Updated
Updated · ZDNet · Jul 8
Claude Cowork Finds 8 Usable Sources in 7,000 Emails After Gemini Fails
Updated
Updated · ZDNet · Jul 8

Claude Cowork Finds 8 Usable Sources in 7,000 Emails After Gemini Fails

3 articles · Updated · ZDNet · Jul 8

Summary

  • A ZDNET journalist used Claude Cowork to scan more than 7,000 Gmail messages and turn them into article research, surfacing 12 PR pitches and narrowing them to 8 validated, publishable sources.
  • Gemini failed on the same task because the request required contextual judgment—finding themed emails, summarizing pitches, and flagging views on Anthropic's Fable 5 restrictions.
  • Claude was connected to Gmail with read-only access, then prompted to identify on-record statements with explicit permission to publish and preserve quotes word-for-word.
  • Links back to the original emails let the journalist verify every quote manually, cutting a task that normally takes hours down to a few minutes of due diligence.
  • The test highlights both the productivity upside of AI inbox tools and the privacy and accuracy trade-offs, with the Gmail connector disabled after the project ended.

Insights

As AI assistants read our emails, how do we balance their power against the profound privacy risks?
AI now handles our grunt work. How long until it automates the creative tasks we value most?
What 'hidden safeguards' in AI models might be secretly altering our results without our knowledge?