Iro AIIro AI Blog
Not on a personal account with the default settings, and not on any plan your firm hasn't approved for client work. Here's what each ChatGPT plan does with what you type, the settings worth changing, and how to strip a prompt of anything that identifies a client.
Not on a personal ChatGPT account with its default settings, and not on any plan until your firm has approved that tool for client data. Those are two separate questions, and most advice online only answers the first.
The first is what the vendor does with what you type: whether it trains on it, how long it keeps it and who can read it. That depends on your plan and your settings, and it's covered below.
The second is whether you're allowed to hand the information to a third party at all. Your professional rules, your client's contract or engagement letter, privacy law and your firm's policy decide that. A better plan can't answer it for you.
Data-handling details come from OpenAI's, Anthropic's and Google's own help and privacy pages, as of October 2026. Vendors revise these pages often, so check the current version before relying on any of it. This is general information, not legal or compliance advice. Last reviewed 23 October 2026.
OpenAI splits ChatGPT into personal plans (Free, Go, Plus and Pro) and managed workspaces such as Business, Enterprise and Edu. The defaults are very different.
| As of October 2026 | Personal plans | Business, Enterprise, Edu |
|---|---|---|
| Used to train OpenAI's models | Yes by default, with an opt-out | No by default |
| How long chats are kept | Until you delete them, then purged within 30 days | Set by workspace admins; deleted chats purged within 30 days |
| Who else in your organization can see them | No one; it's your account | Business admins can view, export and delete members' chats. Enterprise admins get an audit log. |
| Shared links | Anyone with the link can view | Only members of the same workspace |
The business tiers also come with paperwork that firms need. OpenAI's enterprise privacy page says it will sign a data processing addendum for Business and Enterprise, that those plans have completed SOC 2 Type 2 audits, and that data is encrypted at rest and in transit. Its API, which developers use to build tools, isn't trained on by default either. For health information, OpenAI says it can sign a business associate agreement for the API, and it offers a separate ChatGPT for Healthcare workspace.
One detail HR teams should notice: a business account keeps your chats out of OpenAI's training, not out of sight of your own admins.
Claude and Gemini follow the same personal-versus-business split. Anthropic asks Claude Free, Pro and Max users whether their chats can train its models, and keeps them for up to five years in de-identified form if they say yes; its Team, Enterprise and API products aren't trained on by default. Google's consumer Gemini app has human reviewers read some chats, keeps reviewed chats for up to three years, and its privacy hub asks users not to enter confidential information they wouldn't want a reviewer to see. Gemini inside Google Workspace isn't human-reviewed or used to train models outside your organization without permission.
If you use a personal account for any work at all, check these five things. They work the same way on the web and in the mobile apps.
Deleting a chat removes it from your view at once and schedules it for permanent deletion within 30 days, unless OpenAI has to keep it for security or legal reasons. That exception isn't theoretical. In 2025, a preservation order in The New York Times' lawsuit against OpenAI required the company to keep consumer ChatGPT content it would normally have deleted. OpenAI says those obligations ended on 26 September 2025, though it's still storing a limited set of April to September 2025 data. Treat a deleted chat as probably gone within a month, not as something that never happened.
Practice this, don't just read it.
Iro AI turns ideas like the ones in this post into 5-minute exercises with feedback. Free tier, Pro from $4.17 a month ($49.99 a year, 7-day free trial).
Even on an approved business plan, some information shouldn't go into a prompt without a specific reason and sign-off. On a personal account, none of it should.
Most useful prompts don't need the client at all. They need the shape of the problem. A five-step routine:
Be honest about what this achieves. Placeholders reduce risk; they don't make data anonymous if it can be linked back. The GDPR, for one, treats pseudonymized data that "could be attributed to a natural person by the use of additional information" as information about an identifiable person. For the wider habits of using AI on real work, see how to use AI at work.
If your firm has an AI policy, read it first. If it doesn't, these questions get you a usable answer quickly:
The professionals who get the most out of AI aren't the ones who paste the most into it. They write prompts with all the context the model needs and none of the client, and they know which jobs to keep for themselves.
That's a learnable skill, and Iro AI trains it in about five minutes a day. Delegate-versus-do-it-yourself exercises train the call on what to hand off, guardrail-writing drills teach you to set limits inside the prompt, and on Pro the Prompt Lab scores your real prompts on clarity, specificity, constraints and context. If ChatGPT is your main tool, start with learning ChatGPT properly.
Practice this, don't just read it.
Iro AI turns ideas like the ones in this post into 5-minute exercises with feedback. Free tier, Pro from $4.17 a month ($49.99 a year, 7-day free trial).
Not on a personal plan with default settings, where conversations can be used for training. Business, Enterprise and Edu plans are not trained on by default, but whether you may share client data with any vendor depends on your professional rules, client agreements and firm policy. Check with your firm before pasting anything identifiable.
On personal plans (Free, Go, Plus and Pro), yes by default. You can turn off Improve the model for everyone in Settings, under Data controls. OpenAI says it does not train on Business, Enterprise, Edu or API data by default. On personal plans, rating a response with thumbs up or down can still send that conversation for training.
Temporary chats do not appear in your history, do not create memories and are not used to train models. OpenAI may keep a copy for up to 30 days for safety purposes. If you save a temporary chat, it becomes a regular chat and follows your normal settings.
Only within their professional duties. ABA Formal Opinion 512 says lawyers need informed client consent before entering information about a representation into a self-learning AI tool that could disclose it, and that boilerplate in an engagement letter is not enough. State rules vary, so check your jurisdiction and firm policy. This is general information, not legal advice.
It has much stronger defaults than a personal account. OpenAI does not train on Enterprise data by default, admins set retention, and the plan has completed a SOC 2 Type 2 audit with encryption at rest and in transit. Approval for client data is still your firm's decision, and your own admins can access conversation logs.
Replace names with roles, remove direct identifiers such as account numbers, addresses and dates of birth, and blur details that identify someone in combination, like a rare job title plus a location. Ask for the method rather than pasting the record. Placeholders reduce risk but do not make data truly anonymous if it can be linked back.