AI Power Users · Singapore
Singapore uses AI more than almost anywhere.
It gets no speed advantage from it.
Two rival AI companies publish what their models are used for. Both put Singapore at or near the top of the world. Neither shows Singapore getting more out of a conversation than anyone else. What they do show is a country using AI to understand things rather than to produce them.
The paradox
Same leverage as everyone else
Anthropic estimates how long each task would take a person unaided, and how long it takes with AI. Run both numbers for Singapore and for the world and you get the same answer to four significant figures.
Singapore
7.077×
4.93 hours unaided, 41.8 minutes with AI
The world
7.074×
4.73 hours unaided, 40.1 minutes with AI
A difference of 0.04%. Being at the top of the adoption tables buys Singapore nothing extra per conversation. In April the release runs it the other way, with Singapore marginally behind. There is no compression advantage in either direction.
What it does buy
Harder problems, not faster ones
If the leverage per conversation is identical, the difference has to be in what gets brought to the model. It is.
12.43
years of education the task needs
World 11.79
24.6%
involve more than one distinct task
World 22.9%
90.0%
could have been done without AI at all
World 87.6%
That last number is the one that matters. Nine in ten Singaporean tasks are ones the person could have completed unaided. This is not AI standing in for capability that was missing. It is augmentation of people who could already do the work.
The behaviour
Singapore asks. The world tells.
OpenAI sorts every message into asking, doing, or expressing. Singapore leans to asking, and it does so in both halves of life.
At work
Outside work
Singapore world average
Singapore has asked more than the world at work in 24 of 24 months, without a single exception, and outside work in 22 of 24. Anthropic's data agrees from a different angle: Singapore over-indexes on using AI to learn, and on working alongside it rather than delegating to it, in both published months.
Who is doing it
Work, not homework
Globally these figures are inflated by students. Singapore's are not. This is what makes the country's position in the tables mean something.
Singapore
46.1%
of messages are work-related
The world
30.4%
of messages are work-related
Stable across every month in the release. The global figure is weighted by message volume, so it is the average message rather than the average country.
What for
To find things out, not to make things
Anthropic groups requests by topic. Ranking Singapore's against the world's gives a clear shape.
Above the world
Below the world
Percentage points against the global share, May 2026. Coding assistants are excluded from both datasets, so software work is undercounted everywhere. The relative position holds; an absolute claim about Singapore and code would not.
The part that reverses
The world is moving on from writing. Singapore, slowest.
Writing is shrinking as a share of what people bring to AI, everywhere. The obvious reading is that Singapore's edge is eroding. It is the opposite: Singapore's writing share is falling too, just more slowly than everyone else's, so the gap has grown.
Over these two years Singapore's writing share fell 6.5 points. The world's fell 8.3. The gap widened rather than closed. Whether that means Singapore is slower to broaden what it uses AI for, or that writing is simply more embedded in the work done here, this data cannot say. It only rules out the reading everyone reaches for first.
Look it up
Find a job
Here is where almost every write-up of this data goes wrong, so the tool shows you the trap rather than walking into it. Anthropic classifies the task in the conversation, then maps it to the jobs that perform that task. It does not know the user's occupation. So for many jobs the conversations are people buying that service, not people doing it.
Try
of Singapore's Claude conversations involve this job's tasks
against the world
Singapore publishes 255 of 718 occupations, so this is the published subset. Figures are May 2026. The work and personal split is measured globally, because Singapore's rows carry the share only.
What we are doing about it
Two commitments
01
We will re-run this every quarter
Both companies keep publishing. The claim that Singapore's lead is narrowing has a clock on it and nobody is currently tracking it. Every figure on this page is generated from the raw releases by a script, so the next update is one command rather than a fresh round of checking. We will publish what moved, including the parts that go against us.
02
We are publishing the workings, mistakes included
The first version of this analysis was wrong in four places. It said Singapore had climbed to number one, when in fact a country had simply dropped out of the dataset. It read an occupation table as a statement about workers. It credited Singapore with a shift that was really the global average moving. Those errors were caught by checking, not by instinct, and the record of that is more useful than a clean report would have been.
AI Power Users is a Singapore community for people putting AI to work in their day-to-day. That is who this was written for.
See the communitySources and licensing
Where these numbers come from
Anthropic Economic Index, release 26 June 2026, covering April and May 2026. Licensed CC-BY. Massenkoff, Lyubich, Sacher, Hitzig, Zhang, Heller and McCrory, “Anthropic Economic Index report: Cadences”, 2026. Source
OpenAI Signals, release 6 August 2026, covering July 2024 to June 2026. Licensed CC BY 4.0. Chatterji, Cunningham, Deming, Hitzig, Johnston, Martin Richmond, Ong, Shan and Wadman, “OpenAI Signals v2.0”. Source
Modifications. All figures on this page were recomputed from the raw released files. Charts are ours, not reproductions. Nothing here is published by Anthropic or OpenAI in this form.
What this data cannot tell you
- Both indices measure one company's product. Neither is a measure of AI use in general.
- Per-capita rankings divide messages by resident population. A city-state with heavy visitor traffic is structurally flattered by that.
- Singapore is first on OpenAI's ranking for the second quarter of 2026, and second on Anthropic's for May, having been first in April. The position is not stable, and we do not claim it is.
- Occupation labels describe the task in the conversation, not the person typing. Anthropic say so themselves.
- Coding assistants are excluded from both releases, so technical work is undercounted throughout.
- OpenAI's figures cover consumer accounts only and, in their own words, understate business use.
- Anthropic's release covers two months. Where the two disagree, we have said so rather than picked one.