AI for lawyers: a practitioner's map

Most of what is sold as AI for lawyers is about research and drafting. The part that eats a partner's week is different: remembering what every client said, and walking into each call current. Below: the three kinds of tool you will meet, what the bar guidance asks you to check before using any of them, what a law firm's AI policy covers, how the ChatGPT training setting works, and where the client side fits.

AI in a law practice does three separate jobs, and most guides only cover the first. There are research and drafting platforms built for legal work. There are general assistants (ChatGPT, Claude, Gemini) that lawyers use for thinking, rewording and summarising. And there is the client-relationship side of practice, which most legal AI is not built for: the settlement position a client mentioned on a Tuesday call, the reply that arrived while you were in court, the three things you promised to send before the next meeting.

One disclosure before going further. This guide is published by Equerry, a personal assistant for that third job. Equerry is not a legal research or drafting tool. It does not search case law, draft pleadings or review contracts, and nothing on this page is legal advice. The first two-thirds of the page cover the whole landscape on its own terms; Equerry comes in at the end, where it fits.

The three kinds of AI a lawyer runs into

Legal research and drafting platforms

This is the category most "AI for lawyers" lists are about, and the vendors describe it in their own words. Harvey's platform page (read 2026-09-23) offers to "Research complex legal, regulatory, and tax questions across domains" and to "Securely store, organize, and bulk-analyze legal documents." Thomson Reuters says of CoCounsel Legal (read 2026-09-23): "From a single query, it handles research, analysis, and drafting, developing the right legal approach, working through each step, and delivering polished, cited work product — no prompt engineering required." These are named as examples of the category, not as recommendations; there are others, and firms usually choose them through procurement rather than one lawyer at a time. The ABA's own list of what these tools can help with is broad: "legal research, contract review, due diligence, document review, regulatory compliance, and drafting letters, contracts, briefs, and other legal documents."

Whatever the platform, the verification duty does not move. ABA Formal Opinion 512 lists the problems seen so far as "citations to nonexistent opinions, inaccurate analysis of authority, and use of misleading arguments," and says "lawyers should review for accuracy all GAI outputs."

General assistants

ChatGPT, Claude and Gemini are general-purpose assistants rather than legal platforms (some now add legal features: Anthropic offers a Legal plugin for Claude, which it describes as built "for in-house legal teams," checked 2026-09-23), and many lawyers use them for work that doesn't involve client information: outlining a talk, explaining a regulatory process in plain English, restructuring a paragraph. The questions they raise are mostly about inputs. What happens to what you type depends on the plan you use and the settings you choose, and the safe assumption is that the account settings, not the brand, decide the answer. The ChatGPT settings are covered step by step below.

The client-relationship side

Every practitioner carries a running record in their head: who said what on which call, where each negotiation stands, what the client's CFO asked for in the last email. It is the least glamorous part of practice and the most expensive to get wrong, and it is where "AI for lawyers" coverage usually stops. More on it below.

What bar guidance asks you to check

Two documents are worth reading in full before adopting any AI tool, and both are short enough to read in one sitting. ABA Formal Opinion 512, issued July 29, 2024, is the ABA's opinion on generative AI tools. It opens by saying lawyers "must fully consider their applicable ethical obligations, including their duties to provide competent legal representation, to protect client information, to communicate with clients, to supervise their employees and agents, to advance only meritorious claims and contentions, to ensure candor toward the tribunal, and to charge reasonable fees." The State Bar of California's Practical Guidance was revised in 2026 and now covers agentic tools as well. Your state's rules, any guidance it has issued and your firm's policy are what apply to you; what follows is a reading list, not advice.

The checks that come up across the two documents:

• Understand the tool you use. Opinion 512: lawyers "must have a reasonable understanding of the capabilities and limitations of the specific GAI technology that the lawyer might use." California adds that this includes "periodic reassessment" because systems change.
• Know what happens to what you put in. Before inputting information relating to a representation, Opinion 512 says lawyers "must evaluate the risks that the information will be disclosed to or accessed by others outside the firm." California says reasonable efforts "require more than reliance on generalized marketing assurances" and "may include reviewing terms of use, privacy policies, or vendor documentation."
• Consent, when it applies, has to be informed. For self-learning tools, Opinion 512 says "a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool," and that "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient." It also notes uses where consent is not needed because no client information goes in, such as idea generation.
• Limit what a tool can reach. California's 2026 guidance singles out tools "configured to access internal firm systems, such as email, messaging platforms, document management systems, knowledge bases, client files, or calendaring systems," and says "Lawyers must carefully evaluate and limit the scope of such access."
• Verify, and keep the judgment. Opinion 512: AI tools "cannot replace the judgment and experience necessary for lawyers to competently advise clients." California: "A lawyer's professional judgment cannot be delegated to AI."
• Supervise and set policy. Opinion 512: "Managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI, and supervisory lawyers must make reasonable efforts to ensure that the firm's lawyers and nonlawyers comply with their professional obligations when using GAI tools." More on this in the law-firm section below.
• Mind the fee. Both documents address billing for AI-assisted work; California says hourly billing "must reflect the time actually spent on the client's matter."

