Using AI in Your Practice: A Beginner’s Guide for California Lawyers

Most of the writing about AI and law is either a sales pitch or a scolding. This is neither. It is a starting guide for a California lawyer who has not really used these tools yet, wants to know whether they are worth the trouble, and does not want to end up in a published opinion as a cautionary tale.

Short version: they are useful, they are not what the marketing says, and the ethics rules that govern them are the ones you already know.

What these tools actually are

A generative AI tool predicts likely text. That is the whole trick. It has read an enormous amount of writing and it is very good at producing the next plausible words.

That one fact explains most of what you need to know:

• It is genuinely good at working with text you already have. Searching it, summarizing it, reorganizing it, tightening it.
• It is genuinely bad at knowing things. It is not checking a database of California law. When it hands you a citation, it is producing a plausible-looking citation, which is not the same thing as a real one.

Treat it as a fast, tireless, slightly overconfident first-year associate who never admits uncertainty. You would not file that associate’s work without reading it. Same rule here.

Play with it before you put it near a case

This is a skill, and like any skill you will be bad at it at first. That is not a sign the tool is useless. It is a sign you have used it for about an hour.

So do your learning somewhere that does not matter. Plan a trip with it. Have it help with a birthday dinner menu, a fantasy draft, a letter to your kid’s school. Ask it to explain something you already understand well, so you can see where it goes wrong. You are building an instinct for what it is good at, and you cannot build that on a client matter.

There are also more free step-by-step guides than anyone could watch. Search the tool’s name plus what you want to do on YouTube and you will find someone walking through it. The resources are not the bottleneck. The practice is.

How do you get to Carnegie Hall? Practice. Same answer here, and it is not a very long walk. Most lawyers are useful with these tools inside a couple of weeks of regular use.

Which tool

My recommendation is Claude. But the honest answer is that the best tool is the one you will actually use, and the differences between the leading products matter far less than whether you have built the habit.

Try two of them before you commit. Run the same task through both and see which one you like arguing with. Nobody is paying for that recommendation, and there is no affiliate link on this page.

Where to start

Two uses earn their keep immediately. Both work on material you already have, so there is nothing to go verify in a reporter.

1. Email, as a very good spell check. Write the email yourself. All of it. That keeps you sharp and it keeps the writing in your voice, which matters more than people think once you have read a few pieces of obviously machine-written correspondence.

   Then, on the ones that actually matter, run your draft through and ask for options on how to respond. Not “write this for me.” More like a second read from someone who is never too busy. You will take maybe one suggestion in three, and that one is often worth the trouble.

2. Searching discovery. This is where the time savings are real. Point it at a document set and it will find things quickly and well.

   But add one step, and do not skip it: make it tell you where it got the answer. Ask for the document, the page, the paragraph. Then go look. An answer without a source is a starting point, not a finding, and the whole value collapses if you cannot check the work.

After those two, the natural next steps are summarizing a document you already know well (so you can immediately see where it drifts), cleaning up your own outline, and asking it to argue the other side of your position. That last one is free and underrated.

Notice what is not on the list: legal research you intend to rely on, and confidential client information in a consumer tool. Those come later, and with setup.

The confidentiality problem, in practical terms

This is the part people get wrong first.

When you type into a consumer AI product, that text goes to a third party’s servers. Depending on the product and the settings, it may be retained, may be reviewed by a human, and may be used to train the model. The rule does not care that it felt like typing into a search box.

And the duty it points to is about as absolute as California gets:

Before you put anything client related into any tool, answer three questions:

• Does this vendor train on my inputs? Consumer tiers often do by default. Business and enterprise tiers often do not.
• How long is my data retained, and can I turn retention off?
• Who at the vendor can see it, and what is their security posture?

The State Bar’s guidance does not treat this as optional. A lawyer must not put confidential client information into a tool that lacks adequate confidentiality and security protection, and should get IT or cybersecurity help evaluating a platform before client data goes near it.

One change worth knowing. The 2023 version of the guidance told lawyers to anonymize client information as a general instruction. The 2026 revision removed that requirement. That is not permission to be careless. It shifts the analysis back onto your own Rule 1.6 duty and onto what the specific tool actually does with your data, which is a harder question than running a find and replace on your client’s name.

So the safe practice while you are learning is simple: use it on your own work product and on public information, and keep client identifiers out. And treat anonymizing as necessary rather than sufficient. A distinctive set of facts can identify a client as surely as a name does.

Read more: vendor diligence, the longer version

What to actually look for before a tool touches client data:

• Training rights. Read the terms. “We do not train on your data” and “we do not retain your data” are two different promises, and a vendor can truthfully make the first while doing the second.
• Retention controls. Can you set retention to zero, or to a short window? Is that setting available on your tier, or only on enterprise?
• Human review. Many providers reserve the right to have staff review flagged conversations. Find out what triggers that.
• Security posture. SOC 2 Type II reporting, encryption in transit and at rest, access controls, subprocessor lists, and where the data physically sits.
• Contract terms. An enterprise or business agreement often disclaims training rights and adds confidentiality obligations that the consumer terms of service do not.

