Every time you paste a client email, a contract, or a spreadsheet into an AI tool, you’re sharing that information with a third-party service. That isn’t automatically a problem. Businesses share data with email providers and cloud storage every day. But AI tools have their own settings and defaults, and many people never look at them.
This guide explains the questions to ask, what the major vendors say about business data as of October 2026, and a practical checklist and policy for small businesses.
Disclosure: This article may contain affiliate links. If you sign up through them, we may earn a commission at no extra cost to you. Vendor information is based on each company’s published documentation and policies as of October 2026, not independent testing. This is general information, not legal, security, or compliance advice.
The Five Questions to Ask About Any AI Tool
- Training: is my data used to train or improve the vendor’s AI models? Can I turn that off?
- Retention: how long does the vendor keep my conversations and files? Can I delete them?
- Human review: can the vendor’s staff or contractors read my content, and when?
- Access: who in my organization can see what? What can the AI itself access, such as email, files, and calendars?
- Contracts and compliance: is there a business agreement, a data processing agreement (DPA), or security certifications such as SOC 2?
The answers often depend on which plan you’re on.
Personal Plans vs Business Plans
The biggest practical difference is between personal and business accounts. According to vendors’ documentation as of October 2026:
| Vendor | Personal plans | Business plans |
|---|---|---|
| OpenAI (ChatGPT) | Content may be used to improve models unless you turn off “Improve the model for everyone” (Settings → Data Controls) | Business, Enterprise, Edu, and the API aren’t used for training by default |
| Anthropic (Claude) | Users choose whether chats are used to improve models. Anthropic says that if you allow it, data may be retained in de-identified form for up to five years | Team and Enterprise plans have commercial terms. Check Anthropic’s documentation for details |
| Google (Gemini) | Turning off “Keep Activity” stops use for training and human review. Google says chats are still kept for up to 72 hours to provide the service | Google says Gemini in Workspace isn’t used to train models and isn’t reviewed by humans |
| Microsoft (Copilot) | Consumer Copilot has its own privacy settings | Microsoft says Microsoft 365 Copilot works within your organization’s permissions and existing Microsoft 365 data protections |
| Grammarly | You can turn off “Product Improvement and Training” | Defaults vary by how the account was bought. Check Data Settings |
Policies change, and the details matter, so read the current documentation for your plan.
The takeaway: if staff use AI for work, a business plan is usually the safer default. If anyone uses a personal plan for work, make sure the training opt-out is turned on.
What Not to Paste Into AI Tools
Unless the tool is approved for it and your contracts allow it, avoid pasting:
- Passwords, API keys, and access codes. Never paste these into any AI tool.
- Payment card and bank account numbers
- Government ID numbers (Social Security numbers, passport numbers)
- Health information about customers or employees
- Confidential client material covered by an NDA or contract terms
- Employee HR records (performance, discipline, medical)
- Unannounced financials or deals
Safer alternatives: remove names and identifying details before pasting (“Client A,” “$X”), summarize instead of pasting raw records, or use an approved business tool with a proper agreement.
Newer Risks: Connectors, Agents, and Permissions
AI tools increasingly connect to your email, files, calendars, and other apps, and agents can act on your behalf. That adds risks:
- Over-broad access: an AI assistant connected to your company drive can surface anything the user has permission to see. Clean up sharing permissions before rolling out AI.
- Prompt injection: content the AI reads, such as a web page, email, or document, can contain hidden instructions meant to manipulate it. Be cautious about letting agents act on untrusted content, and supervise agents that can send messages, make purchases, or change data.
- Browser agents in logged-in accounts: AI browsers and agents can act within accounts you’re signed into. Limit what they can access.
- Third-party plugins and apps: each one is another vendor with your data. Review before enabling.
Regulated Data Needs Extra Care
If your business handles health, financial, legal, children’s, or other regulated data, general-purpose AI tools may not be appropriate without specific agreements and safeguards. Some vendors offer enterprise agreements for regulated industries. Talk to a qualified advisor before you use AI with this kind of data.
Client Contracts and Confidentiality
Freelancers and agencies: many client contracts restrict how you handle their information, and some now explicitly address AI. Before using AI on client work:
– Check your contracts for confidentiality and AI clauses
– Ask the client if you’re unsure
– Be transparent about how you use AI in your process
A One-Page AI Policy for Small Businesses
Adapt this outline:
- Approved tools: list the AI tools staff may use for work, and on which plans.
- Data rules: what may never be entered (see the list above), and what’s OK.
- Accounts: work use on business accounts. Training opt-outs on for any personal accounts.
- Human review: a person checks all AI output before it reaches customers or is relied on.
- Connectors and agents: approval needed before connecting AI to email, files, or other systems.
- Disclosure: when you tell customers AI was involved.
- Offboarding: remove AI tool access when people leave.
- Owner: who maintains the policy and answers questions.
Pre-Paste Checklist
- [ ] Am I on a business plan, or a personal plan with training turned off?
- [ ] Do I know how long this tool keeps my data, and how to delete it?
- [ ] Have I removed names, account numbers, and identifying details I don’t need?
- [ ] Does this content include passwords, keys, or payment details? (If yes, stop.)
- [ ] Does a client contract or NDA restrict sharing this?
- [ ] Is this regulated data that needs special handling?
- [ ] Will a human review the output before it’s used?
FAQ
Is ChatGPT safe for business use?
It can be, with the right setup. OpenAI says ChatGPT Business and Enterprise aren’t used for training by default. On personal plans, turn off “Improve the model for everyone.” Follow your own data rules either way.
Does deleting a chat delete my data?
It depends on the vendor. For example, Anthropic says deleted Claude chats are removed from its backend storage within 30 days, and Google says Gemini chats are kept for up to 72 hours even with Keep Activity off. Check each vendor’s retention policy.
Are business plans worth it just for privacy?
For teams handling client or customer data, usually yes. They add training protections by default, admin controls, and business terms.
Can AI tools leak my data to other users?
Reputable vendors design their systems to keep accounts separate, but no system is risk-free. Data used for training could in theory influence future outputs, which is one reason to opt out or use business plans for sensitive work.
Related Reading
- Choose an assistant: ChatGPT vs Claude vs Gemini
- AI in your office suite: Microsoft 365 Copilot vs Gemini for Google Workspace
- Build your stack: The Best AI Tools for Small Business
Policies change often. Everything here is as of October 2026, so check each vendor’s current documentation. This article is general information, not legal, security, or compliance advice.
