Americans appear comfortable using artificial intelligence, but much less inclined to check what it creates. EY's AI Sentiment Index 2025 shows a gap between confidence in AI applications and the habits people use to verify AI-generated output.
The pattern is not that U.S. respondents reject AI outright. Instead, the survey points to a more specific behavior: many see AI as useful and responsive, yet relatively few review or refine its texts, images, or translations before relying on them.
U.S. users check AI output less often
According to EY's AI Sentiment Index 2025, just 24 percent of U.S. respondents say they review AI-generated texts, images, or translations. That places the United States among the lowest countries in the study and below the global average of 31 percent.
The finding matters because reviewing is one of the simplest ways users can catch mistakes, mismatches, or weak outputs. The source data does not say why each person skips that step, but it does show that the behavior is less common in the U.S. than in several major markets.
Other countries reported higher review rates. South Korea reached 42 percent, while China and India each reported 40 percent. Germany stood at 27 percent. Only Sweden and France reported lower rates than the U.S.
That comparison puts the American number in perspective. U.S. respondents are not just slightly below the most active reviewers; they are below the global average and near the bottom of the surveyed group.
High confidence may reduce the urge to verify
The survey also found that 74 percent of people in the U.S. believe AI applications understand their needs. That is slightly above the global average of 73 percent.
Those two findings sit side by side: many U.S. respondents think AI tools understand them, while relatively few say they review the results. The data suggests that perceived usefulness and accuracy may reduce the felt need to check AI-generated output.
There is also a practical tension. If every AI result had to be carefully inspected, the speed and efficiency that make the technology appealing would be weaker. That does not mean review is unnecessary, but it helps explain why users who find AI helpful may not want to slow down and examine every answer.
The same pattern appears when the question shifts from reviewing to editing. Only 14 percent of U.S. respondents say they fine-tune AI-generated content. That again places the United States near the bottom internationally.
China and India reported much higher revision behavior, with 32 percent of users regularly revising AI output in each country. France, the U.K., and Japan reported even lower rates than the U.S.
The survey behind the findings
EY's AI Sentiment Index 2025 is based on a global survey of more than 15,000 people in 15 countries. The survey was conducted between December 2024 and February 2025.
The research used random stratified sampling with quotas for age, gender, and location. EY used that approach to ensure representative results in each country.
The survey was also supported by chat-based interviews with 135 participants in February 2025. Those interviews were used to explore attitudes toward AI in more detail.
All results were statistically weighted so each country had equal influence in the findings. That weighting matters because the report compares attitudes across countries rather than simply aggregating all respondents into one overall pool.
Medical AI draws more caution in the U.S.
The same survey shows that American attitudes change when AI moves into more sensitive areas. In medical settings, U.S. respondents were less comfortable than the global average.
For example, 49 percent of U.S. respondents said they would feel comfortable with AI predicting potential health problems based on their data. The global average was 57 percent. India reported 76 percent, and China reported 74 percent.
Comfort was lower for basic medical consultations with an AI "general practitioner." Just 31 percent of U.S. respondents said they would consider it, compared with a global average of 37 percent.
China and India again reported higher comfort levels. In both countries, 61 percent of respondents were comfortable with this kind of AI-driven care.
The contrast is important. U.S. respondents may be willing to accept AI output in everyday contexts without much checking, but that does not mean they are broadly comfortable with AI in every domain. Health-related uses appear to prompt more hesitation.
Misinformation remains a shared concern
Concerns about AI-powered misinformation are widespread in the United States. Seventy-five percent of Americans said they are worried about fake news and deepfakes.
That figure matches the global average of 75 percent. In other words, U.S. concern about fake news and deepfakes is not unusually low, even though U.S. review and revision rates are low.
Taken together, the findings show a complicated AI relationship. Americans often see AI applications as useful and personally responsive. They are among the least likely to review or revise AI-generated output. Yet they also show caution around medical AI and share global concerns about fake news and deepfakes.
For anyone building, deploying, or using AI tools, that combination is the central takeaway. Trust, convenience, and caution are not moving in one simple direction. In the U.S., the survey suggests that many people accept AI assistance quickly, but reserve deeper skepticism for areas where the stakes feel more personal or socially sensitive.