling, which matters more than the strengths. AI is not responsible for the outcome. It can cite, with full confidence, a statute that was repealed years ago, or invent a case number out of thin air. It does not know recent amendments if they came after its training. It cannot see your client or the full circumstances. So the role is simple: AI prepares the draft and clears the routine; the lawyer reviews, refines, and takes the responsibility. The whole article rests on that rule.

Section 02Documents: contracts, demand letters, complaints

Documents are half of a lawyer's work and the biggest time sink. This is where AI pays off fastest. Here are the tasks in order, with prompts you can copy and adjust to your own practice.

Task 1. A contract draft from your inputs

You dictate the substance of the deal, and AI builds the structure and the first version. What it saves is less the typing and more the "remembering": which sections you need, what you cannot forget, and where the risks usually hide for your side.

You help a lawyer in the US draft a contract.
Governing law: general US contract law, no
citations to specific statutes.

Draft a services agreement between a company
(client) and a sole proprietor (contractor).
Scope: website development. Term: 45 days.
Fixed fee, paid 50% upfront and 50% on
acceptance. IP transfers to the client on
full payment.

Structure: parties, scope, rights and duties,
fee and payment, timeline, liability, IP,
force majeure, dispute resolution, termination,
miscellaneous.

At the end, as a separate list, flag the points
I as the lawyer must review and confirm with the
client. Do not invent statute numbers; if you
reference a rule, mark it [verify].

Notice the last two lines of the prompt. Asking it to flag the open points and not invent statute numbers is your insurance. AI will highlight where to look first, on its own.

Task 2. Reviewing the other side's contract for risk

The counterparty sends their template, and it needs a read-through with your side's eyes. In a minute AI finds the one-sided terms, the vague language, and the clauses that hurt your client. Then you check its findings and add what it missed.

Read the contract below through the contractor's
eyes. Find the clauses that are unfavorable or
dangerous for the contractor: unilateral
termination by the client, uncapped penalties,
vague acceptance criteria, hidden deadlines,
IP transfer before payment.

For each clause: the clause number, the risk,
and how to rewrite it in the contractor's favor.
Separately: what the contract is missing.

Contract text:
[paste the anonymized text]

I have watched this kind of review in practice: what used to take an hour of focused reading, AI highlights in a couple of minutes. The lawyer spends the time on the decision, the searching is already done. Double-check the findings; the final word is yours.

Task 3. A demand letter and a reply to one

A demand letter follows a clear shape: the facts, the breach, the demand, the deadline, the consequences. AI holds that shape in mind and writes clean, businesslike text with no filler. You supply the facts, it supplies the frame.

Draft a pre-suit demand letter.
From: a sole proprietor (buyer of services).
To: a company (service provider).
Substance: the provider missed the delivery
deadline by 30 days, the work was not accepted,
a $1,500 retainer was paid.
Demand: refund the retainer and pay the late
fee set in the contract.
Give a response deadline of 10 calendar days.

Tone: businesslike, no threats. Structure: facts,
breach, demand referencing the contract, deadline,
consequences of refusal. Mark any amounts or rules
that need checking as [verify].

Task 4. A draft complaint and answer

When the demand letter does not land, AI assembles the shell of a complaint: caption, statement of facts, legal grounds, requested relief, exhibits. It will not build your legal theory, but it will put the facts in the right order and propose a structure. In the same way it drafts an answer to someone else's complaint when you are on the defense side.

Task 5. Internal documents and policies

A privacy policy, a data-processing consent, a job description, a confidentiality policy, an internal procedure. These are standard documents that AI assembles from your inputs quickly and cleanly. The risk of error here is lower than in court papers, and the time savings are large.

Task 6. Translating legalese into plain English

A client reads the contract and understands half of it. AI rewrites the heavy clause in plain words, and the client sees what they are paying for and where the risk sits. That removes half the distrust before anyone signs.

Explain this contract clause in plain English,
as if you were telling a client with no legal
background. What it means in practice, what the
client risks, what to watch out for. No jargon.

