Monday morning, I was digging through a client interview in Gemini
At 11 p.m. last Monday, I pasted in a set of customer interview summaries I’d compiled myself, hoping Gemini could help me spot the common themes. The next morning, I suddenly wondered: would content containing names, companies, and operational issues end up training the model? I paid for a membership and assumed that meant my content wouldn’t become “product.” I only realized later that payment and privacy are two different things.
This isn’t something that happens to a few users
HN user Chris noted that with a personal Google AI subscription, chat history may be used to improve Google services when Gemini’s “Keep Activity” setting is on. Google also says some conversations are reviewed by human reviewers, and even after you delete activity, human-reviewed content may remain for up to three years. Names, client details, quotes, and unpublished plans are not things to paste in casually. Temporary Chat doesn’t save history, so it works well for experimentation, but you don’t get the full conversation history afterward.
What it cost me to replicate this today: $0, about 10 minutes
Money: $0—no new membership needed. Time: about 10 minutes. Technical barrier: knowing how to tap through a phone or web settings. First step: open “Activity” in Gemini on the web, or “Settings” in the mobile app, find “Keep Activity,” and turn it off. Then run a test in Temporary Chat with no sensitive information. Note: after you turn it off, existing chats usually disappear within 72 hours, so save anything important locally or in your own encrypted notes. I could also just pause here; there’s no rush to test this, and not entering sensitive information yet is the more important move.
If I were in one of these three stages, this is how I’d handle it
Just getting started: I’d use only Temporary Chat, practice with public web pages, and avoid pasting client lists, contracts, or revenue screenshots.
Working with 1-2 clients: I’d replace names, company details, and core figures with “Client A” or “Product B” before asking Gemini to summarize, keeping the originals in my own document library.
Scaling up: I’d make a separate privacy checklist for every AI platform the team uses, review activity records, human review, and data-retention terms one by one, and add “Never paste client data directly” to our team agreement. This isn’t about panicking. It’s about not handing business secrets to a tool whose privacy rules you haven’t reviewed yet.