US Adults Who Have Used ChatGPT (Pew, Mar 2025)
34%
Share of US adults; Feb–Mar 2025 survey
+20pp since May 2023
Global Share — Ever Used Any Standalone Gen AI (Reuters Institute, Q1 2025)
61%
Pooled across ~30 countries; Jan–Feb 2025 survey
+21pp from Q1 2024
EU Individuals Using Gen AI Tools (Eurostat, 2025)
32.7%
Adults 16–74 who used gen AI in past year
First year with this Eurostat measure
US Adults Under 30 Who Have Used ChatGPT (Pew, 2025)
58%
Share of US adults under 30; Feb–Mar 2025 survey
+25pp since 2023

Data

Demographic / Geography% Using Gen AISourceYearDefinition
US Adults Under 3058Pew 20252025Ever used ChatGPT; Feb–Mar 2025 survey
US Postgrad Degree52Pew 20252025Ever used ChatGPT; Feb–Mar 2025 survey
US All Adults34Pew 20252025Ever used ChatGPT; Feb–Mar 2025 survey
US High School or Less18Pew 20252025Ever used ChatGPT; Feb–Mar 2025 survey
US Adults 65+10Pew 20252025Ever used ChatGPT; Feb–Mar 2025 survey
Global 18–24 (weekly)59Reuters Institute 20252025Used any standalone gen AI at least weekly
EU — Denmark42.0Eurostat 20252025Used gen AI tools in past year; ages 16–74
EU — Finland37.8Eurostat 20252025Used gen AI tools in past year; ages 16–74
EU — Sweden35.0Eurostat 20252025Used gen AI tools in past year; ages 16–74
EU Average32.7Eurostat 20252025Used gen AI tools in past year; ages 16–74
EU — Poland8.4Eurostat 20252025Used gen AI tools in past year; ages 16–74
EU — Romania5.2Eurostat 20252025Used gen AI tools in past year; ages 16–74
Global 55+ (weekly)20Reuters Institute 20252025Used any standalone gen AI at least weekly

About this Dataset

By February–March 2025, 34% of US adults had used ChatGPT — roughly double the 18% recorded in the summer of 2023, according to Pew Research Center. The pace of that doubling is the sharpest signal in this dataset: it took approximately 20 months to move a technology from early-majority trial to a level where roughly one in three American adults has direct personal experience with it. For context, Pew estimates smartphone ownership reached similar penetration levels over a period roughly three times as long.

Before reading further, note a critical methodological point. The three series in this page use different definitions and should not be treated as interchangeable. Pew Research Center asks whether US adults have ever used ChatGPT specifically — a brand-level, lifetime-use question. The Reuters Institute Digital News Report asks whether respondents across roughly 30 countries have ever used any standalone generative AI system, a broader definition. Eurostat asks EU individuals aged 16–74 whether they used generative AI tools in the past 12 months — a recency-bounded question. These distinctions are material: a person who uses only Microsoft Copilot or Google Gemini would be counted by Reuters and Eurostat but not by Pew. A person who tried ChatGPT once two years ago counts in Pew’s lifetime-use figure but may not meet Eurostat’s past-year threshold. The chart plots these series together to show the overall trajectory of consumer AI diffusion; the levels are not directly comparable.

The Reuters Institute’s 2025 report, drawn from surveys conducted in January and February 2025 across approximately 30 countries, recorded a global pooled figure of 61% who had ever used a standalone generative AI system — up from 40% in its January–February 2024 survey. That 21-percentage-point increase in 12 months is the largest single-year jump in the Reuters Institute’s tracking of any digital technology category. The sample skews toward higher-income countries with strong internet penetration, which is a limitation of panel-based cross-national surveys, but even with that caveat the directional signal is clear.

Reuters Institute data shows that among adults aged 18–24, 59% were using a standalone generative AI system at least weekly by early 2025 — a rate that suggests trial has largely converted to habitual use in that cohort. Among adults aged 55 and over, the weekly-use rate was 20%, indicating a larger and persistent gap between those who have tried the technology and those who use it regularly.