For law firms: who decides, and what a firm AI policy covers

If you came here looking for AI for law firms rather than for one lawyer, both documents answer the first question the same way: the decision sits with the people who run the firm. ABA Formal Opinion 512, quoted in the checklist above, puts the firm's AI policy on its managerial lawyers and reasonable efforts to ensure compliance on its supervisory lawyers. California's 2026 guidance, written for California lawyers, says: "Managerial and supervisory lawyers should establish clear policies regarding the uses of generative AI, including more autonomous or agentic AI tools, and make reasonable efforts to provide reasonable assurance that the conduct of the lawyers and nonlawyers under supervision complies with the lawyers' professional obligations when using such technologies." A firm policy does not take the question away from the individual lawyer, though. California also says: "A subordinate lawyer must not use AI-enabled tools, including agentic AI systems, at the direction of a supervisory lawyer in a manner that violates the subordinate lawyer's professional responsibility and obligations and must exercise independent professional judgment in evaluating whether such use is consistent with those obligations."

Neither document supplies a policy template. Read together, they point to topics a firm's AI policy could cover:

• What is permitted. Which tools, and which uses of them, the firm allows. That is the "permissible use" the ABA opinion refers to.
• Client information, consent and verification. The checks in the previous section, set once for the whole firm rather than decided afresh by each lawyer: what may go into which tool, when informed consent is needed, and how output is checked before anyone relies on it.
• What each tool can reach. California's point about tools connected to email, document management, client files and calendars, quoted above, is a firm-level decision as much as a personal one.
• Training. Opinion 512: "Supervisory obligations also include ensuring that subordinate lawyers and nonlawyers are trained, including in the ethical and practical use of the GAI tools relevant to their work as well as on risks associated with relevant GAI use." It adds: "Training could include the basics of GAI technology, the capabilities and limitations of the tools, ethical issues in use of GAI and best practices for secure data handling, privacy, and confidentiality." California, on managerial and supervisory lawyers' duties: "This includes providing appropriate training on the ethical and practical aspects, and pitfalls, of any AI-enabled tools that perform tasks with limited or no real-time human direction." And "…lawyers should ensure that their lawyer and nonlawyer staff receive periodic training on the appropriate use of AI and safeguards surrounding the use of AI."
• Keeping it current. California: "As AI technologies continue to evolve, managerial and supervisory lawyers should review and update applicable firm policies, controls, and associated training accordingly to address new risks and use cases, including those that may reasonably be anticipated as AI systems become more autonomous or are integrated into legal workflows."
• Bias. California: "Firms should establish policies and mechanisms to identify and address potential AI biases."
• Billing. How AI-assisted work is charged, which both documents address.

For a practitioner, a sensible first step is the firm's own policy; if there isn't one, whoever runs the firm or its general counsel is the person to ask before putting client work into any tool, the ones on this page included. Which platforms a firm licenses is a procurement decision this guide does not cover, and how the ABA opinion and California's guidance apply in your jurisdiction is a question for your firm and your state bar, not for this page.

If you use ChatGPT: the training setting, step by step

On personal ChatGPT plans, whether your conversations can be used to train OpenAI's models is controlled by a setting. OpenAI's help center (read 2026-09-23) says: "When Improve the model for everyone is off, your new conversations won't be used to train OpenAI models. They can still appear in your chat history." To find it:

• On the web: open your account menu, select Settings, select Data controls, then turn off Improve the model for everyone and select Done.
• On iPhone and Android: open the sidebar, select your profile icon to open Settings, select Data controls, then turn off Improve the model for everyone.

