When Your VA Starts Using AI Quietly
The tell is always the same
You get a report back faster than usual. The phrasing is a little too smooth. There's a sentence structure your VA has never used before, something like "it's worth noting that" or a bullet list that appeared out of nowhere in a task that used to come back as three rushed paragraphs. You don't have proof. You just have a feeling that something changed.
That feeling is usually correct. Somewhere in the last year, most VAs working with content, admin, or customer support picked up a habit of pasting your instructions into ChatGPT and pasting the output back to you. Nobody trained them to do it. Nobody told them not to. They just found it, it worked, and they kept using it without mentioning it.
This isn't a scandal. But it changes what you're actually paying for, and if you run any kind of remote team, it's worth understanding exactly what's happening instead of guessing.
Why they don't tell you
Put yourself in their position for a second. If a VA tells you "I used AI to draft this," the two outcomes they can picture are: you get annoyed that they're not doing the work themselves, or you decide the task is worth less money now that a tool is doing part of it. Neither outcome benefits them. So the safer move is to say nothing and just deliver the work.
There's also a simpler reason. Most VAs don't think of pasting a task into a chat window as "using AI" in any formal sense. It's the same category as looking something up on Google or using a spell checker. They're not hiding a secret system, they're using a tool the same way they use any other tool that makes them faster.
What's actually happening on their end
If you want to know what this looks like in practice, it's almost always one of a few patterns.
For writing tasks, they take your bullet points or a rough brief and drop it straight into a chat interface, then copy the response with light edits. For research tasks, they ask a model to summarize a topic instead of reading five sources themselves. For customer replies, they draft the reply, run it through AI to make it sound more polished, then send it. For repetitive admin, some have started using AI to write short scripts or spreadsheet formulas they don't fully understand themselves.
None of this requires any technical setup. It's a browser tab. There's no agent wired into your systems, no automation touching your files, nothing running unattended. It's a human using a tool between two manual steps, which is a very different risk profile than what people picture when they hear "AI in my business."
The actual risks, not the imagined ones
The risk isn't that your VA is being replaced by a robot or that they're secretly doing nothing all day. The risk is narrower and more boring than that, which is exactly why it's easy to miss.
First, quality drift without your knowledge. If a VA is having AI draft customer-facing replies and only lightly editing them, tone and accuracy can drift in ways you won't catch until a customer flags something wrong. AI models get facts wrong confidently. If nobody who understands your business is checking the output before it goes out, errors ship under your name.
Second, data exposure. If someone pastes your internal notes, customer details, financial figures, or draft contracts into a public AI chat tool, that data has left your control. Most consumer AI chat products retain conversation history by default unless it's explicitly turned off, and you have no way to audit what's been pasted in over the last six months. This is the part that should actually worry you, not the writing style change.
Third, skill atrophy on tasks you're paying for expertise in. If you hired someone for judgment, like triaging support tickets or making editorial calls, and they've quietly outsourced the judgment part to a model, you're paying for a layer of human review that may not be happening the way you think it is.
None of this means AI use is bad. It means undisclosed AI use removes your ability to manage the actual risk.
What I actually do about this
I don't run a large VA team, but I do run agents and automation into real production systems every day, rendering content, publishing across brands, wiring tools together with MCP. So I think about this from both sides: as someone who builds automation and as someone who has to trust a human doing manual work on my behalf.
The fix isn't a ban. Banning AI use just pushes it further underground and you lose even the illusion of visibility you had before. The fix is making disclosure the default and building a boundary around what can and can't go into a prompt.
Concretely, that means telling every VA directly: using AI tools is fine, tell me when you do it, and never paste customer data, financial numbers, passwords, or anything under NDA into a public AI chat tool. That last part is the one that actually matters. Most VAs will follow this without pushback once you say it plainly, because most of them have never been told it's a problem in the first place.
For anything customer-facing, I want a human check on the actual facts before it ships, not just a read-through for tone. AI-polished writing sounds confident even when it's wrong, so a fluent sentence is not the same as a verified one.
For repetitive tasks, if a VA has found that AI genuinely speeds up a task, that's useful information. I'd rather know so I can decide if the task should be restructured or automated properly, with a controlled tool instead of an ad hoc chat tab that has no audit trail.
Where this connects to bigger automation decisions
This is a small version of a question that comes up constantly once you start automating more of a business: where does the AI have access, and what happens if it gets something wrong. A VA quietly using ChatGPT in a browser tab is the lowest-stakes version of that question. It's a human still in the loop, reading the output, deciding whether to send it. Compare that to an agent wired into your CRM or your file system with permission to act on its own, and the stakes go up by an order of magnitude, because now there's no human reading step in between.
The lesson scales in both directions. If a browser-tab AI habit with zero technical setup can quietly change what your VA delivers without your knowledge, that should make you more careful, not less, about what you connect agents to when you're the one setting up the automation. Every tool an agent can call, every file it can read, every system it can write to, is a version of the same question: what happens when this is confidently wrong and nobody catches it before it acts.
What to actually say to your team
If you manage VAs and haven't addressed this directly, do it this week. It's a short conversation, not a policy document. Tell them AI tools are fine to use for drafting and research. Tell them to flag when they've used it, especially on anything customer-facing. Tell them never to paste sensitive data into a public AI tool. That's the whole conversation, and it closes most of the real risk without pretending the tools don't exist.
The VA using AI quietly isn't the problem. The quietly part is.
If you want more on how I actually run AI tools and agents in production, without the hype, [head back to the homepage](/).
Get new guides and videos first — join the Telegram channel.