The State of AI Literacy · A yearly report · 2026
We analyzed 4 independent research reports from the first half of 2026, plus new Canadian business and consumer data, measuring who uses AI, who trusts it, and who's worried about it. Their findings all point the same way: adoption is racing ahead of understanding. This is a mirror, not a scoreboard. Find yourself in it.
Half of American adults use an AI chatbot, but fewer than one in five people feel confident doing it.
This probably describes you or someone close to you. You opened an AI bot because your manager said to, or because everyone else was doing it. You got a decent answer. Underneath it, you wondered: do I trust this, and do I understand what I just did?
That voice is a sign you're thinking. Four separate research teams spent 2026 measuring people's understanding of AI, and their findings were consistent. We're using AI far faster than we understand it. Younger people use it the most and trust it the least. The experts who build it are twice as optimistic as the public who lives with it.
I built AI Literacy Guide on one idea: AI literacy isn't learning every tool, it's knowing how to think. This requires three skills: understanding roughly how AI works, using it on real tasks, and questioning what it produces. The charts in this dashboard show that people are good at using AI, but they're skipping the other two.
So read this as a mirror, not a scoreboard. Find your age, your income, and your area of work. See where you sit.
Find yourself in the data
Pick your age bracket. Every chart below with an age breakdown, using Pew's U.S. survey data, will highlight your group.
Tap an age group above. See how you compare to your age group's adoption rate, confidence level and AI anxiety level.
The same data lands differently depending on where you're sitting. Find the version that's yours.
If you use AI at work
You're probably using it quietly, half-wondering if you're allowed to or doing it right. The confidence you feel you're missing is learnable, and it's mostly about knowing what to check in an answer, not what to type into the box. Start there and the rest gets easier fast.
Start here: the four questions to ask any AI answer →If you lead a team
Adoption is already at 49% and climbing, but only 18% feel confident, and the experts-versus-public optimism gap (73% vs 23%) tells you the understanding is landing unevenly across your org. That gap is a training problem, not a tools problem, and it's the one you can fix.
Read: AI at work →If you're a parent
The loudest anxiety in this data is about what AI does to how we think and whether we can trust what's real online. You're not overreacting. And helping your kids here is less about understanding the tech than about teaching them to question it, which you already know how to do.
Read: what to tell your kids about AI →Share of U.S. adults who say they ever use an AI chatbot such as ChatGPT, Gemini or Copilot. Every bar on this dashboard shares the same 0–100% scale so magnitudes stay honest across panels.
Daily use (almost constantly, several times a day, or about once a day) is shown below for the two cuts Pew published: gender and race/ethnicity. Overall daily use is 24% of all adults.
A second, independent measure: OECD.AI tracked deduplicated monthly visitors to ChatGPT, Claude and Gemini across the 37 GPAI member countries, divided by population. This is observed web traffic, not a survey, and it is a global population, not a U.S. one, so treat it as a companion signal rather than a direct comparison to the Pew panel above.
GPAI-country average. The Q4 2025 quarterly figure was 27%.
Netherlands also reported near or above 50%. Japan and Turkiye more than doubled usage rates in 2025.
Share of all chatbot visitors 35 or older, not an adoption rate. Under-35 share fell from a majority (62%) to 52%.
Stanford HAI's AI Index compiles its own population-adoption analysis, likely built on related web-traffic methodology but computed and reported separately from the OECD.AI piece above. Its numbers are close to, but not identical to, OECD.AI's (Singapore 61% here versus 63% above), a reminder that "adoption rate" depends on exactly how and when it's measured. Treat the two as two independent estimates of the same rough phenomenon, not a single dataset.
Generative AI adoption correlates strongly with GDP per capita, per the AI Index. Estimated consumer value of generative AI tools to U.S. users reached $172 billion annually by early 2026. Separately, Ipsos found that 58% of employees surveyed worldwide use AI at work on a semiregular or regular basis, rising above 80% in India, China, Nigeria, the United Arab Emirates and Saudi Arabia. Source: Stanford HAI, "The 2026 AI Index Report," Economy chapter, April 2026.
Pew's 2026 survey does not ask chatbot users to rate their trust in the accuracy of AI-generated answers. The closest published measure is self-reported confidence in one's own ability to use an AI chatbot, plus separate questions on confidence in government and companies to handle AI responsibly.
Pew published confidence-in-use figures only by gender and age. Race/ethnicity, income and party breakdowns were not reported for this question, so they are omitted rather than estimated.