Four things the same article says that are easy to miss. The setting covers new conversations. "Turning off Improve the model for everyone does not delete or hide saved chats." If you give a thumbs up or thumbs down on a response after opting out, "the entire conversation associated with that feedback may be used to train OpenAI models." And Temporary Chats are not used to improve models but "may be retained for up to 30 days for safety purposes." For firm accounts, OpenAI states that "By default, OpenAI does not use content from ChatGPT Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces to train its models," with the organization's own retention and access policies on top.

Note what the setting does not answer. Training is one of the questions the guidance raises; who else can access the information, how long it is kept and whether client consent is needed are the others. Other assistants publish their own equivalents, and the same questions apply to each.

The part nobody covers: remembering what every client said

Research tools answer questions about the law. The question that fills a partner's week is usually about the client: What did she say about settlement on the last call? Did the other side's counsel reply to the draft? What did I promise to send before Thursday? The answer is scattered across a phone call nobody wrote up, a thread in your inbox, a PDF the client attached and a note on your phone.

You can keep that record well without any AI at all, and most good practitioners already do some version of this:

• Write the post-call note in the first five minutes. Four lines: what was decided, what the client asked for, what you owe and by when, and the one sentence you will want to quote accurately later. Our guide to keeping a record of client calls covers this in detail.
• Send the recap email. A two-line summary to the client after an important call puts the record where both of you can find it, and invites a correction while memories are fresh.
• File it with the matter, the same day. A note in the matter file beats a note in the wrong app. The system matters less than doing it the same way every time.
• Brief yourself before the next call. Ten minutes with the last note, the last two emails and the open items. Our client-meeting prep guide has a checklist.

The weak point is always retrieval. The note exists, but finding it four months later means remembering where you put it, and the email that changed the position arrived after you wrote it.

Where Equerry fits: a personal assistant for the client record

Equerry is a personal assistant who helps you get things done, and for a lawyer that means the third job on the map, not the first two. It does not research, draft or give advice. It keeps track of what you share with it and briefs you from it, the way a good assistant keeps the file straight.

• Debrief the call by voice. Walking out of a client call, dictate a sixty-second voice note in Equerry: what was decided, what the client said about settlement, what you owe. It transcribes it and remembers it. Equerry does not join or record your calls. The voice note is yours, made after the call.
• CC it on the correspondence that matters. Equerry has an address of its own at mail.getequerry.app. Forward it a thread, or add it to the CC line of a live one, and for as long as it stays on the thread each reply arrives on its own. It never asks for access to your inbox; the original stays in your own mailbox.
• Save the documents a client sends. Share a PDF, a photo of a marked-up page or a screenshot from your iPhone's share sheet, or type a note.
• Ask in plain language. "What did the client say about settlement?" is answered from the voice notes, emails and notes you gave it. When an answer draws on a note you saved, it carries a source chip: tap it to see the passage Equerry used and open the original.
• Walk into the meeting briefed. If you allow calendar access, Equerry keeps a snapshot of your events from today through the next 14 days (titles, start and end times, locations and attendee display names, never event notes, meeting links or email addresses), and before a client meeting it puts together where things stand from what you have shared: the latest reply, what is open, what was agreed last time. Calendar titles often name clients, so include the calendar in the question you put to your firm's policy.

Some honest limits, in the spirit of the checklist above. Equerry only knows what you send it, plus your calendar if you allow it; it cannot find the email you never forwarded. It is not a practice-management system and doesn't connect to one. Anyone on a thread can see the Equerry address on the CC line. And forwarding or recording something is still sharing it: what you send is processed to make it searchable and answerable, including by the outside providers named on our AI disclosure page. Equerry makes no claim about privilege or about meeting any bar rule. Whether and how client material goes into any tool, this one included, is a decision for you under your firm's policy, your jurisdiction's guidance and, where it applies, your client's informed consent. What it does change is scope: the question becomes about the specific threads and notes you chose to send (and your calendar, if you allowed it), rather than a whole mailbox.

If email is where most of your client work happens, read our guide to choosing an AI email assistant next. For the broader picture of client-facing work, see Equerry for client-facing executives.

Sources, all read 2026-09-23; the ABA opinion and the California guidance were re-read 2026-09-25 for the law-firm section. American Bar Association Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024). The State Bar of California, Standing Committee on Professional Responsibility and Conduct, Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (2026 revision, replacing the November 2023 version). OpenAI Help Center, Data controls in ChatGPT. Harvey, Platform. Thomson Reuters, CoCounsel Legal. Anthropic, Legal plugin. Vendors change their products and pages often; where a vendor's current page disagrees with this one, the vendor's page is right. This guide is general information, not legal advice.

Last updated 2026-09-25.

All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.