Several common beliefs about AI chat privacy are partly wrong. Deleting does not erase everything, incognito does not mean nothing is stored, and a shared link is public. Here are seven myths checked against what Anthropic and Google say in their own help pages.
Myth 1: Deleting a chat erases it right away
Fact: In Claude, a deleted chat leaves your history at once and is removed from back-end storage within 30 days, and longer holding is possible for legal or policy reasons. In Gemini, Google says reviewed chats are kept up to three years and deleting activity does not remove them.
Myth 2: Incognito or temporary means nothing is saved
Fact: Google says Temporary Chats are kept 72 hours. Anthropic says incognito chats are not used to improve Claude. Neither claim is the same as storing nothing.
Myth 3: Turning off training removes my old chats from models
Fact: Anthropic says data already in training runs started, or in models already trained, remains included.
Myth 4: A shared chat link is private to the person I sent it to
Fact: Anthropic says anyone with the link can view the snapshot. Google says anyone with the link can read the chat, and recipients can reshare it.
Myth 5: Voice chats keep my voice forever
Fact: For dictation in Claude's mobile apps, Anthropic says the audio recording is deleted after speech is converted to text, and it does not use your voice for training. The transcript is kept like any message.
Myth 6: Companies cannot tell where I am
Fact: Anthropic says it estimates your country or region for abuse prevention, which cannot be switched off, and may estimate city level location for features like web search, which can.
Myth 7: A well written message must be from a real person
Fact: CISA says perfect grammar is not a reliable sign of safety, because AI can produce flawless messages.
What to take away
The pattern is consistent. Controls work going forward, they have exceptions, and the safest data is data you never typed. For a short list, see our posts on what not to paste and on which control protects what.
Teaching it at home
Pick one myth a week and look at the actual setting together. Children and teenagers learn more from finding the toggle than from a lecture. Parents can check the same pages for any tool the family uses, since the details differ by vendor and change over time.
Quick check
Which single habit helps most?
Keep sensitive details out of the chat. Every vendor control has limits, and none can unsay what you typed.
Should I trust a summary like this?
Treat it as a pointer. Read the vendor's current help page, because settings and terms change.
Sources
- How long do you store my data? (Anthropic Privacy Center)
- Is my data used for model training? (Anthropic Privacy Center)
- Share and unshare chats (Anthropic Privacy Center)
- Dictation on the Claude mobile apps (Anthropic Privacy Center)
- Does Claude use my location? (Anthropic Privacy Center)
- Gemini Apps Privacy Hub (Gemini Apps Help)
- Share your Gemini chats (Gemini Apps Help)
- Recognize and Report Phishing (CISA Secure Our World)
Last reviewed: 19 October 2026.
