Lead
Google has pushed back against viral posts and some articles that say Gmail messages and attachments are now being used to train its AI models. A Google spokesperson told reporters the claims are misleading, that Gmail Smart Features predate the reports, and that Gmail content is not used to train the Gemini model. The company also notes a January settings update that separated personalization choices for Workspace and other Google products. Still, at least one Verge staffer reported being re-opted into smart features after previously opting out, prompting privacy concerns.
Key Takeaways
- Google says it has not changed anyone’s Gmail settings and denies using Gmail content to train Gemini AI, per spokesperson Jenny Thomson.
- Gmail Smart Features — which include spell check, order tracking and quick-add flight detection — have existed for many years, Google says.
- In January Google updated smart feature personalization to let users toggle Workspace features separately from other Google products.
- A Verge staffer reported they had been opted back into some Smart Features after earlier opting out, suggesting possible settings confusion or sync issues.
- Malwarebytes and viral social posts circulated claims that Gmail content and attachments were being used to train AI models and that opt-out required disabling Smart Features.
Background
Gmail includes a suite of convenience tools collectively called Smart Features. These have long offered spell checking and automated helpers such as detecting and summarizing travel itineraries or tracking orders from message content. For many users, those features enhance productivity by surfacing relevant information without manual input. Google has a separate set of personalization controls for Google Workspace customers that state Workspace content and activity may be used to personalize the Workspace experience when enabled.
In January Google adjusted its personalization settings to allow users to control smart features for Workspace independently from those in other Google products such as Maps and Wallet. That change was intended to give organizations and individual users clearer control over where personalization is applied. Still, the overlap of labels and the phrasing on settings pages has previously been a source of confusion. Security and privacy publishers monitor such changes closely because small UI or policy shifts can be widely interpreted as changes to data-handling or training practices.
Main Event
This week viral posts and an article referenced by security outlets claimed Google had altered its policy to use Gmail messages and attachments to train AI models, and that the only way to opt out was to disable Smart Features entirely. The specific allegation spread across social platforms and prompted wider media attention. Google disputed the characterization, with spokesperson Jenny Thomson telling reporters the reports are misleading and that Gmail content is not used to train the Gemini AI model.
Google also emphasized that Gmail Smart Features have been available for years and that recent personalization controls were designed to give users more granular choices rather than to change how content is applied to model training. Nonetheless, the episode was amplified when a Verge staffer reported being re-opted into some Smart Features after having previously opted out, highlighting a gap between company statements and some users’ experiences.
Security and privacy outlets such as Malwarebytes published accounts summarizing the viral claims, which increased public scrutiny. Google responded publicly to stem the spread of misinformation and to encourage users to verify their settings. The company reiterated that Workspace personalization language reflects personalization of the experience and not handing over emails to train Gemini.
Analysis & Implications
The incident spotlights a recurring tension in consumer tech between convenience features and user expectations about data use. Smart Features provide tangible benefits—autocomplete, order tracking and calendar suggestions—but they require analyzing message content. Even when companies say that content is not used to train external AI models, users may mistrust opaque settings or ambiguous language around personalization.
Separating Workspace personalization from other Google product personalization was a step toward finer user control, but it also creates new complexity. Users and administrators must now understand multiple toggles and how each relates to data retention, processing and downstream uses. That complexity increases the chance of accidental opt-ins and fuels speculation when public-facing statements are brief or technical.
From a regulatory and reputational perspective, high-profile confusion like this can accelerate calls for clearer privacy disclosures and stronger default protections. Regulators in several jurisdictions are already scrutinizing large models and the data used to train them; perceived ambiguity in settings or messaging can prompt investigations or policy proposals that demand greater transparency about training data sources.
Comparison & Data
| Setting | Main effect when ON | Main effect when OFF |
|---|---|---|
| Gmail Smart Features | Enable spell check, order tracking, auto-add flights to calendar | Loss of convenience helpers; no additional message analysis for features |
| Workspace personalization | Workspace content used to tailor Workspace experience | Workspace experience not personalized using Workspace content |
The table above summarizes what Google describes as the user-facing differences between the main toggles. It does not assert anything about internal Google training pipelines beyond Google’s public statements. Users who require strict separation of business content and personalization should review both Workspace and product-level toggles and consult organizational policies.
Reactions & Quotes
These reports are misleading — we have not changed anyone’s settings, Gmail Smart Features have existed for many years, and we do not use your Gmail content for training our Gemini AI model.
Jenny Thomson, Google spokesperson (official statement)
Google has changed its policy to use your Gmail messages and attachments to train AI models.
Malwarebytes (security outlet reporting the claim)
Public response included both calls for immediate settings checks and requests for clearer explanations from Google. Some users said they would disable Smart Features until they received firmer guarantees; others accepted Google’s clarification but urged simpler, clearer controls.
Unconfirmed
- Whether any internal or historic Google training datasets ever included aggregated, de-identified Gmail-derived signals beyond Google’s public statements is not confirmed here.
- The precise technical cause of the Verge staffer being re-opted into Smart Features (UI bug, sync issue, or other) has not been independently verified.
Bottom Line
Google’s official position is that it has not changed Gmail settings en masse and that Gmail content is not being used to train the Gemini model. The company points to longstanding Smart Features and a January personalization setting change as context for the confusion. Nevertheless, the incident underscores how quickly ambiguity around privacy settings can spread and erode trust.
For users concerned about data use, the practical steps are simple: review your Gmail and Workspace personalization settings and adjust the Smart Features toggles to match your privacy preferences. For policymakers and product teams, the episode is a reminder that clearer, more consistent language and defaults are needed to prevent misunderstandings and to maintain user confidence.