68% of America's daily AI users told Gallup they are worried about the technology

Somebody opened a chatbot this morning to pull together a deck, shipped what came back, and would tell a pollster the same afternoon that the technology worries them. Gallup put both questions to the same people. Its survey covers recipes, homework, work research and presentations as the everyday uses ^[21], and it finds 68% of Americans who use AI daily are worried about it ^[1].

Set that against the figure for all Americans. Gallup has 74% reporting feelings of worry, so the heaviest users are six points less worried than the country as a whole ^[6][16].

Trust in the output is a separate question with a separate answer. In the US, 45% of daily users said they trust AI results a lot ^[11]. The other 55 of every 100 heavy users are working with material they do not fully believe ^[17]. Canada runs the same pattern at a smaller scale: 29% of respondents there said they use AI daily, more than half of those had concerns, and 64% of all Canadians aware of AI said it worried them ^[7].

The global picture is not gloomy. In 34 of the 37 countries, the most common feeling reported was curiosity, about 32% of all respondents said they felt worried, and positive emotions were most common in China at 90%, Singapore at 89% and Kenya at 87% ^[13].

"One of the clearest findings is that attitudes toward AI are multidimensional. People can be curious about AI, expect it to be beneficial and use it frequently while still being worried about it. Similarly, they can see considerable potential in AI while remaining skeptical about whether the information it provides is accurate," Pablo Diego-Rosell, senior scientist at Gallup, told TechCrunch ^[14].

The worry has referents outside anyone's product. TechCrunch points to tech companies laying off hundreds of thousands of employees in order to invest in AI, shrinking entry-level openings, and Anthropic CEO Dario Amodei's warning that the technology will wipe out jobs en masse ^[19].

What a product team can act on is the split between two measures that usually get collapsed into one number. The first is return without prompting: the share of users who come back inside seven days when no notification, email or banner has gone out. The second is acceptance of output: the share of generated text that ships unedited, plus citation clicks and re-prompts per completed task. Four cells come out of that. Returns and ships unedited is trust, and returns and rewrites is reliance without trust. Does not return but ships once is an errand, and the user who does neither is a dead feature.

Measuring the second axis has a price, and it is paid in the demo. You have to leave the seams visible: sources on the claim, a diff against the original, an edit path that costs one click rather than a copy-out into another tool. That makes first-run time-to-value longer and the walkthrough less impressive. Take the seams out and you lose the signal along with the users who were checking.

Gallup's US numbers say where those users are concentrated. In a weekly retention chart, cell one and cell two look identical.