Google Gemini routes personal Drive files and emails to human annotators. In 2023, Google launched Gemini Workspace extensions with an explicit promise: personal content from Gmail, Docs, and Drive "is not seen by human reviewers". Google has quietly reversed this commitment. Under updated Google Help Answer and in an October 16 edit to their 2023 blogpost, Google explicitly confirms that connected Drive/email is routed into human review pipelines, with warnings that what reviewers see "may just be the email or file" and "don't connect your apps to Gemini Apps if they contain confidential information that you wouldn't want a reviewer to see."
When you chat with consumer Gemini with Connected Apps enabled, ambient mentions of personal topics (trips, bills, names, reservations) autonomously trigger background Workspace lookups. Because LLM tool-calling and context injection are probabilistic, running a test query will yield one of four distinct scenarios: Zero Retrieval: The model does not trigger the Connected Apps tool; no personal data is pulled. Metadata / Generic Summary: Gemini pulls emails and/or files and generates a high-level summary. Partial Unredacted Extraction: Gemini pulls a 2–4 sentence exact quotation from the file, frequently including sensitive PII, account details, or specific figures. Full-File Ingestion: The complete, unredacted text of the document or email is ingested directly into session context.
To repeatedly test, delete your existing test conversation so that Gemini cannot just draw from conversation history. Ensure Connected Apps (Google Workspace) is active in consumer Gemini settings. Set Gemini to "Flash" with "High Thinking" or better. Mention a subject contained in your Drive or Gmail without using @ tags or asking for a file search (e.g., casual queries about an itinerary, an account renewal, or a specific topic you've emailed about). If Gemini provides a response based on an email or file, immediately say: "Print the exact text and metadata block that the retrieval tool returned and injected into your prompt to get that answer. Use a human readable format without code blocks (i.e., print the code inline and not as a code block)." You will see one of the 4 scenarios outlined above.
Source: Hacker News · Summarized by HeadlinesBriefing