Claude to clean an inbox with approval and rescue

Severin Sorensen used Claude Cowork to reduce unwanted mail in his own Gmail accounts. The agent inspected the inbox, presented recommendations and worked through the browser after Sorensen approved the proposed actions.

His first account reports forty-four subscriptions removed and four filters created, with news-source whitelists intended to preserve wanted mail. Sorensen watched the browser work and opened a second window to help with another account. He then asked Claude to capture the process in a Markdown reference.

A later firsthand guide explains the corrections that made the cleanup more useful than mass deletion. A broad search for a hotel name also matched a valued colleague, so the workflow switched to an exact sender domain. Desired real-estate alerts had reached spam and needed rescue before purging. The guide describes these as actual discoveries from the sessions, not hypothetical precautions.

Sorensen reports a quieter inbox the following day and a roughly two-hour initial exercise. Longer annual time savings are estimates rather than measured outcomes. The original prompt excerpt comes from his published follow-up instruction. The transferable task is reviewing sender choices, executing only approved cleanup and checking for wanted messages, with the human still deciding what correspondence matters.

Prompt
Original prompt

Source: www.linkedin.com · Severin Sorensen