Redact the Client's Details Before You Paste That Email Into ChatGPT

Copy, paste, and your customer's data just left the building

Using ChatGPT, Copilot or Gemini to draft a reply, summarise a long thread, or untangle an awkward complaint is now routine. The habit that's less examined is what gets pasted in along with the task: the full email, headers, signature block and all — names, phone numbers, home addresses, account numbers, sometimes payment details.

The scale of this is measured, not anecdotal. LayerX's 2025 Enterprise AI and SaaS Data Security Report found that 77% of employees paste data into GenAI tools, 82% of that from personal accounts outside any company oversight, and roughly a fifth of what gets pasted contains PII or payment card data. A separate TELUS Digital survey in early 2025 found 57% of enterprise employees had entered sensitive information into a public AI assistant, most commonly personal data — names, addresses, emails, phone numbers — followed directly by customer information such as order history and support chat logs. Cyberhaven's tracking puts the trend in motion: the share of corporate data entering AI tools that's genuinely sensitive has more than tripled in two years, from 10.7% to 34.8%.

None of this requires anyone to be careless in any dramatic sense. It's a support agent pasting a customer's full complaint email to get help writing a reply, or a consultant dropping a client thread into an AI tool to summarise six months of back-and-forth before a call. The task is legitimate. The email just happens to be full of someone else's personal data, and most free-tier AI tools have no data processing agreement with your organisation covering that.

A concrete example

The 30-second habit that avoids the problem

For anyone who regularly pastes real customer or client email into an AI tool to draft, summarise or analyse.

The situation

A customer support agent wants an AI assistant to help draft a firm but polite reply to an escalated complaint. The original email includes the customer's full name, home address, order number and a partial card reference in the signature block from an earlier automated receipt they quoted.

Source and destination you choose

Where it comes from

The raw email or thread

Copy-pasted text, or a .eml/.msg export if you have the file

Why this one: You want the redaction done before the content goes anywhere else, so starting from the original is simplest.

Also works with

Forwarded copy from a shared support inbox
PDF export of a webmail thread

Where the result goes

Redacted plain text, ready to paste

Names, addresses, phone numbers and account-style numbers replaced with placeholders

Why this one: A placeholder like [CUSTOMER NAME] keeps the context an AI tool needs to draft a useful reply, without the actual identifying detail.

Or choose

Redacted PDF for a permanent record
Redacted CSV if you're batching several emails at once

What happens before you ever open the AI tool

1

The email is scanned for personal and financial detail

Names, addresses, phone numbers, email addresses and anything that looks like an account or card reference.
2

Suggested redactions are shown for review

Double-click any word, or select text, to redact or restore it — nothing is removed without you seeing it first.
3

You download the cleaned version

As plain text, PDF or CSV, keeping the tone and substance of the message but not the identifying detail.
4

That's what goes into the AI tool

The assistant gets everything it needs to help draft or summarise, minus the data you didn't need to hand over.

Automatic detection is a starting point, not a guarantee — review the suggested redactions before you rely on them, particularly for anything sensitive enough that a mistake would matter.

Why this is worth the extra step even when it feels unnecessary

It's tempting to treat this as excessive caution for a routine task. The counter-argument is in the numbers above: this isn't a rare mistake by a careless employee, it's the default behaviour of most people using AI tools at work, and most of it happens through personal accounts that IT and security teams have no visibility into at all. A habit of redacting before pasting doesn't require trusting every AI vendor's data policy — it just removes the question.

Practical questions about this workflow

Will redacting the email make the AI tool's reply worse?

Usually not — placeholders like [CUSTOMER NAME] or [ADDRESS] keep enough context for drafting or summarising. You reinsert the real detail yourself afterwards if the final message needs it.

Does this stop data leaving the building completely?

It removes personal and financial identifiers from what you paste elsewhere. It doesn't monitor what happens after that, and it isn't a replacement for a company policy on which AI tools are approved for work use.

Is this only useful for customer support teams?

No — consultants, recruiters, solicitors and anyone else regularly handling other people's personal information in email have the same exposure whenever they use an AI assistant on that content.