1 in 3 Americans Received Wrong Calculation Results Using AI. A Study of 1014 Participants (2026)
Report Highlights
-
1 in 3 Americans have received incorrect math results from AI, yet 38% now use it as their primary calculation tool.
-
Only 21% of users trust AI completely, with 57% citing potential mistakes as their primary reason for skepticism.
-
Gen Z is twice as likely to use AI for math as Boomers, despite 46% fearing it will erode their personal skills.
-
Users prioritize AI for explaining complex “word problem” steps and financial context over simple numerical accuracy.
-
AI models like DeepSeek and ChatGPT show instability rates up to 69%, often replacing one error with another.
Digital habits are shifting as AI enters our work and personal routines. This trend raises a basic question: Can we actually trust these tools to do the math? We ran a survey and asked over 1000 people.
Our study found that 62% of Americans now use AI for calculations. However, even with that high adoption, about 1 in 3 users (35%) say they’ve received incorrect answers.
While 55% of users say they trust AI for math, 45% remain skeptical.
The trust in AI is also fairly shallow.
Only 21% trust AI “completely,” meaning they expect 90%-100% accuracy. Nearly half (46%) place their trust in the 60%-90% range, while 34% trust it only slightly or not at all.
There are regional differences in the United States, too. People in the South are 13 points more likely to “mostly” trust AI compared to those in the West.
How much do you trust AI with calculations? | Response rate | West | Midwest | Northeast | South | |
|---|---|---|---|---|---|---|
Completely 90%-100% | 21% | 28% | 24% | 20% | 19% | |
Mostly 60%-90% | 46% | 36% | 44% | 47% | 49% | |
Moderately 30%-60% | 20% | 17% | 20% | 21% | 20% | |
Slightly 10%-30% | 8% | 14% | 6% | 8% | 6% | |
Not at all 0%-10% | 6% | 5% | 7% | 5% | 6% |
When asked what reasons they would distrust AI, respondents primarily cited the risk of AI making mistakes, with 57% citing errors as a top concern.
This baseline skepticism is fueled by a fear of losing human skills through over-reliance (30%) and a smaller but notable unease regarding privacy (14%) and the black-box nature of AI reasoning (13%).
Interestingly, a subset of the population manages this distrust by compartmentalizing AI’s role; while they remain wary of its judgment, 28% explicitly trust the technology for objective calculations.
Reason for distrusting AI | Percentage of total |
|---|---|
AI can make mistakes | 57% |
Fear of over-reliance / losing skills | 30% |
Security/privacy concerns | 14% |
Lack of transparency in reasoning | 13% |
Prefer human judgment or online tools | 13% |
I trust AI for calculation | 28% |
A paradox exists between the youngest and oldest generations. Gen Z uses AI more than any other group for learning, yet they harbor an underlying anxiety: 46% of Gen Z fears that using AI will lead to a loss of their own skills. Contrast this with Boomers (ages 62-79). Even though Boomers use AI less frequently, 35% say they trust it for calculations, compared to only 20% of Gen Z.
There is also a large 40-percentage-point gap regarding educational value. 54% of Gen Z use AI specifically because it "explains the steps," compared to just 14% of Boomers.
For the younger generation, AI is a multi-purpose mentor, while for Boomers, it remains a specialized tool they approach with surprising optimism.
The benchmark backs up these concerns and highlights AI’s limitations when it comes to calculations.
A crucial measure of AI performance is not just whether an answer is right or wrong, but how stable those answers are. We call this the instability metric. It looks at how often a model changes its output from one attempt to the next, even when the answer remains incorrect.
This behavior can happen in three ways:
- A wrong answer changes to a different wrong answer;
- A correct answer becomes wrong; or
- A wrong answer becomes correct.
The first two cases highlight why AI cannot yet replace a traditional calculator. You might get an answer that looks plausible, but it could change if you ask the question again.
AI does not have the fixed reliability of a standard calculator. Our testing found that ChatGPT corrected 65% of its repeated mistakes, but the new answers were still often incorrect. DeepSeek was the most unstable, changing its output 69% of the time. While Gemini and Grok had instability rates of 46% and 55%, respectively.
Standard calculators follow fixed rules. Large Language Models are probabilistic, meaning they predict the next likely word rather than calculating in a traditional way. This approach causes the AI to “drift,” often fixing one error only to create a new one.
Relying on AI for calculations means accepting a margin of error where the logic can literally shift mid-sentence