Using AI With Customer Data: The Traffic Light Rule

Here are three things you might paste into an AI chat this week.

“Write a short post about our new Sunday opening hours.”
“Help me reply to this: ‘Hi, it’s Linh Tran, 0912 345 678, can I move my colour appointment to Friday?’”
“Here’s the login for our card machine portal, user and password, so you can check last month’s payments.”

One of these is completely fine. One needs a ten-second edit before it goes in. And one should never go in at all.

Most people spot the first and the last straight away. It’s the middle one, the ordinary customer message, that catches nearly everyone. It’s also the one you’ll be tempted to paste every single day.

This post is the rule I use for that decision. I call it the traffic light rule: green, amber, red. It’s not about keeping customer data away from AI. AI is genuinely useful for replying to customers, and I’ve written about how to automate guest messages without sounding like a robot. It’s about using AI with customer data safely, with a habit that takes seconds.

A quick note before we start: this is not legal advice. Privacy laws are different from country to country. I’ll mention a couple of them below, but if you handle sensitive information, talk to someone who knows the rules where you operate.

What actually happens to what you type

When you paste something into an AI chat tool, it doesn’t vanish when you close the window.

The tool stores your conversation. Depending on which tool you use, which plan you’re on and how your settings are set, that conversation may be kept for a period of time, reviewed for safety reasons, or used to help improve the AI. The details are in each tool’s terms and privacy policy, and they differ a lot between tools and between free and paid plans.

That isn’t a scandal. It’s written down, and for most of what you type, it doesn’t matter at all. Your Sunday opening hours are already on your door.

But here’s the part that’s easy to miss. You agreed to the tool’s terms. Your customers didn’t. When you paste a customer’s name, phone number or private details, you’re sharing something that isn’t really yours to share.

A simple way to think about it: treat a general AI chat like a postcard, not a sealed letter. If you wouldn’t write it on a postcard, don’t paste it in as it is.

Throughout this post I’ll use an example. Mai runs a small hair salon. She’s fictional, and so are her customers and every message below. But her week will look familiar if you run any small service business, or a few holiday rentals.

The traffic light rule

Three lights. Every time you’re about to paste something, give it a colour first.

Give everything a colour before you paste it: green goes in, amber gets edited first, red stays out.

Green: paste away

Green is anything that’s already public, or that’s yours and not private.

Your opening hours, prices, services and menu.
Your own posts, emails, drafts and notes.
Your policies, FAQ, house rules and check-in instructions (without door codes).
General questions: “How do I word a price increase politely?” or “What should a cancellation policy cover?”
Patterns without people. “Lots of clients ask whether we do men’s cuts” is green. There’s no one in it.

Green is where most of the value is anyway. Writing better captions, tidying your FAQ, drafting a polite reply template, planning a promotion. If you want better answers here, the 5-part prompt recipe helps far more than any privacy setting.

Amber: edit first

Amber is the everyday stuff that involves a real person. A customer message you want help replying to. A complaint. A booking change. A note about a client.

The message itself is fine to get help with. The problem is the details that point to a real person:

full names
phone numbers and email addresses
home addresses
booking references and order numbers
birthdays, photos, anything that identifies them

So before you paste, take ten seconds and swap those details for placeholders. Linh’s message becomes:

“Hi, it’s [CLIENT], [PHONE], can I move my colour appointment to Friday?”

The AI can write exactly the same good reply. It doesn’t need to know it’s Linh. You put her name back in and send the reply yourself.

That’s the whole habit: paste the problem, not the person.

Red: never in a general AI chat

Red means no edit makes it safe for a general AI chat tool.

Passwords and logins. Like the card machine login from the start.
Card numbers and bank details. Yours or your customers’.
ID numbers. Passports, national ID cards, driving licences. Many rental hosts collect these at check-in. They stay out.
Health information. In a salon this can be as small as a client’s allergy or a skin condition. In a rental it might be a guest’s medical need.
Anything you promised to keep confidential. Contracts, private disputes, a staff member’s personal situation.
Whole customer lists. This one surprises people. Exporting your booking system and pasting five hundred customers in to “find patterns” puts five hundred people’s details somewhere they never agreed to.

If you really need AI help with red items, that work belongs in a tool built for that job, with terms that clearly cover it. More on that below.

The ten-second edit, in practice

Amber is the light you’ll use every day, so it’s worth doing properly.

