Your office manager discovers that ChatGPT can draft client emails in seconds. Your salesperson starts pasting in proposal templates to get them polished up. Your bookkeeper feeds it a spreadsheet to summarize expenses. Nobody cleared any of this with you.
This scene is playing out in small businesses everywhere right now — and in most of them, nobody has stopped to ask a basic question: where does that data actually go?
This isn't a reason to ban AI tools. Used correctly, they're genuinely useful. But "correctly" requires understanding what happens on the other side of that chat window — especially when you're a business that handles customer information, contracts, or anything regulated. Here's what you actually need to know.
What Happens When You Type Something Into a Chatbot
When you type a message into a consumer AI tool — ChatGPT (free tier), Claude (free tier), Gemini, and others — your input travels over the internet to the provider's servers, gets processed by their model, and a response comes back. Simple enough.
The less obvious part: depending on which product you're using and which settings are enabled, that input may also be stored and may be used to train or improve future versions of the model. This isn't hidden — it's in the terms of service that essentially nobody reads.
OpenAI's free ChatGPT tier, for example, historically used conversations for model training unless users opted out. The opt-out exists, but it's not the default for every user in every context, and it varies by product, region, and account type. Other providers have similar policies with their own nuances.
The practical upshot: if you paste a client's name, address, and medical history into a free consumer chatbot to help you draft a letter, there is a real possibility that information is being stored on someone else's servers and potentially reviewed by humans for quality-assurance purposes. That is a problem — both for your client's trust and potentially for your legal obligations under HIPAA, GDPR, state privacy laws, or your own contracts.
The Difference Between Consumer and Business/Enterprise Tiers
This is the single most important distinction most small business owners don't know about.
Consumer/free tiers (ChatGPT Free, Claude.ai free, etc.) are built for individual use. Privacy controls are minimal by default. Data handling is governed by consumer terms of service, which are written to give the provider significant latitude.
Business and enterprise tiers operate differently — and meaningfully so.
- ChatGPT Team and Enterprise: OpenAI states that data from these tiers is not used for training by default, and conversations are not retained beyond the session window unless you specifically enable memory features.
- Claude for Business / Anthropic API: Anthropic's API and business agreements include data processing agreements (DPAs) that give customers more control and clearer commitments about how data is handled.
- Microsoft Copilot (via Microsoft 365): For businesses already on Microsoft 365, Copilot routes data through Microsoft's existing enterprise data protection commitments — generally a stronger privacy posture than using a standalone free tool.
The gap between these tiers isn't just about features. It's about accountability. A paid business tier usually comes with a Data Processing Agreement — a legal document where the vendor commits to specific data-handling practices. Without a DPA, you have no contractual basis to enforce privacy obligations on the vendor. For any business handling client data, that matters.
The rough rule of thumb: If your business has a professional or legal obligation to protect client data, you should be using a paid/business tier with a signed DPA — or not using AI for that data at all.
What Should Never Be Pasted Into Any AI Tool (Without Careful Vetting)
Even on a well-configured business tier, some categories of information warrant serious caution. For free consumer tools, treat these as hard stops.
Client personally identifiable information (PII)
Names + addresses + account numbers + dates of birth — any combination that uniquely identifies a real customer. Especially sensitive: health information, financial account details, Social Security numbers, or anything covered by a specific regulation (HIPAA, GLBA, CCPA, etc.).
Contracts and NDAs
These documents often themselves contain confidentiality clauses. Pasting a fully executed contract — including counterparty names and deal terms — into a third-party AI service may technically violate the agreement you just signed.
Login credentials and API keys
This one should be obvious, but it happens: people paste in configuration files or setup instructions that include passwords or secret keys. Never do this. Not once.
Proprietary financial data
Revenue figures, margin data, payroll details, or anything that would be material if it became public or landed with a competitor.
Unpublished intellectual property
Unreleased product plans, draft patents, or trade secrets. Putting these into an external service creates an ambiguous paper trail around ownership and disclosure.
Employee personal information
Performance reviews, salary data, disciplinary records, or anything from an HR file.
A useful test before pasting anything: Would I hand this document to a stranger in a coffee shop and ask them to summarize it? That's closer to the reality of what free consumer AI tools represent than most people realize.
A Simple One-Page AI Data Policy for Your Team
You don't need a 40-page security manual. You need something your team will actually read and follow. Here's a straightforward framework you can adapt:
[Your Company Name] AI Tool Usage Policy
Effective date: [Date] | Owner: [Name or Role]
Purpose
We use AI tools to work more efficiently. This policy ensures we do that without putting client data, company data, or our reputation at risk.
Approved Tools
[List the specific tools you've vetted and approved — e.g., "ChatGPT Team account (company login only), Microsoft Copilot via Microsoft 365."]
Do not use personal or free accounts for work tasks.
What You Can Use AI For
- Drafting communications, then reviewing before sending
- Summarizing internal documents that contain no client PII
- Brainstorming, ideation, research synthesis
- Editing your own writing
- Creating templates and frameworks
What Never Goes Into an AI Tool
- Client names combined with any personal identifiers
- Any data covered by a confidentiality agreement
- Login credentials, passwords, or API keys
- Payroll, financial, or HR records
- Unreleased product or business strategy documents
When in Doubt
Remove or replace specific details before using AI. Describe the situation in general terms rather than pasting real data. ("A client in the healthcare industry with a billing dispute of approximately $X" rather than pasting the actual invoice.)
Questions
Direct any questions to [Name/Role]. If you're unsure whether something is appropriate to use, ask first.
One page. Print it. Post it near the coffee machine. Do a ten-minute walkthrough with your team. That's the entire implementation.
The Practical Setup: What to Actually Do This Week
If you want to move from "we're just winging it" to "we have this under control," here's a concrete sequence:
-
Audit what your team is currently using. Ask around. You'll be surprised. People are using free accounts, personal accounts, and tools you've never heard of.
-
Decide on one or two approved tools. Pick business-tier accounts for anything that touches client work. This is usually $20–$30/user/month for tools like ChatGPT Team or Copilot. That's a reasonable cost for the liability reduction.
-
Check whether your highest-risk vendors have a DPA available. For the tool you use most, look for "Data Processing Agreement" or "DPA" in their legal docs. If you handle health data, financial data, or data covered by state privacy laws, make sure you have one signed.
-
Write and share the one-pager above. Customize it, send it to your team, and spend fifteen minutes talking through it.
-
Create a "clean version" habit. When drafting AI prompts that involve real situations, train yourself and your team to anonymize before pasting. It takes thirty seconds and eliminates most of the risk.
This is the kind of AI readiness review we do with clients — walking through what tools the team is actually using, where the data exposure risk sits, and how to build simple guardrails that don't get in the way of the productivity benefits.
The Bottom Line
AI tools are genuinely useful for small businesses. They save real time and do real work. The data privacy risk isn't a reason to avoid them — it's a reason to use them thoughtfully.
The main things to internalize:
- Free consumer tiers have the weakest data protections. Treat them accordingly.
- Business and enterprise tiers are meaningfully different, and for client-facing work, that difference matters.
- Certain categories of data — client PII, contracts, credentials, proprietary financials — warrant either a hard stop or serious vetting before they go anywhere near a third-party AI.
- A simple written policy, actually shared with your team, closes most of the gap between "winging it" and "responsible use."
None of this requires a security team or a six-figure compliance budget. It requires a little deliberate thinking, which is almost always the part that gets skipped.
If you'd like help thinking through how your specific business should be using AI — which tools make sense, what your data risk actually looks like, and where the real productivity opportunities are — book a strategy call and we can work through it together.