Short answer: it depends. It depends on which plan you use, which settings are on, and what your people paste in.
"Is ChatGPT safe?" is the wrong question, because it treats three different things as one. The free app on someone's phone, a paid personal plan, and a business plan are different products with different data rules. The safe move is to know which one your team is using and to write down what is allowed.
This post covers what the vendors say about your data, where the real risk is, and a one-page policy you can adopt this week. It is part of the AI implementation for small business series.
This is general information, not legal advice. If you handle regulated data (health records, payment cards, certain client contracts), talk to a lawyer or your compliance contact.
Two things people mix up
Training. Some AI companies use what you type to improve their models. That means pieces of your conversations could shape future versions of the AI. Whether that happens depends on the plan and the settings.
Access and storage. Separately from training, your chats are stored somewhere. Someone at the vendor may be able to see them in limited cases, and people at your company can see them too, depending on how accounts are set up.
Both matter. Most owner worry goes to training. Most real trouble comes from the second one, plus people pasting in things they shouldn't.
What the vendors publish
As of November 2026, here is what we could read. Vendors change these pages, so check the links before you rely on any detail.
OpenAI (ChatGPT)
OpenAI's pricing page has a comparison table. For the Free, Go, Plus and Pro plans, it lists "Content is used to train our models: Opt-out available." That means on personal plans, your content may be used for training unless you turn it off in settings. The Business and Enterprise plans are not on that row of the table. Source: OpenAI.
We tried to read OpenAI's dedicated business privacy pages for this post. They returned an access error to our reader, so we are not going to quote them. Go to OpenAI's business privacy page yourself and read what it says about training, retention and ownership before you put customer data into a Business or Enterprise plan.
Anthropic (Claude)
Anthropic's privacy center says that on its consumer plans (Free, Pro and Max), chats may be used for training if you opt in through its model-improvement setting, if a conversation is flagged for safety review, or if you opt in another way. Incognito chats are never used for improvement. Thumbs up or down feedback is stored for up to 5 years and may be used for training after it is de-linked from your user ID. This article is dated March 16, 2026. Source: Anthropic.
For its commercial products, Anthropic's article dated August 18, 2026 says it will not use inputs or outputs from them to train models. The exception is if you send feedback or otherwise opt in. Owners on Team and Enterprise plans can turn off the feedback setting. Source: Anthropic.
Google (Gemini)
For the Workspace business product, Google's privacy hub says Workspace "does not use customer data for training models without customer's prior permission or instruction." It also says your conversations with Gemini are visible only to you, even inside shared files, and that admins control history retention. Feedback may be retained for up to 18 months. Source: Google.
Google's Gemini plans page says that for its enterprise editions, "You own your data, not Google," and that prompts and outputs are not used to train Google's models or models for other customers. It lists the free tier of its developer tool, Google AI Studio, as "Content used to improve our products," and it does not spell out training terms for personal plans on that page. Source: Google.
What the pattern tells you
On the pages we could read, business-grade plans have stronger default protections than personal plans, and personal plans lean on settings you have to find and change yourself. That is the whole point of buying a business plan. It moves the safe setting from "you have to remember" to "it's the default."
None of this makes any tool safe for everything. It lowers one risk. It does not touch the others.
Where the real risk sits
Shadow AI
"Shadow AI" is employees using AI tools on their own accounts without telling you. The Pax8 SMB AI Pulse Report for Q3 2026, surveying 1,150 small and midsize business leaders in four countries, found 35% of AI-engaged businesses had discovered shadow AI in the past six months. Pax8 sells AI software through IT providers and the figure blends four countries, so use it as a signal. Source: Pax8.
The U.S. Chamber of Commerce Foundation and Ipsos, in a June 2026 survey of 1,070 workers at small businesses (2 to 499 employees), found half already use AI at work. In 19% of cases the workers said adoption was driven mostly by employees exploring on their own, compared with 11% driven mostly by organizational guidance. Source: U.S. Chamber of Commerce Foundation.
Your team is probably already using it. The question is whether you know how.
Pasting in what shouldn't be pasted
The typical leak is not a hacker. It is a well-meaning employee pasting a customer list, a signed contract, a password, or a photo of an ID into a chat window to "get a quick summary."
Concerns owners already have
Goldman Sachs' 10,000 Small Businesses survey, fielded January and February 2026 among program alumni, named lack of technical expertise, difficulty choosing tools and data privacy concerns as barriers to using AI. Source: Goldman Sachs. You're not alone in worrying about it.
The one-page AI use policy
Copy this. Change the parts that don't fit. Put it where people will see it, and go over it once in a team meeting.
1. Approved tools. We use [name the tool and plan]. Nothing else for work, unless the owner says yes. Work accounts only, never personal accounts.
2. Never paste these.
- Passwords, keys, codes, bank or card numbers
- Social Security numbers, ID photos, medical information
- Full customer lists or exports
- Signed contracts or anything under NDA
- Anything a customer told us in confidence
3. Okay to paste.
- Public information and your own marketing copy
- Rough notes with names and details removed
- Template text with the specifics left blank
4. Strip it first. Replace names with "Customer A" and numbers with ranges when you only need help with the wording.
5. A human checks everything. AI drafts. A person reads before anything goes to a customer. AI is wrong sometimes and sounds sure anyway.
6. Tell the AI nothing you'd mind seeing on a billboard (if you are on a personal or free plan).
7. Settings. Training and memory settings stay set the way the owner chose. Check them when you start and every quarter.
8. Tell us. If something sensitive goes in by mistake, say so right away. Nobody gets in trouble for reporting it. People get in trouble for hiding it.
9. Offboarding. When someone leaves, remove their access to every AI tool the same day.
10. Review. The owner reviews this page every quarter. Vendors change their terms, and so will we.
That is one sheet. If you would rather build a version tailored to your business, use our free AI use policy generator. You answer eight questions and it builds a one-page, plain-English policy you can copy. Your answers are not saved or sent by the tool.
Make the policy real
A policy no one remembers is a decoration. Three habits keep it alive.
- Pick the tool and buy the right plan. If the team needs AI, give them an approved account. People go around rules that have no easy alternative.
- Train once, in person if you can. Twenty minutes with real examples beats a PDF. Use a made-up customer record and show what "strip it first" looks like.
- Check on a schedule. Look at who has accounts, what settings are on, and whether new tools have appeared. Our managed AI explainer covers what a monthly check includes.
A note on regulated data
If you handle health information, payment card data, or data covered by a client contract or a state law, a general policy is not enough. Whether a particular tool or plan is allowed for that data is a legal and compliance question, and it depends on your situation. Ask a qualified lawyer or your compliance contact. We can help you map where AI touches your data. We don't give legal advice.
The next step
Tonight, ask three people on your team what AI tools they use for work and which account they use. You may be surprised. Then use the policy above, or the generator, to put the rules in writing.
If you want someone to look at your setup for you, start with the free AI audit, ten questions and a free PDF. See how we help with rollouts on our AI implementation page.