These two questions were reported by party affiliation only; Pew did not publish age, gender, race/ethnicity or income breakdowns for institutional confidence in this report.
A separate 2025 survey, the Civic Health and Institutions Project (CHIP50), asked all 50 states whether federal AI regulation goes too far, doesn't go far enough, or if they're not sure. Unlike the Pew panel, this one publishes an education breakdown.
Nationally, 41% said regulation doesn't go far enough versus 27% who said it goes too far, with the rest unsure. Concern about too little regulation rises steadily with age, peaking at 51% among adults 65 and older, and with education, from 34% among those with a high school education or less to 46% among college graduates. Source: CHIP50 2025, as compiled in the Stanford HAI AI Index Report 2026.
A 30-country Ipsos survey asked whether people trust their own government to regulate AI responsibly. The United States ranks last.
A separate Pew global survey asked a related but different question: how much people trust the EU, the U.S. and China specifically to regulate AI effectively, regardless of their own nationality. Across 25 countries, the median respondent trusted the EU most (53%), then the U.S. (37%), then China (27%). Within the U.S. itself, opinion split close to evenly between trusting (44%) and distrusting (47%) its own government on this question. Sources: Ipsos AI Monitor 2025 (30 countries, n=23,216, fielded Mar 21–Apr 4, 2025) and Pew Research Center's Spring 2025 Global Attitudes Survey (25 countries), both as compiled in the Stanford HAI AI Index Report 2026.
Share of U.S. adults who predict AI will have a very or somewhat negative impact on society, and on them personally, over the next 20 years. Full demographic breakdown available for both questions.
"AI is advancing too quickly" was asked of all adults (63% overall). Pew published an exact breakdown only by gender: men 58%, women 68%. Age-group and race/ethnicity differences were described only as ranges in the report text, so specific figures are not charted here.
Stanford's AI Index compiles a separate strand of research: matched surveys of the U.S. general public and AI experts, run by Pew Research Center, the Forecasting Research Institute and Elon University. This is not the same survey as the Adoption, Trust or Anxiety panels above, so its figures sit in their own section rather than folded into those charts.
Share saying AI will have a positive impact over the next 20 years, by domain.
U.S. public
AI experts
% saying AI will lead to fewer jobs for this occupation
Public and experts roughly agree on cashiers and software engineers. They diverge sharply elsewhere: experts see far more risk for truck drivers (62% vs. 33%) and lawyers (38% vs. 23%), while the public sees more risk for teachers (43% vs. 31%) and doctors (28% vs. 18%) than experts do. Source: Pew Research Center, "How the U.S. Public and AI Experts View Artificial Intelligence" (McClain et al., 2025, based on a survey of 5,410 U.S. adults and 1,013 U.S.-based AI experts, Aug 2024) and "How Americans View AI and Its Impact on People and Society" (Kennedy et al., 2025, 5,023 U.S. adults, Jun 2025), as compiled in the Stanford HAI AI Index Report 2026.
A separate national survey of 1,842 Canadian adults, fielded May 7–11, 2026 (margin of error ±2 points). This is a different country and a different survey instrument than the Pew and OECD.AI panels above, so read it side by side, not merged in.
The Angus Reid AI Assessment Index sorts Canadians into these four groups based on trustworthiness and perceived impact across four questions. Two in three land Negative or Critical. Age and gender are not strong predictors of which group someone falls into; the more consistent divide is political (Conservatives lean more positive, New Democrats much more negative) and economic, shown below.
Disagreement is lowest among the lowest earners (62%, tied with the overall figure) and peaks in the upper-middle band ($100K–$150K, 73%), not a clean straight line by income. Households under $50K and over $200K land at nearly the same point.
"Active" use means typing prompts directly into a tool. "Passive" means encountering AI-generated content without asking for it. Angus Reid found that more active use correlates with more favorable views of AI, so this gap in engagement style may be feeding the gap in trust.
Among Canadians who never use an AI chat tool, a third see no benefit at all, versus 1–3% among monthly-or-more users.
Frequent use is climbing regardless: the share of Canadians using AI several times a day rose from 10% (Nov 2025) to 16% (Jun 2026). Trust in AI-generated information has not caught up: almost no one calls it "very trustworthy" (1%), though 43% call it "trustworthy enough." Source: Shachi Kurl and Dave Korzinski, Angus Reid Institute, "AI Divide: Canadians worry technology will widen inequality," published June 9, 2026, based on an online survey of 1,842 Canadian adults conducted May 7–11, 2026 via the Angus Reid Forum panel, margin of error ±2 percentage points for comparison purposes.
