By Alberto Alatorre ·
How do I keep company data out of Ai tools?
Free chatbot plans and business plans treat your data differently. That difference is the whole answer. Know which plan your team is on, write down what stays out, and show everyone where the setting lives.
Training fixes the habit of pasting. Rules fix what is allowed to go in at all, and which plan the team uses. Both matter, and they are different jobs.
What is the difference between free and business plans?
On ChatGPT, the split is clear. ChatGPT Business data is excluded from training by default, and it is encrypted in transit and at rest, according to the OpenAI Help Center. Consumer plans may be used to improve models unless the user turns off Improve the model for everyone in Settings, under Data Controls.
Read that again. On a free or personal plan, your chats can become training data unless someone changed a setting. Most people never changed it. Most people do not know it exists.
Gemini works the same way, with its own setting. With Keep Activity on, Gemini chats can be used to improve Google services, including training. A subset is read by human reviewers, and reviewed chats are kept up to three years even if activity is deleted. That comes from the Google Gemini Apps Privacy Hub, updated October 2026, as reported by QuickTopicGuide in November 2026.
The pattern is the same on both tools. The free plan is the product learning from you. The business plan is the product working for you. If your team is doing company work on free accounts, the data rules are already broken, whatever your policy says.
What stays out, no matter the plan?
Some data should never go into a chatbot, on any plan. Write the list on one page and put it where the team can see it.
Customer names with account details stay out. Financial information stays out. Contracts stay out. Prices not yet published stay out. Anything a client would not want read by strangers stays out. When in doubt, leave it out.
The practical trick is placeholders. Instead of pasting a real client email, paste the same email with the name, the account number, and the dollar figure swapped for brackets: [client], [account], [amount]. The chatbot can still fix the wording. It does not need the real details to do that.
Where do the settings live?
For ChatGPT, the setting is Improve the model for everyone, in Settings, under Data Controls. Turn it off on every personal account the team uses for work. Better yet, move the team onto ChatGPT Business, where data is excluded from training by default and the setting is not something each person has to remember.
For Gemini, the setting is Keep Activity. Turning it off stops chats from being used to improve Google services, but note the exception: chats already reviewed by a human can be kept up to three years. The setting prevents future use. It does not erase what was already reviewed.
Vendor terms change often. The dates above were checked November 11, 2026. Check the settings again when a tool updates its terms, and check them once a quarter even if nothing changed.
What is the fastest fix for a small team?
Three moves, in one afternoon. First, move the team onto a business plan, so the default is on your side. Second, write the one page list of what stays out. Third, walk the team through the settings on their own accounts, so each person sees the switch with their own eyes.
A rule nobody has seen is not a rule. A setting nobody has touched is not a setting.
What to do next
Ai 101 teaches Ai limits and when to rely on human judgment, free, in about 15 minutes, with no signup, in English and Spanish. It is the right starting point before staff paste real company details into a tool. Read the course overview on the Ai 101 page, then use its “Start Ai 101 Now” button to open the course.