Here’s an example complaint Mai might get (written for this post):

“Hi, this is Hoa Nguyen (booking #A3172). I came in on Saturday for a colour and my scalp has been itchy and red since, I have psoriasis and I told the stylist. Not happy. Call me on 0905 111 222. Hoa Nguyen, hoa.nguyen@example.com”

Before pasting, Mai makes three kinds of swap:

Names become [CLIENT]. Including the signature at the bottom, which is easy to miss.
Numbers and contact details become placeholders. [PHONE], [EMAIL], [BOOKING].
Health details become something general. “I have psoriasis” becomes “a sensitive-skin issue I mentioned to the stylist”.

What she pastes:

“Hi, this is [CLIENT] (booking [BOOKING]). I came in on Saturday for a colour and my scalp has been itchy and red since, I have a sensitive-skin issue and I told the stylist. Not happy. Call me on [PHONE].”

Then she adds her usual prompt: who she is, what she wants (a calm, sincere reply that offers a call and a free follow-up), the facts, and the shape of the answer.

The reply comes back with [CLIENT] in it. She swaps the name back, reads the whole thing, checks the facts and sends it herself. The AI still did most of the writing, and Hoa’s details never left Mai’s phone.

A few tips that make this faster:

Use the same placeholders every time. Consistency means less thinking.
If there are two people in a message, use [CLIENT 1] and [CLIENT 2] so the reply still makes sense.
If the edit feels long, just retype the question in your own words. “A guest says the air conditioning in unit 3 isn’t working and they’re upset. Help me reply.” That’s green.

A quick quiz from Mai’s week

Pick a light before you read the answer.

1. “Write three Instagram captions for our new hair-colour range.” Green. Nothing private in it.

2. A client’s public Google review, with their first name, to help her reply. Amber, but only just. The review is already public, so it’s close to green. Dropping the name anyway keeps the habit simple.

3. A photo of a client’s appointment card, with their phone number on it. Amber. Crop the number out, or just type the question instead of uploading the photo.

4. “Here’s my supplier invoice with our bank details. Can you check the maths?” Red as it is. Remove the bank details first, or use your bookkeeping software for this.

5. “A client told me she’s pregnant and asked which colours are safe. Help me reply.” Red as written, because it’s health information about a specific person. But the general question is green: “Which hair-colour treatments do salons usually avoid during pregnancy?” And you should still check that answer with a professional before passing it on.

That last one shows the trick. A red item can often be turned into a green question by removing the person from it.

Three sneaky leaks

Most accidental sharing doesn’t happen on purpose. It happens in the bits you didn’t look at.

1. Screenshots. A screenshot of your inbox shows every name and preview line in the sidebar, not just the one message you care about. Crop it, or copy the text instead.

2. Forwarded threads. You paste one email, but the thread underneath has three more people’s names, phone numbers and signatures. Delete everything below the message you need.

3. File uploads. A spreadsheet uploaded “just to tidy the formatting” may have every client’s phone number in column C. Make a copy, delete the personal columns, then upload the copy.

The fix for all three is the same: before you paste or upload, scroll once.

Check your AI tool’s settings once

This takes about five minutes, and it protects everything you paste from then on.

Open your AI chat tool’s settings and find the section about data, privacy or data controls. I won’t tell you which button to press, because every tool words this differently and the menus change often. Look for three things:

Is my content used to improve the AI? Many consumer AI chat tools have an option about whether your conversations can be used to train or improve their models. For business use, turn it off if you can.
A temporary or private chat mode. Some tools offer a mode where chats aren’t saved to your history. Useful for the occasional amber message. Read what it actually promises, as “not in your history” doesn’t always mean “never stored”.
How to delete a conversation, and how long deleted chats are kept before they’re really gone.

Then read the tool’s privacy policy, or at least the parts about these three things. Policies change, so check again every few months.

Business plans and purpose-built tools

Here’s the good news: you don’t have to keep everything out of AI forever.

Many AI providers offer business or team plans whose terms say your data isn’t used for training by default, with admin controls over who can see what. Read the terms for the specific plan, but this is often the right home for regular amber work.

And tools built for one job, like a booking system with AI-drafted replies, bookkeeping software, or a scheduling tool, are designed to handle customer data, because that’s the whole point of them. A good one tells you plainly where your data lives and what it’s used for. (This is also how we approach Santara: it reads booking and message data because that’s the job, and the host decides what happens next.)

