Can AI erase your personal data from the internet? What the viral prompt actually does.

No, it cannot. An AI assistant cannot submit opt-out requests to data brokers for you, cannot log in anywhere as you, and cannot compel any website to delete anything. The prompt doing the rounds promises all three. The exposure it is pointing at is real. The one-click fix is not.

What makes this worth writing about is not the prompt. It is why so many people shared it without ever checking whether it worked.

What the prompt claims

The format varies, but it is always some version of: paste your name and email into an AI assistant, ask it to find everywhere your personal data is exposed, and have it submit removal requests to the data brokers holding it. It reads as empowering. It went a long way.

What actually happens when you run it

Two things, and both are instructive.

The first is that a careful assistant refuses. Someone who tried it posted the reply they received, and it is worth reading closely:

"What you're asking me to do is search the open web for a specific person's name and email combination, scrape data-broker profiles, and compile whatever personal data I find, addresses, phone numbers, relatives, into a table."

Read that back as a description rather than a complaint. Compiling a named individual's address, phone number and relatives into a single profile is precisely the mechanism behind doxxing and stalking. The assistant has no way to verify that the name in the box belongs to the person typing it. Refusing is the correct behaviour. It is not the tool being broken or over-cautious, it is the tool declining to build a dossier on a stranger.

The second is simpler and was flagged underneath the same post by someone in the comments: the tools cannot do the task anyway. They cannot file opt-outs on your behalf. Where an assistant does not refuse, it tends to produce something worse than a refusal, which is a confident, tidy, plausible answer that nobody has checked.

Why this keeps happening

This is the pattern worth taking away, because it will repeat with the next viral prompt.

A person who is unsure hesitates. They say "I think so" or "let me check". An AI assistant does neither. It produces a fluent, confident answer whether or not it is correct, and it cannot tell the difference. So a prompt that promises a capability the tool does not have still returns something that looks like a result, and that output gets screenshotted and shared as proof.

Then the second failure stacks on the first. Nobody verified it. The prompt spread on the strength of how good it felt to read, not on whether anyone had run it and checked. That is the whole loop, and it is the same loop that empties advertising budgets: confident output, nobody checking, real consequences. We wrote about that pattern in the context of AI search and how engines actually decide who to name.

The exposure is real, though

None of this means you should relax. If anything the viral post is right about the problem and wrong about the cure. Search your own name properly and most people find more than they expected: old accounts, directory and classified listings, domain registration records with a home address in them, business filings, and passwords from a service you forgot you signed up for in 2016.

Here is what actually reduces it. All of it is manual, and that is the point.

• Search yourself deliberately. Your full name in quotes, then your phone number, then each email address you use. Check the image results too, not only the web tab.
• Check your credentials against known breaches. If a password of yours has appeared in a breach, change it everywhere you reused it. Reuse is the actual vulnerability, not the original breach.
• File data broker opt-outs one at a time. There is no shortcut and no assistant that does this for you. Each broker has its own form and most require you to confirm from your own email.
• Turn on WHOIS privacy for your domains. Domain registration is a common and completely overlooked source of a home address and personal phone number.
• Clean your own website. Staff email addresses published in plain text, old admin accounts nobody closed, and test pages left online are exposure you control completely.

Where this stops being personal and starts costing your business

Here is the part that gets missed, and it is the reason we care about this at all.

Reconnaissance comes first in most attacks on a small business website. Nobody starts by attacking your server. They start by reading what is already public: which email addresses exist, who has admin access, what your staff post, what your domain record reveals, and which of those passwords already appeared in a breach.

A reused password from an old leak, an admin address printed on a contact page, or enough personal detail to make a convincing password-reset call is how most of the sites we recover were actually entered. The malware is the last step, not the first. By the time you see casino spam on your homepage, the interesting part already happened weeks earlier.

So the honest version of the viral prompt is much less exciting and much more useful. You cannot delegate this to an assistant. You can go and look, methodically, at what is exposed, and close it.

And if your site has already been hit, the order matters: containment first, not panic. Our guide to hacked website recovery and malware removal walks through what actually happens, including a real recovery you can verify independently.

Frequently asked

Related reading