Clause: [paste the clause text]

Section 03Client work and consultations

The second layer of the work is talking to people. Here AI is not about replacing your judgment. It lightens the prep and the routine correspondence. You walk into a consultation prepared, instead of burning time untangling a client's messy messages.

Task 7. Working up a client's situation before the consult

A client sends ten voice memos and three screenshots of a thread. AI builds structure out of that chaos: what happened, who the parties are, the dates and amounts, where the tight spots are. You sit down for the consult already seeing the picture.

The client described the situation in a jumble.
Build a structure for a legal consultation:
1. Parties and their roles
2. Timeline with dates
3. Key facts and amounts
4. The core of the dispute
5. Which documents are mentioned
6. A list of questions I should ask the client
   to close the gaps

Do not give a legal assessment - I will do that.
Client's text: [paste the anonymized text]

That last line matters. Keep the legal characterization for yourself; give AI the collecting and the structuring. That way you do not risk it steering you sideways with an invented rule.

Task 8. Question lists and document checklists

For common matters, AI prepares a checklist: which documents to ask the client for, which questions to ask so nothing slips. Build one such checklist per category of matter once, and prep for a meeting speeds up several times over.

Task 9. Template answers to common client questions

Half of the inbound messages are the same questions. How to terminate a contract, what to do about a missed deadline, which documents a deal needs. AI writes draft answers in your tone, you review them once, and after that you use them as templates in your direct messages.

Task 10. A case summary and timeline

On a long matter, by the third hearing your head is a mess. From your notes and documents, AI assembles a short summary: where we are now, what is done, what comes next. Handy before a hearing, and handy when a matter is handed to a colleague.

Section 04Content and growing your practice

Clients reach a lawyer through two doors: referrals and content. Referrals are self-explanatory; content is what most lawyers skip, because "there is no time to write." AI closes that objection: a draft post is born in five minutes, and all you add is your experience and a living detail.

Task 11. Posts and case studies for your blog

A lawyer's strongest content is a walk-through of a real situation with no names. A client got into trouble, you solved it, the reader recognizes themselves and writes to you. AI helps turn the matter into a post: the setup, the problem, what you did, how it ended, the takeaway for the reader.

Write a post for a lawyer's Telegram channel from
a case. Audience: business owners and individuals
with no legal background. Tone: calm, confident,
no scare tactics and no jargon.

Case: a client bought a home; six months later the
prior owner surfaced and challenged the sale. We
sorted it out and held the sale. What helped:
checking the property's history before the purchase.

Structure: a hook, the situation, what was at stake,
what we did, how it ended, a short takeaway, and a
soft call to reach out. No invented details - where
you need a fact from me, leave a blank [___].

From there it is worth running the draft through a cleanup to strip the machine phrasing, so the text reads in a living voice. I have a separate walk-through on that: how to remove AI traces from text. The voice of your practice should be yours, not an average.

Task 12. Newsletter, website, and consultation scripts

AI assembles emails for your client list, website copy for your firm aimed at search queries, and a script for the first paid consultation. On emails that get read to the end, I go deep in the piece on writing a newsletter with AI - the mechanics for a lawyer are the same. For the website, the "common client question, then a full answer" pairing works: it is useful to the reader and it brings search traffic.

Section 05The prompt formula for lawyers, before and after

The quality of AI's answer rests 80% on how you set the task. For a lawyer, a prompt stands on five parts. Put them together and you get a document, not a pile of generalities.

Here is how it looks in practice. Compare two prompts for the same task.

The formula is simple, but worth drilling until it is automatic. A breakdown with before-and-after examples for any task is in my piece on how to write prompts - it is useful well beyond legal text.

Section 06Attorney-client privilege and client data

This is the most important section, and you cannot skip it. When you paste text into a cloud AI, it travels to a third party's server. For a lawyer that is a high-risk zone: attorney-client privilege and client confidentiality both demand that a client's information does not leak.

