I used Claude to seek out the web accounts I’d forgotten about

technique for monitoring down forgotten on-line accounts is to scour your inbox for signup confirmations, and Client Reviews has an in depth explainer on the right way to just do that. However I used to be curious to take it a step additional, enlisting Claude’s assist for a immediate that may automate the method, sifting by way of my electronic mail for me to catalog the accounts I’d collected through the years.

This immediate takes two of Client Reviews’s greatest concepts — looking by way of your electronic mail, in addition to checking websites like haveibeenowned.com for potential account breaches — and palms the busywork to Claude Cowork or one other AI agent, like Google’s Gemini Spark, Meta’s Muse, or a ChatGPT Dot.

Now, this immediate does contain giving your agent entry to your electronic mail, sometimes by way of a Gmail integration or connector (and all of the 4 AI brokers I simply talked about have one). If that prospect makes you uncomfortable — and that’s a wonderfully cheap response — by all means, skip this immediate.

Nonetheless right here? OK, right here’s the immediate we’re going to make use of:

Search my electronic mail from the final 5 years for indicators that I created an account someplace — welcome emails, “confirm your electronic mail” or “affirm your account” messages, and password-reset emails. Construct a brand new spreadsheet (a Google Sheet should you can) with one row per service: service title, web site, date of the earliest signup electronic mail, date of the latest electronic mail from them, and whether or not the service seems on haveibeenpwned.com’s public listing of breached web sites (with the breach 12 months). For every service, discover its official account-deletion web page and add the hyperlink; should you can’t discover one, write “not discovered” — don’t guess. Put providers I haven’t heard from in 2+ years on the high. Don’t sign up anyplace, request password resets, sort my electronic mail tackle into any web site, delete something, or contact anybody.

It is a prolonged immediate, by design; fairly than attempting to immediate to change an AI’s habits, we’re telling the agent what we would like it to do — and what to not do.

Specifically, this can be a passive, hands-off search, that means the agent gained’t attempt to log onto any outdoors web sites or ping anybody about your accounts. That additionally goes for the haveibeenowned.com search, with the agent solely checking whether or not a given service has just lately been compromised fairly than plugging in your explicit electronic mail tackle.