ChatGPT, Claude and Gemini: which version can hold a client's confidences

The chart

How much risk each version puts on what you type into it.

"I pay for it" doesn't mean "it's the business version"

This is where most solo and small-firm lawyers get caught. A $20-a-month subscription buys more usage and better models. It doesn't change the contract. ChatGPT Plus, Claude Pro and Google AI Pro are all governed by the same consumer terms as the free tiers. Anthropic says it plainly: Free, Pro and Max are under its Consumer Terms, and Team, Enterprise and the API are under its Commercial Terms.

It's the same split as Gmail. A firm on Google Workspace has a business agreement with Google. A lawyer running a practice out of a personal gmail.com address doesn't, even if they pay for extra storage.

The same test applies to your email and your files

Most lawyers who worry about AI already put client files in places that raise the same question. A police report emailed from a personal Gmail account, or a discovery folder in a personal Dropbox, sits on consumer terms too. Ask the same thing of all of them: is there a business agreement that says the company handles your data only to run the service for you? If your email runs on Google Workspace or Microsoft 365, you already answered yes once, for email. A business AI plan is the same decision made again.

Is ChatGPT Plus with training turned off enough?

It fixes one problem and leaves the rest. Turning training off stops OpenAI from using your future chats to train its models. You're still on consumer terms, and there's still no data processing agreement. Your chats still sit in your account until you delete them. Deleted chats take up to 30 days to clear, and OpenAI can keep them longer "for security or legal obligations."

That last exception is not theoretical. In May 2025, a federal court in the New York Times copyright case ordered OpenAI to keep ChatGPT conversations it would otherwise have deleted. The order covered Free, Plus, Pro and Team (now Business). ChatGPT Enterprise and Edu were exempt. The order ended September 26, 2025, but OpenAI says it still holds the conversations it captured from that period, under legal hold.

The other two have their own catches:

Claude: if training is left on, Anthropic keeps your chats for up to 5 years.
Gemini: people at Google read a sample of conversations on personal accounts. A reviewed conversation is kept for up to 3 years, and deleting your activity doesn't remove it.

So the practical answer to the question: use a personal plan, with training off, for work that has no client information in it. That covers research questions, a draft of a blog post, or a motion template you'll fill in yourself. Anything with a client's name, facts or documents in it goes in a business plan.

What it costs a solo

The business versions are sold to teams, so a one-lawyer practice pays a little extra for the protection.

ChatGPT Business and Claude Team both require 2 seats. At $25 a seat that's $50 a month for one person, or $40 billed yearly. Plus and Pro are $20.
Gemini in Google Workspace is the cheapest route if your email already runs on Workspace. It's part of Business Standard, $14 a month for one person, which you may already be paying. Business Starter, at $7, only gets Gemini inside Gmail.

Pick one. You don't need all three.

How to check which version you're on

ChatGPT

Your plan: click your profile icon, then Settings, then Account.
What counts: Business and Enterprise are the business versions. Free, Plus and Pro are personal plans, whatever you pay. A solo can buy Business, but it has a 2-seat minimum.
Turn training off on a personal plan: Settings, then Data controls, then switch off "Improve the model for everyone." On the phone app it's in the same place, under your profile icon.
Temporary Chat isn't saved to your history and isn't used for training, but OpenAI may keep a copy for up to 30 days.

Claude

Your plan: click your name in the lower left, then Settings, then Billing.
What counts: Team and Enterprise are the business versions. Free, Pro and Max are personal plans. A solo can buy Team, with a 2-seat minimum.
Turn training off on a personal plan: Settings, then Privacy, then switch off "Help improve our AI models." It stops training on future chats only. Anything already used stays used.
Incognito chats aren't used for training on any plan, and are kept for 30 days.

Gemini

Which account you're signed in with matters most. Check the account shown in the top right of the Gemini app. A personal Google account is on consumer terms, even if it pays for Google AI Pro.
Gemini in Workspace: you're signed in with your Workspace account, the one on your firm's own domain, not a gmail.com address. Your plan shows in the Google Workspace admin console under Billing.
Turn training off on a personal account: go to Gemini Apps Activity (myactivity.google.com/product/gemini) and turn off Keep Activity. Google still keeps new chats for up to 72 hours.

Before you upload a client's file

You're on a business version, not a personal plan you happen to pay for.
The data processing agreement is part of the terms you accepted. For ChatGPT Business, Claude Team and Gemini in Workspace it comes with the commercial terms. You don't have to ask for it.
On Enterprise plans, set how long chats are kept. Claude Enterprise with no retention setting keeps chats indefinitely.
You don't use thumbs up or thumbs down on that chat. Anthropic keeps a rated conversation for up to 5 years, on any plan.
You've read the rules below and decided whether the client should be told.

What about legal AI tools?

Most legal AI products run on these same companies' APIs, which fall under business terms. Ask any vendor three things: which AI company processes your data, whether it's under that company's commercial terms with training off, and what agreement the vendor has with you.

Plani's answer: Anthropic's API, commercial terms, no training, and our own terms with you on top.

What the rules say

No Massachusetts ethics opinion addresses generative AI yet. In December 2025 the SJC's rules committee looked at whether the rules needed changing for AI and said no, because the existing rules "provide ample guidance." Here is that guidance.

Rule 1.6(c) requires "reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to" client information. Comment 18 weighs the sensitivity of the information, the likelihood of disclosure, and the cost and difficulty of better safeguards.
Rule 1.1, comment 8 expects you to keep up with "the benefits and risks associated with relevant technology."
Rule 5.3, comment 3 covers outside services that hold client information. You must make reasonable efforts to see that the service is "compatible with the lawyer's professional obligations," which turns partly on the contract terms protecting that information.
MBA Ethics Opinion 12-03 (2012) is the closest Massachusetts analog. It allowed cloud storage if the provider's terms bar access "by the provider itself for any purpose other than" serving you. It also said to get the client's express consent before putting "particularly sensitive" information online. A criminal defense file usually qualifies.
ABA Formal Opinion 512 (July 2024) isn't binding here but is the most detailed. Lawyers should "read and understand the Terms of Use, privacy policy, and related contractual terms" of any AI tool. Before putting client information into a self-learning tool, one that may use your input to train, the client's informed consent is required. Boilerplate in an engagement letter isn't enough.

Bar Counsel's own article on AI puts it in one line: "don't enter confidential information when you are using a non-private system."

Practicing in Maine? Opinion 207 (2013) on cloud computing still stands, and it also asks for a careful read of the vendor's terms and security policy.

This is a guide to the terms these companies publish, not legal advice. The terms change often. Check the pages linked below before relying on any line of this chart.