· OpenAI · Report Translation · 8 min read
From Asking to Doing: How the World Puts ChatGPT to Work
OpenAI has released country-level data on ChatGPT usage for the first time. In the accompanying Signals dataset, Singapore ranks first among 147 countries in per-capita message volume. The proportion of "getting it to do things" in work contexts is more than twice that in non-work contexts; multimedia is the fastest-growing use case; and the share of messages from people aged 35 and above is rising in nearly every country—Singapore is one of the rare exceptions.
Original: From asking to doing: How the world is putting ChatGPT to work Data: OpenAI Signals — individual data (CC BY 4.0) Release date: August 6, 2026 Translation: Nix (AI-assisted translation, human-reviewed). The “Site Note” section at the end was added by Singapore AI Observatory and is not part of the original.
AI is shifting from being something people ask questions of to something people use to actually get things done.
OpenAI has released country-level data for the first time, showing how people around the world use ChatGPT. These findings describe a technology that is moving beyond simply “providing answers” toward “doing things”—it is stepping out of the early adopter circle and becoming useful in more ways to more people.
- From asking to doing: In work contexts, people are more than twice as likely to use ChatGPT to complete a task or produce an output—writing, coding, analysis—than in non-work contexts.
- AI adoption is going global: Countries in Latin America, Africa, and Oceania are catching up with early adopters, and the global adoption gap is narrowing.
- AI is going beyond text: Multimedia is the fastest-growing use case globally, now accounting for 7.8% of all messages, exceeding one in ten messages in countries like Brazil and Colombia.
- AI is reaching beyond early adopters: Usage among people aged 35 and above is rising in nearly every country. In France and Czechia, this demographic’s message share has increased by more than 10 percentage points over the past year.
Together, these data point to a clear shift: AI is no longer just helping people find answers—it is helping more people in more places actually get things done.
This new data is available to view or download on OpenAI Signals—OpenAI’s economics research team’s data, research, and analysis hub. These country-level statistics can help the public, policymakers, and researchers in nations worldwide better understand how over 1 billion people are using ChatGPT.
The Signals dataset reflects only messages sent from ChatGPT Free, Go, Plus, and Pro accounts—that is, accounts typically managed by individuals rather than organizations.
In Work, “Doing” Dominates
This data confirms a trend that has appeared in recent news, especially since the launch of ChatGPT Work: people are increasingly using this technology for professional tasks. In these scenarios, users more frequently engage in “doing”—using ChatGPT to produce a result or execute a task.
Such work includes editing, coding, and analysis. Our data shows that globally, people are more than twice as likely to use ChatGPT in this way in work contexts as in non-work contexts. Outside of work, usage is more exploratory: “asking”—seeking information and clarification—remains the largest category.
The Adoption Gap Is Narrowing
Continuing the steady global upward trend we recorded in June, countries and regions that started with lower per-capita ChatGPT adoption rates are gradually catching up with established early adopters.
For the second quarter of 2026, our team updated the cross-country usage ranking. The accompanying chart shows how the per-capita leaderboard has changed since the first quarter of 2026.
Notably, in the second quarter, usage growth in Latin America, Oceania, and parts of Africa outpaced the rest of the world, with Peru, Uruguay, and Costa Rica rising the most in global rankings. Adoption in North America and Europe continues to rise, but parts of the Southern Hemisphere are catching up in per-capita adoption rates.
Note: The ranking change chart covers 144 countries, with blue indicating ranking increases, brown indicating decreases, and gray indicating no change; hatched countries are excluded or do not meet criteria.
Multimedia Is ChatGPT’s Fastest-Growing Use Case
As new capabilities emerge, specific use cases capture people’s attention and drive adoption. The generation, analysis, and retrieval of multimedia is one example. Since the release of ChatGPT Images 2.0 in April 2026, the global share of messages focused on multimedia has risen to 7.8%. It still lags behind leading use cases like practical guidance, writing, and information queries, but has been steadily climbing since the start of the year.
Latin American Countries Stand Out in “Doing and Analyzing Media”
Diving deeper into the multimedia message category, the data reveals clear differences between countries and regions. In Latin American countries like Brazil and Colombia, more than one in ten messages sent to ChatGPT are classified as multimedia.
Note: Multimedia message share in the second quarter of 2026, by country. This updated version includes 126 countries with qualifying second-quarter multimedia estimates; hatched countries are outside the scope of this release or lack qualifying data.
Global ChatGPT Adoption Among 35+ Year-Olds Is Growing
Quarter-over-quarter changes in per-capita adoption rates vary not only by usage patterns and geography, but data also reveals differences across age groups.
Trend lines over the past year show that the share of messages sent by 35+ year-olds is rising in nearly every country. Year-over-year, these users now account for 5% more of messages compared to 12 months ago.
This analysis covers only users who self-reported their age on the ChatGPT platform.
