ChatGPT for lawyers: what it is good for, and what never to paste into it

Plenty of lawyers already use ChatGPT for the first draft, the plain-English explanation, the second opinion late at night. This guide covers what it costs, where it earns its place, the court sanction that followed invented citations, what to keep out of it, OpenAI's own settings step by step, and prompts for client work.

This is the ChatGPT chapter of our AI for lawyers guide, which maps the whole landscape and the bar guidance. Here the question is narrower: you have ChatGPT, or you are about to, and you want to use it on real client work without regretting it.

Two disclosures first. This guide is published by Equerry, a personal assistant for the client record, and Equerry is not a legal research or drafting tool. It comes in at the end, where it fits. And nothing here is legal advice: what you may put into any tool is decided by your jurisdiction's rules, your firm's policy and your client, not by a vendor. Every statement about what ChatGPT does or offers comes from OpenAI's own pages, read in a browser on 29 September 2026 unless another date is given.

What ChatGPT is, and what it costs a solo lawyer

ChatGPT is OpenAI's general-purpose assistant, used on the web and in its iPhone and Android apps. It is not a legal research platform. For a solo or small-firm lawyer the plan matters less for its model than for what it does with your conversations and how much it remembers. OpenAI's pricing page shows prices in your local currency. Its help article "What is ChatGPT Plus?" gives Plus as "$20/month (billed monthly)", and "About ChatGPT Pro tiers" offers "three monthly Pro plans" at $100, $200 and $500 (both read 1 October 2026). Here is what the pricing page lists for each plan on the points that matter to a practitioner:

For a solo lawyer the real choice is usually a personal plan with training switched off, or a Business workspace where the default runs the other way; OpenAI's pricing page notes that "Business plans are available starting at 2 users" (read 29 September 2026), so it is not a one-seat option. Neither settles a professional-responsibility question on its own. If you use Claude instead of, or alongside, ChatGPT, the same ground is covered for Anthropic's assistant in Claude for lawyers. For the same questions in a different client business, see ChatGPT for real estate agents.

Where ChatGPT saves a lawyer time

The work it does well is the work where you already know the answer and want a faster first version, or where no client information needs to go in at all:

• Plain-English explanations. What happens at a mediation, what a closing involves. You check the draft; the client gets a clearer email.
• Tightening your own drafts. A status update, a fee explanation, a cover letter, with names replaced by placeholders. Ask it to list every change so you can review the edits rather than reread the whole thing.
• Preparing for a first meeting. The questions to ask about a type of matter, grouped by facts, documents and goals.
• Arguing with yourself. Describe an argument in your own words and ask for the strongest objections to it. It is a sparring partner, not a source.
• Working through public documents. Summarising a regulation or an opinion you pasted in full, with each point tied to quoted language you can check.
• Everything that isn't client work. CLE outlines, a talk, a bio.

Where it falls short: citations, and the case every lawyer has heard about

Its answers sound equally sure whether right or wrong, and in law the costliest wrong answer is a citation. In its Opinion and Order on Sanctions in Mata v. Avianca, Inc. (No. 22-cv-1461, S.D.N.Y., June 22, 2023), the court found that two lawyers and their firm "submitted non-existent judicial opinions with fake quotes and citations created by the artificial intelligence tool ChatGPT, then continued to stand by the fake opinions after judicial orders called their existence into question." It imposed a $5,000 penalty and ordered them to send the opinion to their client and to each judge falsely named as the author of a fake opinion. Its opening paragraph also says: "there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings."

The working rule is simple: never file or send a citation you have not pulled and read yourself. Treat any case name, quotation or statute section ChatGPT gives you as a lead to verify, not a finding.

What to paste into ChatGPT, and what never to paste

This is a working default, not a rule of professional conduct, and your firm's policy overrides it. The test before you paste: could this identify a client or their matter?