What not to upload into a cloud AI in the clear:

• names of clients and counterparties;
• Social Security numbers, tax IDs, addresses, phone numbers;
• company account and bank details;
• case numbers, court identifiers, and other unique markers;
• trade secrets and internal documents where a contract forbids disclosure.

Anonymizing is your baseline habit

The fix is simple and becomes a habit fast. Before uploading, replace everything personal with generic labels: "Client A," "Company B," "City N," "amount X." To draft a contract, AI does not care whether it is Smith or Jones, it cares about the structure of the deal. You put the real data back into the finished draft yourself, offline.

Anonymizing before upload:

Before: John A. Smith, SSN 123-45-6789,
        Acme LLC, EIN 82-1234567, lease at
        10 Main St, amount $3,000/mo
After:  Tenant (individual), Landlord (LLC),
        space downtown, amount $3,000/mo

Extra measures

• In your account settings, turn off using your conversations for training where you can. Paid and business tiers usually offer it.
• For sensitive data, prefer tools that run on your own machine without sending anything to the cloud, or a business tier with a no-training agreement in place.
• Do not keep a matter's correspondence in the AI chat like an archive. Save what you need on your side, clear the rest from the thread.

Section 07Why AI invents laws

A story that went around the profession: in Mata v. Avianca (2023), lawyers filed a brief citing cases that did not exist, cases ChatGPT had "found" for them. The text looked convincing, the citations were in place, and the cases were pure fiction. The result was sanctions and a public black eye. A lesson for everyone who uses AI in law.

The reason is in the nature of the technology. AI does not look a rule up in a database of laws. It predicts what text plausibly comes next, based on what it saw in training. So it confidently produces the plausible: it looks like a statute, it looks like a case number, and inside it is invention. That is a property, not a bug.

A word on freshness. The model knows the law as of its training. Amendments passed later do not exist for it. If you are working with recent changes, AI is unreliable there, and you cannot lean on its memory of dates. Check the current version yourself.

Section 08Which AI to choose

Different models behave differently on a lawyer's tasks. Here is the short version of where each is stronger.

• Claude - careful with long documents and reasoning, drifts into invention less often, holds the structure of a contract and a risk review well. My main tool for complex text.
• ChatGPT - a versatile all-rounder, strong on content and quick drafts, with a familiar interface and image generation.

A practical setup: run drafting and review on anonymized data in Claude, and lean on ChatGPT for content and fast first passes. For long legal text and careful review, Claude is the pick; you can register at claude.ai. Both have free tiers that cover the first tasks, so you can test the pairing before you pay for anything. If you want a fuller comparison, I lay it out in Claude vs ChatGPT.

Section 09Your first working day with AI

Do not try to adopt everything at once. Take one task and carry it to a result. From there it rolls on its own.

Section 105 mistakes that make AI let you down

1. Trusting citations blindly. The most expensive mistake. AI confidently cites a case that does not exist. Verify every authority in an official source before you file.
2. Uploading client data as-is. Names, SSNs, account details in the cloud is a direct risk to privilege. Anonymize before uploading, every time.
3. Expecting a finished document with no edits. AI gives a draft, not a final. Whoever sends its text unread will, sooner or later, send an invention. Editing is mandatory.
4. Asking for "a legal opinion." A generic ask gets a generic, often wrong answer. Set a concrete task with facts and a format, and you get something usable.
5. Keeping every matter in one chat. The context tangles, and AI mixes the circumstances of different matters. One matter, one thread or one project.

RecapThe whole path in a minute

AI for a lawyer covers three layers of work. Documents: drafts of contracts, demand letters, and complaints, a risk review of the other side's text, a plain-English translation. Client work: working up a client's situation, question checklists, template answers to common asks, case summaries. Content: posts, case studies, a newsletter, website copy.

It runs on one rule: AI prepares the first version in minutes, the lawyer reviews and takes responsibility. Hold two insurances always - anonymize client data, and verify every citation in an official source. Build the prompt on five parts: role, facts, task, format, limits. Start with one task today, collect a library of prompts that work, and within a week AI hands back the evenings that used to go to routine.