Looking at countries individually, adoption among 35+ year-olds is growing particularly rapidly in several nations. For example, the relevant charts show that the share of messages from 35+ users has shifted notably in several European countries. In France and Czechia, this share has increased by over 10 percentage points over the past year.
Nearly three-quarters of European countries saw growth above the average, while in some Southeast Asian countries—such as Singapore—the opposite pattern emerged. Overall, six out of eight countries in the region saw this share rising, though the magnitude was very small.
Note: Three-month moving average of the message share from self-reported users aged 35 and above, with changes relative to each country’s own average for Q2 2025. The gray line represents 111 countries with complete estimates; the dashed gray line represents the median across countries. The June 2026 endpoint compares the most recent quarter to Q2 2025. Data source: OpenAI Signals updated public CSV release.
About OpenAI Signals
OpenAI Signals is a platform through which OpenAI’s economics research team shares data and research with the public. Beyond research, we also regularly release updates on how individuals and businesses use ChatGPT. Data can be downloaded from the Data and Methodology page.
Site Note: Singapore’s Position in This Data
The original article mentions Singapore only once, and as a counterexample—the message share of 35+ year-olds barely changed. The number that is actually relevant to Singapore is in the accompanying data.
In the ranking of per-capita message volume for Q2 2026 (April–June), Singapore ranks first among 147 countries.
| Rank | Country/Region |
|---|---|
| 1 | Singapore |
| 2 | Malta |
| 3 | Netherlands |
| 4 | Montenegro |
| 5 | Latvia |
| 6 | United Arab Emirates |
| 7 | Estonia |
| 8 | Lithuania |
| 9 | Azerbaijan |
| 10 | Luxembourg |
Other references in the same ranking: Australia 19th, South Korea 25th, New Zealand 40th, United Kingdom 41st, United States 51st, Japan 57th, Malaysia 59th, Vietnam 88th, Philippines 92nd, Thailand 96th, Indonesia 101st, India 109th. The gap within Southeast Asia spans roughly ninety positions.
This is already the third vendor’s dataset placing Singapore at or near the top:
- Anthropic (March 2026): AI Usage Index 5.53, 1st among 116 regions. See our analysis
- Microsoft (January 2026): Generative AI adoption rate 60.9%, 2nd globally, second only to the UAE. Full translation
- OpenAI (August 2026): Per-capita message volume, 1st among 147 countries. The UAE ranks 6th in this ranking.
Three companies, three different metrics, three different denominators, one consistent conclusion. This essentially rules out the explanation that “a particular product simply happened to sell well locally.” Singapore is small, wealthy, English-speaking, densely packed with knowledge workers, and has had seven consecutive years of consistent policy pushing in one direction—the accumulation of these conditions yields a per-capita ceiling.
That +1.2 Percentage Point
Singapore’s entry in the original is that users 35 and older saw their message share rise by only 1.2 percentage points, while the median across countries is 5.1. Other countries marked on the same chart: Czech Republic +12.6, France +10.1, South Korea +8.0, United States +7.8.
This figure is easy to misread. It measures message composition, not adoption level. A flat share could mean that users 35 and older haven’t kept up, or it could mean that younger users’ usage grew at the same rate—both sides expanding in sync, ratio unchanged. This data cannot distinguish between the two, and moreover, age is self-reported by users, and the sample itself carries inherent bias.
In the Singapore context, the second interpretation makes more sense: in a market where per-capita usage already ranks first globally, people across all age groups have likely already adopted it, leaving limited room for incremental growth. What’s truly worth pursuing is not this adoption percentage, but rather the “depth”—what it’s being used for, how much work time it has replaced. That belongs to enterprise-side data, which the personal version of Signals cannot show.
Policy Implications
In Singapore’s public discussions, “AI adoption rate” is still often treated as a KPI. These three pieces of independent data show that this phase has basically ended: widespread adoption itself is no longer the bottleneck.
The next question is that word in the original title—from asking to doing. Globally, the proportion of “getting AI to do things” in work scenarios is more than twice that of non-work scenarios. For an economy where per-capita usage has already hit its ceiling, the marginal value does not lie in getting more people to open ChatGPT, but rather in getting those already using it to shift from “asking” to “doing,” and in making this happen within formal enterprise processes, not just on individual employee accounts.
The personal version of Signals explicitly does not include enterprise accounts and Codex. In other words, Singapore’s number-one global ranking only measures personal usage. The enterprise-side accounts have not yet been opened.
Reference Sources
- From asking to doing: How the world is putting ChatGPT to work (OpenAI, August 6, 2026)
- OpenAI Signals — individual data (data updated on August 6, 2026)
- Signals Data Download and Methodology
- Citation format: Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoe Hitzig, Drew Johnston, Alex Martin Richmond, Christopher Ong, Carl Yan Shan, and Kevin Wadman, “OpenAI Signals v2.0” (CC BY 4.0)