Usually raises fewer questions:

• Public material: statute text, a published opinion you retrieved yourself, a regulator's guidance, a court's local rules.
• Your own drafts with every name, date, amount and distinguishing fact replaced by placeholders such as [Client], [Counterparty], [Amount].
• Work with no client in it at all.

Keep out, unless your firm has decided otherwise for a specific tool and plan:

• Client names together with the facts of their matter.
• Privileged communications, and your notes of them.
• Settlement positions, deal terms and negotiation strategy.
• Personal identifiers: account numbers, health information, dates of birth, identity documents.

A separate case: anything you hold under a protective order or a confidentiality agreement, including the other side's documents. Check the terms of the order or agreement, and your firm's policy, before it goes into any tool.

One more input deserves a decision of its own: connected apps. ChatGPT can connect to email and other accounts, and OpenAI's Memory article lists "content from connected apps, such as Gmail" among the sources Memory can draw on, depending on plan and region. OpenAI's Google app FAQ says that for an individual connection "ChatGPT can access only content available to the account you connect, after you authorize access," and adds its own caution: "Google app data may contain information you might consider sensitive, so only enable the Google app if you're comfortable with ChatGPT using this data." A lawyer's mailbox is mostly other people's confidences. We cover what that connection grants, and how to take it back, in can ChatGPT read my emails.

Is ChatGPT confidential?

Whether a use is consistent with your confidentiality duties is a legal judgment about your jurisdiction, your client and the matter, so no vendor can answer it. What OpenAI can tell you is what happens to what you type, and its help center (read 29 September 2026) says this:

• Training, personal plans: "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." And turning it off "does not delete or hide saved chats."
• Feedback overrides it: after opting out, if you give a thumbs up or thumbs down, "the entire conversation associated with that feedback may be used to train OpenAI models."
• Temporary chats "do not appear in your chat history," "do not create or update memories," "are not used to improve OpenAI models" and "may be retained for up to 30 days for safety purposes," all while they remain temporary.
• Memory: "Sensitive information can appear in Memory if you share it with ChatGPT."
• Firm workspaces: no training on Business, Enterprise, Edu or ChatGPT for Healthcare content by default, and "Your organization's privacy, retention, and access policies still apply."

Those are controls over training, history and memory, set per account. None of them says a conversation is privileged or confidential in the legal sense, and this guide makes no such claim. Put them in front of whoever sets your firm's policy.

The three settings, step by step

1. Keep one conversation out of history and memory. OpenAI's steps: "Open a new chat." "Select Temporary." "Choose Personalized or Unpersonalized before you send your first message." Choose Unpersonalized: it "Does not use memory, custom instructions, or plugins." You can't change it once the chat has started, and if you save the chat it becomes a regular one that follows your account settings. Our guide to ChatGPT Temporary Chat has the details.

2. Stop new conversations being used for training. On the web: open your account menu, select Settings, then Data controls, 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 and turn it off. Signed in, the setting follows your account across devices.

3. See or switch off memory. Open Settings, select Personalization, then Memory. The controls there vary by account. If a client matter has already gone into a regular chat, removing it takes several steps; our guide on how to delete ChatGPT memory walks through them.

ChatGPT prompts for lawyers' client work

Good prompts share three habits: a narrow job, no client identifiers, and output you can check. Fill every bracket with a placeholder, not the real detail.

• Explain a process to a client. "Explain to a client with no legal background what happens at a [mediation] in a [commercial dispute]. Under 300 words, no citations. Flag anything that varies by court or state so I can confirm it."
• Tighten an update without moving a commitment. "Here is my draft update to a client, names removed: [paste]. Make it shorter and plainer. Do not change any date, number or promise. After the draft, list every change you made."
• First-meeting questions. "I'm meeting a new client about a [lease dispute]. Give me the questions to ask in the first meeting, grouped under facts, documents, deadlines and what the client wants."
• Stress-test an argument. "Here is an argument in my own words, facts anonymised: [paste]. Give me the three strongest objections the other side will raise and the weakest point in my reasoning. Do not cite any case." Then check anything it states as law anyway.
• Summarise a public document you can verify. "Summarise the [regulation] I've pasted below in ten points. For each point, quote the exact sentence it comes from."
• Turn anonymised notes into a task list. "Turn these notes into a list of next steps with who owns each one and when it is due: [paste notes with placeholders]."

