Blog · · Benatar Brands
Data privacy questions to ask before you plug AI into customer data
The moment an AI tool can read your customer list, your order history or your inbox, it is handling other people's information on your behalf. Your customers never agreed to that tool. They agreed to you. So before you connect anything, answer a short list of questions in writing, in your own words.
Owners already sense this. In a March 2026 survey of small business owners from Goldman Sachs 10,000 Small Businesses Voices, data privacy concerns were named among the barriers to using AI more deeply, alongside lack of technical expertise and difficulty choosing tools. The fix is not to avoid AI. It is to ask better questions before the data moves.
This checklist is not legal advice. It is the set of questions we work through before connecting a tool to anything with customer names, emails, addresses or payment details in it. If the answers point to a law that may apply to you, talk to a lawyer before you go further.
Start with the data, not the tool
The Federal Trade Commission's guide for businesses, Protecting Personal Information, builds a security plan on five principles: take stock, scale down, lock it, pitch it and plan ahead. The guide predates today's AI tools, but the order still works. You cannot judge a tool until you know what it would touch.
1. What customer data would this tool actually see?
Write down the systems, then the fields. "Shopify" is not an answer. "Order history with names, emails, shipping addresses and what each person bought" is. Include the places data hides: inbox attachments, CRM notes fields, spreadsheets someone exported last year.
2. Does the job need all of it?
The FTC's second principle is to keep only what you need, and it applies to access as much as to storage. The guide calls this the principle of least privilege: each person gets access only to what their job requires. Treat the AI tool the same way. A tool that summarizes weekly sales may not need names at all. If you can connect a narrower view, a single folder, a filtered list, a read-only report, do that instead of the whole account.
3. Can it change or send anything, or only read?
Reading and writing are different risks. A tool that can only read can leak data. A tool that can also send, edit or delete can leak it and act on it. Decide which one this job needs before you connect. We covered permissions, logs and kill switches in more detail in our October tools roundup, so we will not repeat that list here.
Then ask the vendor
4. Is my data used to train models?
Look for a direct sentence, not a paragraph about "improving our services." A clear answer looks like the one in Google's privacy hub for generative AI in Workspace: business users' chats and uploaded files won't be reviewed by human reviewers or otherwise used to train generative AI models outside the customer's domain without permission. Note what that sentence covers and what it does not. The same page says that using third-party apps with Gemini is governed by different terms. Every connection you add can come with its own answer.
5. Which plan does that answer apply to?
Vendors often have different terms for free, individual and business plans. Confirm the answer applies to the plan your team is actually logged into, not the one on the sales page. If staff use personal accounts for work tasks, the business terms may not cover them.
6. How long is the data kept, and can I delete it?
Ask how long prompts, uploaded files and connected data are retained, and whether you can delete them yourself. Ask what happens to it when you cancel. "We keep it as long as necessary" is not a retention period.
7. Who else touches it?
Most AI products run on other companies' infrastructure and models. Ask for the list of subprocessors and where the data is processed. The FTC guide's advice on contractors and service providers fits here: investigate their security practices before you outsource, put your expectations in writing in the contract, then verify rather than take their word for it.
8. Will they tell me if something goes wrong?
The same FTC section says to insist that service providers notify you of security incidents, even ones that may not have led to an actual compromise of your data. Find the notice clause in the vendor's terms or data processing addendum. If there is no commitment, write that down as a risk.
Then check your own promises
9. Does this match what we told customers?
Read your own privacy policy. If it says you share data only with providers who help you run the store, ask whether this tool fits that description. If you promised customers something about how their data is used, the tool has to fit inside that promise, not the other way around.
10. Could a privacy law apply to us?
You do not need to become a privacy lawyer, but you should know whether a law with specific rules might cover you. One example: the California Attorney General's CCPA page says the law applies to for-profit businesses that do business in California and meet any of three tests: gross annual revenue over $25 million; buying, selling or sharing the personal information of 100,000 or more California residents or households; or earning 50% or more of annual revenue from selling California residents' personal information. Covered businesses must, among other things, handle deletion requests and tell their service providers to delete the data too.
California's privacy agency has also adopted regulations, effective January 1, 2026, that cover risk assessments, annual cybersecurity audits for certain businesses, and consumers' rights to access and opt out of businesses' use of automated decisionmaking technology. Most small shops will not hit those thresholds. Some will, and other states and countries have their own rules. If yours might, ask a lawyer before the tool goes live.
Make the answers one page
Put the ten answers on a single page per tool, with the date you checked. When the vendor changes its terms or you add a new connection, check the page again.
If an answer is missing, that is information too. A tool that cannot tell you whether it trains on your data, or how long it keeps it, should not be reading your customer list yet. Start it on something that has no personal data in it, such as drafting product copy from your own notes, and revisit later.
Where we fit
Choosing what to connect, and writing these answers down before anything is connected, is part of our embedded AI partnership. If you are about to connect an AI tool to customer data and want a second pair of eyes on the answers, book a free session and we will go through the list with you.
Related service: embedded AI partnership.
Get our notes on AI and how businesses run in your inbox: subscribe to The Beyond Brief.