Eurostat’s 2025 reading — the first year its harmonised household ICT survey included a specific question on generative AI tool usage — showed 32.7% of EU individuals aged 16–74 used such tools in the past year. The EU-wide figure masks wide cross-country variation: Denmark leads at 42.0%, followed by Finland at 37.8% and Sweden at 35.0%. Romania records 5.2% and Poland 8.4%, placing them more than 30 percentage points below the Nordic leaders. This dispersion is consistent with broader EU digital economy indicators and reflects differences in infrastructure quality, language-model availability in local languages, and the sectoral composition of each economy.

The demographic gradient within the United States, as documented by Pew, is particularly pronounced. Adults under 30 report a 58% usage rate; those aged 65 and over report 10%. Postgraduate degree holders report 52% usage; those with a high school diploma or less report 18%. These gaps are comparable in magnitude to the early-adoption divides recorded for other major consumer technologies, but they have persisted further into the diffusion curve than those historical precedents might suggest — roughly two and a half years after ChatGPT’s public launch, the age gap has narrowed only modestly.

Key facts about this dataset’s coverage and scope:

  • US series (Pew): Three confirmed data points — May 2023 (14%), approximately August 2023 (~18%), and February–March 2025 (34%). A Q3 2024 estimate of approximately 24% is included as an interpolated intermediate point, consistent with the Pew report’s note that adoption roughly doubled from summer 2023 to early 2025.
  • Global series (Reuters Institute): Two confirmed annual snapshots — Q1 2024 (40%) and Q1 2025 (61%). Pooled across approximately 30 countries using online panels; sample skews toward higher-income, higher-internet-penetration markets.
  • EU series (Eurostat): One confirmed data point — 2025 (32.7%). The generative AI question is new in this survey cycle; no prior-year comparison from Eurostat is available.
  • Definitions differ across all three series — see the Q&A section and the methodology note above for full disclosure.

For strategy and investment teams, the most practical framing of this data is not the absolute adoption level — which is contested by definitional differences — but the rate of change and the demographic structure of who is adopting. The Pew doubling from 2023 to 2025 among US adults under 30 (from roughly 33% to 58%) suggests that the sub-30 cohort is approaching saturation on the brand-level trial question. The slower adoption among older and less-educated cohorts represents either a structural ceiling or a lagged diffusion curve; the data available through 2025 cannot distinguish between the two with confidence.

