Verdict first
● Spending — ChatGPT 52.5%, Claude 29.4%
● Subscribers — ChatGPT 71.2%, Claude 27.1%
● The gap — Claude takes more money per subscriber
● Market — about 1.07 billion pounds a year, UK consumers
Most comparisons of these two argue about benchmarks. This one uses money. A working paper by Nikhil Datta and Peter John Lambert of the University of Warwick and Patrick Schneider of Imperial Business School analysed de-identified records from 14.2 million UK accounts, which makes it one of the few public datasets on what people actually pay rather than what they say they prefer.
The numbers
| Measure |
ChatGPT |
Claude |
Period |
| Share of spending, Dec 2025 |
73.5% |
7.4% |
start |
| Share of spending, Jul 2026 |
52.5% |
29.4% |
end |
| Change over seven months |
-21.0 points |
+22.0 points |
— |
| Share of active subscribers, Jul 2026 |
71.2% |
27.1% |
end |
| Spending share incl. app store estimate |
not stated |
22.8% |
end |
Why the two shares disagree
ChatGPT holds 71.2% of subscribers but 52.5% of spending. Claude holds 27.1% of subscribers and 29.4% of spending. Read those together and the conclusion is arithmetic rather than interpretation: Claude's average paying user spends more.
That points at tier mix. A product whose paying base skews toward higher tiers will take a larger share of money than of people, and the reverse is true for one with a large entry-tier population. Neither pattern is better on its own. A broad base of low-tier subscribers is a stronger position for raising prices later; a concentrated high-tier base is more revenue now and more exposed if those users leave.
Where these numbers stop
It excludes employer-purchased licences
This is consumer card and account data. Every seat a company bought for its staff is outside it. For two products sold heavily into teams, that is a large exclusion, and it means the study measures the consumer market rather than the market.
It counts accounts, not people
One person with two accounts appears twice. The direction of that bias is not obvious, which is a reason to treat the subscriber split as less firm than the spending split.
The app store adjustment is material
Including the authors' estimate of app store payments lowers Claude's July spending share from 29.4% to 22.8%. That is a 6.6 point swing on the headline figure. If you are citing a conservative number, 22.8% is the one to use, and anyone quoting 29.4% without the caveat is quoting the friendlier version.
What this does and does not tell you about which to buy
Almost nothing, directly. Market share is not a quality signal, and seven months of share movement in a market this young tells you more about pricing and distribution than about model capability. What it does tell you is that the two-horse framing is real at the consumer level in at least one large market, and that the gap is closing on money faster than on people.
If you want a capability comparison instead, our ChatGPT vs Gemini on real work takes that approach, and the METR assessment of Anthropic R&D covers how fast one side is moving.
Frequently asked questions
How big is the UK consumer AI market?
About 1.07 billion pounds a year, based on July 2026 spending rates, across 14.2 million de-identified UK accounts. This excludes employer-purchased licences.
Is Claude overtaking ChatGPT?
Not on these numbers. Claude closed the spending gap from 66 points to about 23 points in seven months, but ChatGPT still leads on both spending and subscribers, and holds 71.2% of active subscribers against 27.1%.
Why does Claude earn more per subscriber?
The data shows the gap but not the cause. The arithmetic points to tier mix: a paying base weighted toward higher-priced plans produces a larger share of revenue than of people.
Who produced this data?
Nikhil Datta and Peter John Lambert of the University of Warwick and Patrick Schneider of Imperial Business School, in a working paper titled Who Pays for AI? It analyses de-identified records supplied by a financial data aggregator.
Does this cover business use?
No. Employer-purchased licences are excluded entirely, which is a significant gap for two products with large enterprise businesses.
Sources