A second Canadian lens, from official statistics: Statistics Canada surveyed 9,251 businesses in April and May 2026. If hands-on experience is what builds comfort with AI, that experience is being handed out unevenly, by employer size, by industry, and by geography.
Tripled in two years. Industry spread is wide: 42.3% in information and cultural industries versus 4.5% in agriculture.
Rural businesses adopt at less than half the urban rate.
Large employers train staff on AI at nearly three times the rate of the smallest ones.
This is a business survey, not a survey of individuals, so it is not merged with the Angus Reid charts above. But the two tell one story: Angus Reid shows active AI use rising with personal income and education; Statistics Canada shows the training and exposure that build that skill concentrating in larger, urban, tech-adjacent employers. Source: Statistics Canada, "Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026," Canadian Survey on Business Conditions, released June 11, 2026.
A third Canadian lens, from EY's 23-market consumer study (1,018 Canadian respondents): a small but growing share of Canadians now let AI act for them, not just advise them, and health is where the stakes show first.
Delegation is growing faster than trust: most Canadians insist on human oversight, and more than three quarters worry about telling real content from AI-generated content, even as one in four takes health guidance from AI without a doctor involved. EY classifies Canada as a lagging market for AI adoption overall. Source: EY Global AI Sentiment Study (wave 2), 18,152 respondents across 23 markets including 1,018 in Canada, as reported in "As AI moves from advice to authority, who defines its limits?", EY Canada, May 6, 2026.
One more Canadian contrast, offered with caveats: in Deloitte's State of AI in the Enterprise (a survey of 175 Canadian business and IT leaders, fielded August to September 2025), 86% expected meaningful productivity gains from AI. Set against the 56% of Canadians who expect AI to widen inequality, that is Canada's own version of the optimism gap between the people deploying AI and the public living with it. Treat it as directional only: the sample is small, leaders only, and fielded months earlier than the surveys above.
Chatbot ever-use plotted against personal-impact anxiety, by age and by race/ethnicity, U.S. data. If adoption calmed anxiety, these dots would converge as adoption rises. They don't.
Ever use a chatbot vs. expect a negative personal impact
Ever use a chatbot vs. expect a negative personal impact
Adults 18–29 lead every group in adoption (66%) but also lead in personal-impact anxiety (37%): heavy use has not translated into comfort. Asian adults show the opposite pattern: highest adoption (70%) paired with the lowest anxiety (20%). White adults sit near the middle on adoption (46%) but report the highest personal-impact anxiety of any racial or ethnic group (33%). Three countries, three surveys, one consistent theme: using AI more and trusting AI more are separate things that move independently, not two names for the same thing.
Same numbers, three readers. Find the one that sounds like you.
Half your coworkers use a chatbot. Fewer than one in five feel confident doing it. The unease you feel in meetings is the majority position, not a gap in your ability.
Your move: build one habit. Treat every AI answer as a first draft, and verify anything that matters outside the chat window.
Start with AI at work →You're in good company. 63% of adults say AI is moving too quickly, and the people most anxious about it are often the ones who understand it least. Nervousness tends to fall as understanding rises.
Your move: trade the worry for one clear question: how does this thing work? Start there.
What AI literacy means →Younger adults use AI the most and trust it the least, and school policy is years behind. Your kids will meet AI long before anyone teaches them to question it.
Your move: skip the ban-or-allow fight. Teach the skill that lasts: asking whether an answer is true, and how they'd check.
The guide for families →Pew Research Center. "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact," published June 17, 2026. Survey of 5,119 U.S. adults, American Trends Panel Wave 187, conducted Feb. 17–23, 2026, margin of error ±1.6 percentage points for the full sample. White, Black and Asian adults include only those who report a single race and are not Hispanic; Hispanics are of any race. Asian adult estimates represent English speakers only. Income tiers are adjusted for household size and cost of living: lower income is below $51,900, middle income is $51,900 to $155,600, upper income is above $155,600, in 2024 dollars scaled to a three-person household. Education was used only as a survey-weighting variable in this report, never published as a demographic breakout for adoption, trust or anxiety questions, so it does not appear here. Where Pew reported a range rather than an exact per-group figure (age or race/ethnicity differences in "AI is advancing too quickly"; race/ethnicity, income and party differences in confidence using chatbots), that cut is omitted rather than estimated.