Frequently Asked Questions

The three series measure related but distinct phenomena and should not be treated as interchangeable. Pew Research Center asks US adults whether they have ever used ChatGPT specifically — a brand-level, lifetime-use question that excludes other AI tools such as Google Gemini, Microsoft Copilot, or Claude. The Reuters Institute Digital News Report asks respondents across roughly 30 countries whether they have ever used any standalone generative AI system, a broader definition that captures the full market. Eurostat's isoc_ci_ai_i asks EU individuals aged 16–74 whether they used generative AI tools in the past 12 months — a recency-bounded question that differs from both the Pew and Reuters lifetime-use framing. The practical consequence is that these series are not directly additive or directly comparable in level terms. The Eurostat 12-month figure of 32.7% is likely lower than a lifetime-use equivalent would be; the Reuters global figure of 61% using any standalone tool is higher than Pew's 34% partly because it uses a broader product definition and partly because it is a pooled global average rather than a US-only figure. Analysts using these data to benchmark adoption should cite each series' specific definition explicitly rather than treating generative AI adoption as a single, universally measured concept.
The Pew Research Center's February–March 2025 survey reveals a pronounced and consistent age-education gradient in US ChatGPT adoption. Adults under 30 report a 58% usage rate versus 10% among those aged 65 and over — a 48 percentage-point gap that is among the sharpest digital adoption divides Pew has documented for any consumer technology. The education gradient is similarly pronounced: 52% of postgraduate degree holders report having used ChatGPT, against 18% of those with a high school diploma or less. These two dimensions tend to compound: younger adults are more likely to have advanced degrees, and technology adoption in professional settings reinforces personal familiarity. The Pew report also notes that adoption roughly doubled overall between summer 2023 and early 2025, with particularly rapid growth among adults under 30 (from approximately 33% to 58%). The persistence of a large age gap through 2025 — two and a half years after ChatGPT's public launch — suggests that the diffusion curve for this cohort is not simply lagged: it may reflect structural differences in use-case relevance, workplace exposure, and digital comfort that narrow slowly.
Eurostat's 2025 data reveals a wide dispersion across EU member states that is not fully captured by the 32.7% EU average. Denmark leads at 42.0%, followed by Finland at 37.8% and Sweden at 35.0% — all Nordic countries, consistent with their generally high digital economy rankings. At the other end, Romania records 5.2% and Poland 8.4%, placing them more than 30 percentage points below the Nordic leaders. This dispersion is likely driven by a combination of factors: digital infrastructure quality, the availability of AI tools in local languages, educational attainment levels, and the sectoral composition of each country's workforce (economies with larger shares of knowledge-intensive service employment tend to score higher). The EU-27 average of 32.7% falls below the Reuters Institute global average of 61%, but methodological differences — Eurostat's 12-month recency window versus the Reuters lifetime-use question — account for a meaningful portion of that gap. Policymakers and strategy teams benchmarking EU digital competitiveness should note that aggregate EU figures mask bifurcated adoption patterns that may affect the distributional benefits of AI-driven productivity growth.
Survey questions about technology adoption typically distinguish between at least three states: awareness, trial, and regular use. 'Ever used' (the framing used by Pew and the Reuters Institute for their headline figures) measures cumulative trial — the share of adults who have attempted the technology at least once. 'Weekly use' is a more demanding threshold that indicates sustained, habitual incorporation into daily behaviour. The Reuters Institute's Generative AI and News Report 2025 reports both: 61% of respondents across ~30 countries had ever used a standalone generative AI system, while the weekly-use rate among adults aged 18–24 was 59% — suggesting that among younger adults, trial has largely converted to habitual use. Among adults aged 55 and over, the Reuters data shows a weekly-use rate of approximately 20%, well below their overall trial rate, indicating a larger gap between those who have tried the technology and those who use it regularly. For investment analysts and strategy teams, 'ever used' figures are the more relevant early-diffusion metric; they indicate market penetration and brand reach. Weekly-use figures better proxy for the engaged user base that generates recurring value for AI platforms and shapes the unit economics of consumer AI products. As the technology matures and the trial cohort saturates, weekly-use rates are likely to become the primary benchmark of meaningful adoption.
Each of the three primary sources carries specific limitations that users of this dataset should weigh carefully. Pew Research Center's surveys are US-only and ask specifically about ChatGPT by name, which may cause respondents who use other AI tools — Gemini, Copilot, Claude, Perplexity — but not ChatGPT to be classified as non-users. Pew's survey samples are large and probability-based, providing high methodological credibility within their defined scope, but the brand-specific question means the figures understate total US generative AI engagement. The Reuters Institute Digital News Report surveys are conducted annually across roughly 30 mostly higher-income countries and use online panels, which tend to oversample internet-connected adults and underrepresent populations with lower digital access. The Reuters Institute's global average thus likely overstates adoption in lower-income markets that are not included in the sample. Eurostat's isoc_ci_ai_i series has the advantage of mandatory statistical reporting standards across EU member states and a consistent questionnaire methodology, but the specific generative AI question was introduced only in 2025, making year-on-year trend analysis currently impossible from this source. The Eurostat definition — used generative AI tools in the past 12 months — is also narrower than lifetime-use questions, which means the 32.7% figure should not be read as a ceiling. Finally, none of the three sources captures enterprise or workplace AI use separately from personal use; workers who use AI tools exclusively through employer platforms may be classified differently depending on how each survey frames its questions.