Swap and return.
Use ChatGPT on a customer’s message without handing over the customer. Swap the names and numbers for labels, ask the AI, then put the real details back into its reply. Everything happens on this phone or computer.
Paste the message
Read it, then copy it into ChatGPT
The yellow labels stand in for the real details. Read it through before you copy it. If a name or number was missed, add it to “Also hide these words” above and press Swap again.
What was swapped
Untick anything that did not need hiding. “Sure” means the detail has a fixed shape the tool can test, like a card number or an email address. “Check this” means it was worked out from clues in the message, like a name after “Hi”, so make sure it is right and nothing is missing.
Paste the AI’s reply
The labels in the reply are turned back into the real details, here on this device. Keep this page open until you have finished: if you close it, the list of swaps is gone.
Your finished reply
What it will not do
It finds names by the clues around them: “Hi Sarah”, “My name is”, “Mr”, a signature, a WhatsApp sender, or a name written next to a phone number, an email or an address. A name with nothing around it can be missed, so always read the swapped message, and add anything it missed yourself.
The AI still sees the rest of the message: the complaint, the product and the amounts. And this is not a legal check of any kind. It does one thing: the details you swap never leave this device.
Nothing you paste here is sent anywhere or saved. Turn the wifi off and it still works.
How well it works
We wrote 70 made-up customer messages in the way people write in South Africa, Nigeria, Kenya, India, the UK and the US: WhatsApps, emails and complaints, with 299 names, numbers and addresses in them. It swapped all 299. It also swapped 6 things that did not need hiding, all of them a company’s own email address.
That is a small test, and we improved the tool on those same messages, so real messages will go less well. That is why we keep testing it on new messages it has never seen, and why you should always read the swapped message before you copy it.
Why this exists
In a March 2026 survey of office workers in seven countries, 52% said they had used AI tools their employer never approved, while nine in ten of their bosses said they knew which tools were in use.
In a 2025 study of 48,000 people in 47 countries, including South Africa, Nigeria and Egypt, almost half of workers admitted using AI against their employer’s rules, including putting customer information into free public AI tools.
Sources: Okta survey, reported by The Register, May 2026 (Okta sells workplace security tools) · KPMG and the University of Melbourne, 2025.