How Virtual Assistants Use ChatGPT on Client Work — Without Risking a Single Name

Let's be honest: if you're a virtual assistant in 2026, you're already using AI. Maybe you draft emails with ChatGPT, summarise meeting notes with Claude, or rewrite client proposals with a quick prompt. You'd be leaving money on the table not to.

But there's a line. And you hit it the moment you need to work with real client data — the email with the client's full name and personal mobile number, the contract that mentions the company's legal name and address, the invoice with bank details.

You know you shouldn't paste that straight into ChatGPT. Your client trusts you with their information. But manually find-and-replacing every name, email, and phone number across a three-page document? That's a twenty-minute detour that kills the whole point of using AI in the first place.

This article walks through how virtual assistants are solving this with a Chrome extension that anonymises client data before it reaches any AI tool — and restores it after.

The confidentiality problem VAs don't talk about

Virtual assistants operate in a unique position: you handle more confidential information, across more organisations, than almost any other role. A single morning might involve drafting a follow-up for Client A using their customer's full name, reformatting a proposal for Client B that contains financial projections, and replying to Client C's customers on their behalf.

Every one of those tasks could benefit from AI. And every one of them contains information your clients didn't agree to share with OpenAI, Anthropic, or Google.

And if you've signed an NDA or confidentiality agreement — which most VAs have — this isn't just an etiquette issue. When you paste client data into ChatGPT, that text is transmitted to OpenAI's servers. Depending on the wording of your NDA, that could count as disclosing confidential information to a third party. It doesn't matter that you're using it as a productivity tool. The data still left your device and reached someone else's infrastructure.

It doesn't take much for AI to connect dots. A client's name, email address, and company mentioned together in a single prompt is more than enough to identify them — and once that text is on someone else's server, you can't take it back.

Most VAs solve this by either not using AI on sensitive tasks (leaving productivity on the table) or by pasting everything in and hoping for the best (leaving trust on the table). There's a third option.

The workflow: anonymise, prompt, restore

The idea is simple. Before your text reaches ChatGPT or Claude, every sensitive detail gets replaced with a placeholder. [PERSON_1] instead of "Sarah Chen". [COMPANY_1] instead of "Acme Holdings". [EMAIL_1] instead of the real email address. The AI gets full context about the structure of your request without ever seeing the content that makes it confidential.

Then, when the AI returns its response, you swap the placeholders back for real names. The final output reads naturally — your client never knows the difference.

Here's what that looks like in practice:

The AI can still rewrite the email, change the tone, translate it, or summarise it. It just can't learn who Sarah Chen is, what company she works for, or how to reach her.

How the Chrome extension works (step by step)

The PromptSafe Chrome extension runs as a side panel alongside ChatGPT and Claude. You don't switch tabs. Drop in a document, review what's detected, and hit "Anonymise & Inject" — safe text goes straight into the chat input. When the AI responds, restore the real names in one click.

Five real VA scenarios

Here's what this looks like across the types of tasks freelancer VAs handle every day.

Your client data never leaves your browser

Here's the part that makes this different from other tools: PromptSafe processes your document content entirely in your browser. The detection engine, the anonymisation, the placeholder mapping, and the restoration all run in JavaScript on your machine. Your pasted text is never transmitted to any server.

This isn't a claim you have to take on faith. You can verify it yourself:

For VAs, this is a selling point. When a client asks "how do you handle my data when using AI?", you can honestly say: "Your names and details are replaced with placeholders before anything reaches ChatGPT. The real data never leaves my machine." That's a confident answer — and it's one most VAs can't give right now.

What PromptSafe does NOT do

If you've been burned by tools that promise privacy and then bury a data-sharing clause in page 14 of their terms, this matters:

What it costs (and why VAs find it worth it)

The free tier needs no sign-up, no email, no credit card. It handles up to 5 sensitive items per document — enough to test the full workflow on a short client email and see exactly how it works. For real client documents with dozens of names, emails, and phone numbers, the Pro plan at $5.99/month removes all limits and adds PDF redaction for formatted documents.

To put that in perspective: if PromptSafe saves you even 30 minutes a week in manual find-and-replace work, and you bill at $25/hour, the tool pays for itself in the first week. More importantly, it removes the mental overhead of constantly second-guessing what you can and can't paste into AI.