Google denies 'misleading' reports of Gmail using your emails to train AI (www.theverge.com)

Google is pushing back against viral posts and articles alleging it quietly changed Gmail’s policy to use users’ emails and attachments to train AI models unless they disable “smart features.” A Google spokesperson told The Verge the reports are misleading: Gmail Smart Features have existed for years and Google does not use the content of users’ Gmail to train its Gemini model. The company also notes a January settings change that separated smart-feature personalization controls across Workspace and other Google products; however, some users (including a Verge staffer) reported being re‑opted into certain features, fueling confusion. For AI/ML practitioners and privacy-conscious users, the episode underscores an important distinction: personalization features (spell check, order tracking, calendar extraction) can use Workspace content to tailor user experience without necessarily contributing to model training datasets. Google’s settings language — agreeing to let Workspace use content and activity to personalize experience — can be read as permissive for feature personalization but, per Google, not as permission to include emails in training corpora. The takeaway for the community: don’t assume corporate product telemetry equals training data availability; transparent, granular controls and clear disclosures about data provenance and use (personalization vs model training) are critical for ethical dataset sourcing and regulatory compliance. Users should still review their settings if they care about personalization or data use.