None of these asks for a case to cite, a conclusion about a real client, or a document to send unread.

What ChatGPT does not hold: what each client said across email and calls

ChatGPT's memory is real, so the gap is not that it forgets. It is about control. Everything it knows about your practice arrives one of two ways: you type, paste or upload it into a chat, or you connect an account and it can reach what that account can reach. The first keeps you in charge but doesn't keep up: the reply that changes a client's position lands while you're in court. The second keeps up, but a mailbox connection is to the account, not to one thread, and for a lawyer that is exactly the question the "never paste" list above is about. And in what OpenAI calls improved Memory, which "continually updates a broader summary of relevant context," the summary you can review is, in OpenAI's words, a "high-level view" that "does not necessarily include every detail or source that Memory can reference." (Where the older saved-memories experience is still offered, it uses "a more explicit list of individual memory items.")

The record a practitioner actually needs sits in between: what did the client say about settlement on Tuesday's call, did opposing counsel answer the draft, what did I promise to send before Thursday? That is the client side of practice, covered more broadly in our practitioner's map of AI in a law practice.

Equerry is a personal assistant who helps you get things done, and for a lawyer that means that record, not the law. It does not search case law, draft pleadings or review contracts. It holds only what you choose to send it:

• CC it on the one thread. Equerry has an address of its own at mail.getequerry.app. Add it to the CC line of a live client thread, or forward a thread after the fact; replies that keep the address on the thread keep arriving. It never asks for your inbox, and the original stays in your own mailbox. Ask about it later. What Equerry remembered from it is listed in Settings → Your memory, each line marked as from that email, and you can delete any line.
• Debrief the call by voice. Walking out of a client call, record a sixty-second voice note in Equerry: what was decided, what the client said, what you owe. Equerry does not join or record your calls; the note is yours, made afterwards.
• Save the documents a client sends. Share a PDF or a screenshot from your iPhone's share sheet.
• Write it down. Type a note in Equerry after a meeting. 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.
• Ask before the next call. "Where do things stand with the lease matter?" is answered from what you sent it.

The honest limits. Equerry knows only what you send it; it cannot find the email you never forwarded. Anyone on a thread can see its address on the CC line. What you send is processed to make it searchable, including by the outside providers named on our AI disclosure page. And Equerry makes no claim about privilege or any bar rule: whether client material goes into it is the same decision as for ChatGPT, made under your firm's policy and your jurisdiction's guidance. What changes is the scope of that decision, from an account to the specific threads and notes you chose.

Keep ChatGPT for the first draft and the plain-English explanation, with the settings above. Use Equerry for remembering what each client said. It is on the App Store for iPhone and iPad.

Sources. OpenAI Help Center, read 29 September 2026 in a browser session (the help center refuses automated fetches): Data controls in ChatGPT (training setting and steps, feedback, temporary-chat retention, managed workspaces); Temporary chat in ChatGPT (steps and personalization); Memory in ChatGPT (sources, settings path, memory summary, deletion; also read 25 September 2026 with the same wording); Google app data controls FAQ (connected Google apps); What is ChatGPT Plus? (US Plus price; read 1 October 2026); About ChatGPT Pro tiers (the three Pro plans and their US prices; read 1 October 2026). OpenAI, ChatGPT pricing (plan features and the Business minimum, read 29 September 2026; prices shown in local currency). United States District Court, Southern District of New York, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions (June 22, 2023; S.D.N.Y. docket entry 54, via CourtListener's RECAP archive, read 29 September 2026). OpenAI changes its products and pages often; where its current page disagrees with this one, OpenAI's page is right. This guide is general information, not legal advice.

Last updated 2026-10-01.

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.