The one most firms miss: your existing vendors are quietly adding AI features. Practice management, email, document management, e-discovery. Some of it is switched on by default. That means you may already be sending client data to an AI system without ever having made a decision about it. Go look at what has been enabled for you.

The State Bar’s guidance is direct on this point: consult IT or cybersecurity help if you are not equipped to evaluate a platform yourself, and confirm the provider is not sharing your inputs with third parties or using them to improve its own product.

The ethics rules that apply

There is no separate AI rulebook. The duties you already have do all the work.

Competence

Technology is expressly part of that. California added Comment [1] to Rule 1.1 on March 22, 2021:

Read that twice. The benefits and the risks. A lawyer who refuses to learn what these tools do sits under the same rule as the lawyer who uses one carelessly.

The State Bar’s guidance puts real content behind this. Before you use one of these tools, you are expected to understand at a reasonable level how it works, what its limits are, and what the terms of use say about your data. Output is a starting point, not a work product: you are expected to review, validate, and correct both what you put in and what comes out. The guidance is blunt that leaning on these tools too heavily is not consistent with actually practicing law, and that your professional judgment is not something you can hand to a machine. It also makes clear that competence takes more than catching the obviously false results.

You are responsible for the output. Not the vendor, not the model. If you file it, you wrote it.

Candor toward the tribunal

This is the sanctions rule. Note the second half: the duty to correct. If you discover after filing that a citation was fabricated, the rule points hard toward telling the court promptly.

Rule 3.3(a)(2) matters here too. It reaches knowingly misquoting a case, statute, or other authority, and it requires you to disclose directly adverse controlling authority the other side has not cited. That is worth remembering because fabrication is not the only failure mode. A real case, cited for something it does not actually say, is the quieter version of the same problem, and a tool connected to a real database can still produce it.

The State Bar’s guidance is explicit that you review AI output, including the analysis and every citation, before it goes to a court, and that you correct errors already made. It also tells you to check whether the jurisdiction, court, or department has its own AI disclosure requirement.

Talking to the client about it

There is no categorical duty to announce to every client that you used AI. The guidance frames it as a judgment call driven by the circumstances: how novel the technology is, what the risks are in this matter, the scope of the representation, and how sophisticated the client is. It says you should consider telling the client you intend to use it, and explaining how, along with the benefits and risks.

Two practical additions. Institutional clients increasingly have their own AI policies, and those bind you, so read the outside counsel guidelines before you assume anything. And some courts have standing orders on AI use in filings, so check the department as well as the client.

Supervision

Your paralegal is probably already using these tools. If you have not asked, you do not know. That is a supervision gap, not a staff problem. Rule 5.1 puts a parallel duty on managerial lawyers.

Fees

This is the area where the guidance changed most, so be careful with anything you read that predates May 2026:

• The flat prohibition is gone. The 2023 guidance contained a bare sentence saying a lawyer must not charge hourly fees for time saved by using generative AI. That sentence was deleted in the 2026 revision. If you have seen it quoted recently, the source is quoting a depublished document.
• You may bill for the time you actually spend working with the tool. Writing and refining prompts, reviewing output, correcting it, and turning it into something usable is real work, and it is billable like any other.
• Cost pass-through narrowed. The 2026 revision is tighter than the 2023 version about charging clients for the tools themselves, not looser. Do not assume a subscription can simply be passed through. If you intend to charge for AI tooling, the basis belongs in the engagement letter and it should be a deliberate decision.

None of that makes the underlying problem go away. Rule 1.5(a) still prohibits an unconscionable fee, and billing a client for hours you did not work is a Rule 1.5 problem whether or not a guidance document spells it out. The deletion changed the guidance, not the rule.

The habit that matters more than any tool choice

Pick whatever tool you like. The habit is the same and it is not complicated:

Nothing reaches a client, an opponent, or a court that you have not read and verified yourself.

That is the whole discipline. Every AI horror story in the legal press is a story about somebody skipping that step. Not a bad model. Not a bad prompt. A lawyer who did not read the thing before it went out.

Where to go next

Once you are using these tools day to day, the companion piece goes into the failure modes in detail, walks through the six rule amendments now pending, and covers how to actually run citation checking: Where AI Helps in a California Practice, and Where It Will Burn You.

The rest of what I have written for lawyers, rather than for clients, sits in the for attorneys section.

This is general information for other lawyers. It is not legal advice, not an ethics opinion, and not a substitute for your own reading of the Rules of Professional Conduct and the State Bar’s guidance. The rule amendments described above were proposals as of July 2026 and had not been adopted. Verify anything here before you rely on it.