DuckDuckGo, a search company built on privacy, asked one thousand nine hundred and forty-four adults in the United States about their relationship with chatbots. The survey ran between 17 and 27 June 2026, with a margin of error of 2.2 percentage points.
What gets told to the machine and not to people
- Thirty-two per cent of those who use these tools had told one something they had not told a close friend, a parent or a doctor.
- That rises to fifty-six per cent among those who describe themselves as very enthusiastic about the tools.
- Among parents it is forty-three per cent, against twenty-five among everyone else.
And what they do not know
Fifty-three per cent did not know, or were not sure, that a conversation goes into training the model automatically unless the setting is changed. Only twenty-five per cent knew these conversations can be subpoenaed. And forty-three per cent were unaware of six basic facts about the privacy of these tools.
How they felt once they knew
When the arrangement was explained to them, fifty-eight per cent said they were uncomfortable with their conversations being used for training, thirty of them “not at all comfortable”. As for trust, fourteen per cent trust the companies behind these tools and thirty-nine do not trust them at all.
The survey is American, and its sample is weighted to the US population, so the percentages read as an indicator rather than a local figure. But the setting each user decides for themselves is the same wherever they are: whether the conversation goes into training or not.
Photo: two hands holding a phone beside a coffee cup. By Kristin Hardwick, CC0 via Wikimedia Commons.
Source: the DuckDuckGo blog, “DuckDuckGo Research: The AI Privacy Problem”, 25 August 2026.
Kuwait One, once a week.
The most useful stories we published this week, in your inbox.
No spam. Unsubscribe anytime.