OECD.AI. Jeff Mollins, Clemence Descubes, Luis Aranda and Sara Rendtorff-Smith, "How people are using GenAI chatbots: Evidence from web traffic data," published June 30, 2026. SimilarWeb traffic data for ChatGPT, Claude and Gemini across 37 GPAI member countries, Feb. 2024 to Mar. 2026, divided by IMF population projections. This measures observed web visits, not survey responses, and covers a global population, not a U.S.-specific one, kept in its own panel rather than merged with Pew. Two figures are derived rather than quoted directly: the "under 35" shares of the chatbot user base are calculated as the complement of the "35 and older" shares the article states directly. This specific OECD.AI article does not report a U.S. figure. An earlier version of this dashboard flagged a widely circulated claim that "the U.S. ranks 24th at 28.3%" as unverifiable and excluded it; that figure has since been traced to Stanford HAI's AI Index Report 2026, not to OECD.AI, and is now shown in its correct place below with its correct source.
Angus Reid Institute. Shachi Kurl and Dave Korzinski, "AI Divide: Canadians worry technology will widen inequality," published June 9, 2026. Online survey of 1,842 Canadian adults conducted May 7–11, 2026 via the Angus Reid Forum, a panel built to be representative of the Canadian population by region, gender, age, household income and education. Margin of error would be ±2 percentage points, 19 times out of 20, for a probability sample of this size, for comparison purposes. This is a Canadian population, distinct from Pew's U.S. sample and OECD.AI's global one. The AI Assessment Index breakdown by age, gender and federal vote is described qualitatively in the report text rather than charted here: the underlying chart's exact per-category values could not be reliably reconstructed from the source PDF's text layer, so rather than guess, those splits are quoted narratively instead of plotted.
Stanford HAI, AI Index Report 2026. Published April 2026, 425 pages, compiling data from many original sources rather than running its own single survey. The figures used on this dashboard trace back to five distinct underlying studies the AI Index cites: Pew Research Center's "How the U.S. Public and AI Experts View Artificial Intelligence" (McClain et al., 2025: 5,410 U.S. adults and 1,013 U.S.-based AI experts, Aug 2024) and "How Americans View AI and Its Impact on People and Society" (Kennedy et al., 2025: 5,023 U.S. adults, Jun 2025); the Forecasting Research Institute's Longitudinal Expert AI Panel; the Ipsos AI Monitor (30 countries, n=23,216, fielded Mar 21–Apr 4, 2025); Pew's Spring 2025 Global Attitudes Survey (25 countries); and the Civic Health and Institutions Project (CHIP50, 2025, fielded across all 50 U.S. states). Each of these is a different survey population and instrument from the "Americans and AI 2026" Pew report used elsewhere on this dashboard, even where both are Pew studies, so figures from this section are not merged with the Adoption, Trust or Anxiety panels above. The AI Index's own Economy-chapter adoption figures (U.S. 28.3%, Singapore 61%, UAE 64%) are shown separately from OECD.AI's adoption figures for the same reason: related methodologies, different measurements, different exact numbers.
Canadian supplementary sources. Three additional Canadian sources appear only in the Equity section, each kept separate from the Angus Reid charts because they survey different populations. Statistics Canada, "Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026" (released June 11, 2026): official statistics from the Canadian Survey on Business Conditions, 21,105 businesses sampled and 9,251 responding, collected April 1 to May 6, 2026; this measures businesses, not individuals. EY, Global AI Sentiment Study wave 2, as reported in "As AI moves from advice to authority, who defines its limits?" (EY Canada, May 6, 2026): 18,152 consumers across 23 markets including 1,018 in Canada, random stratified sampling with quotas by age, gender and location; the article reports toplines only, so no demographic breakdowns are charted. Deloitte Canada, "State of AI in the Enterprise: A Canadian perspective" (March 2026): 175 Canadian business and IT leaders within a 3,235-respondent global survey, fielded August to September 2025; because the sample is small, leaders-only and fielded earlier than every other source here, it contributes one clearly-caveated contrast sentence and no charts.
No values on this dashboard are inferred, extrapolated or merged across studies. Every figure is attributed to its source at the point of use. If you spot a number that looks off, the full dataset lists the exact question wording and source page for every data point shown.