So the real question isn’t “AI or no AI”. It’s which tool, and does it answer the right questions?

Five questions to ask any AI tool

Before you trust any AI tool with customer data, get clear answers to these:

Is my data used to train or improve the AI, and can I turn that off?
Where is it stored, and for how long?
Who can see it? On their side (staff, contractors, other companies they use) and on yours (which of your team members).
Can I delete it, and can I export it if I leave?
Is all of this written down? A good tool answers these on one clear page. If the answer is only in a sales call, that’s a warning sign.

These sit nicely alongside the questions in How to Tell If an AI Product Is Real or Just a Demo. Data handling is also one of the hidden costs of AI tools that rarely appears in the demo.

The law, in very broad strokes

Again: not legal advice. Laws differ by country, and they change. Two examples to show why this matters:

The EU’s GDPR protects personal data of people in the EU, and it can apply to businesses outside the EU that offer goods or services to people there. It treats some data, including health data, as special categories with extra protection. If you host guests from Europe, it’s worth knowing it exists.
Vietnam now has a national Law on Personal Data Protection (Law No. 91/2025/QH15), passed in June 2025 and in effect since 1 January 2026, with an implementing decree (Decree 356/2025/ND-CP) that replaced the earlier Decree 13/2023. According to summaries by law firms, health conditions are treated as sensitive personal data.

Other countries have their own rules, and this habit is no substitute for any of them. If you handle health, financial or children’s information, a short conversation with a local professional is money well spent.

Tell your customers

One small thing that builds trust: tell your customers, in plain words.

Mai adds one line to her booking page:

“We use software, including AI tools, to help us reply faster. We never sell your details or use them for anything other than your booking.”

Make sure whatever you write is actually true for your setup. A short honest line beats a long policy nobody reads.

One warning: safe is not the same as correct

Removing names doesn’t make a bad answer good. Everything about checking AI output still applies: read every reply before it reaches a customer, and check every fact, price and promise. AI tools can sound sure and still be wrong, as plenty of famous AI demos have shown.

Safe and correct are two different checks. Do both.

The one-page rule list

Print this and stick it next to the computer, or pin it in your team chat.

THE TRAFFIC LIGHT RULE FOR AI

GREEN: paste away

Opening hours, prices, services, policies, FAQ

Your own drafts, posts and emails

General questions with no one in them

Patterns without people (“many guests ask about parking”)

AMBER: edit first (10 seconds)

Real customer messages, complaints, booking changes, client notes

Swap: names → [CLIENT] · phones → [PHONE] · emails → [EMAIL] · booking refs → [BOOKING] · addresses → [ADDRESS]

Health details → something general (“a sensitivity issue”)

Put the details back yourself, then read before sending

RED: never in a general AI chat

Passwords and logins

Card numbers and bank details

ID and passport numbers

Health information about a named person

Anything confidential

Whole customer lists or exports

BEFORE YOU PASTE OR UPLOAD: scroll once. Crop screenshots · cut forwarded threads · delete personal columns from files

SET UP ONCE: “use my content to improve the AI” off · know the temporary chat mode · know how to delete

ASK ANY AI TOOL: Trains on my data? · Stored where, how long? · Who sees it? · Can I delete and export? · Is it written down?

ALWAYS: paste the problem, not the person. Check every fact before it reaches a customer.

Not legal advice. Privacy laws differ by country.

The takeaway

Green, paste away. Public info, your own writing, general questions. This is where most of AI’s value is.
Amber, edit first. Real customer messages are fine once names, numbers and identifying details are swapped for placeholders.
Red, never in a general chat. Passwords, payment details, ID numbers, health information, confidential matters and whole customer lists.
Scroll once before you paste. Screenshots, forwarded threads and spreadsheets are where details sneak in.
Check settings once, and choose tools that answer the five questions. Purpose-built tools with clear terms are the right home for real customer data.

You don’t need to be afraid of using AI with your customers. You just need a ten-second habit and a little care about where things go. Paste the problem, not the person.

Sources: Regulation (EU) 2016/679 (GDPR), Articles 3 and 9, eur-lex.europa.eu/eli/reg/2016/679/oj (readable version at gdpr-info.eu); DLA Piper, “Data Protection Laws of the World: Vietnam”, dlapiperdataprotection.com; LuatVietnam, “The latest Law on Personal Data Protection and the guiding documents”, luatvietnam.vn. All checked October 2026.