Britain's £400 billion AI growth figure came from Google

For two years British ministers said AI would add £400 billion to the economy by 2030. On 26 October the AI minister was asked what the number rested on, and answered in writing: a report commissioned by Google. The government's own central estimate is £120 billion — less than a third of the promise.

The answer, in a footnote

The question came from Ben Spencer, the Conservative shadow Cabinet Office minister, and it was narrow: what is the evidential basis for the statement in the AI Opportunities Action Plan that AI adoption could add £400 billion by 2030, and will the government publish the underlying analysis? Kanishka Narayan's written answer is four sentences long and does two things at once.

First, provenance: the £400 billion "was drawn from a publicly available 2023 report commissioned by Google and produced by Public First". Second, a re-basing. AI adoption and data investment could add 0.4 to 1.3 percentage points to productivity growth, he said — "hugely uncertain" — equivalent to £55 billion to £140 billion of gross value added in 2030, with a central estimate of £120 billion. He did not withdraw the old figure or call it wrong. As Wired noted when it reported the exchange, this appears to be the first time a minister put the two numbers side by side, and the first time the Google commission was stated on the record.

The provenance was never actually hidden, which is the part worth sitting with. The Action Plan, published in January 2025, puts the £400 billion claim in its section on private-sector adoption barriers and footnotes it to Public First's report, Google's Impact in the UK 2023. Follow that footnote and you land on Google's own announcement of the research: a study of Google's contribution to the British economy, which concluded that generative AI could create "over £400 billion in value" by 2030, equal to 2.6% of annual growth, saving the average worker more than 100 hours a year and 700,000 hours of admin for doctors and teachers.

The method behind those numbers, as The Times reported at launch, was to analyse 17,000 tasks and assume AI cuts the time each one takes by half. That is a scenario, conditional on adoption, and the report said so. A policy document then quoted it as a projection, and two governments repeated the projection.

What actually moved was the adoption rate

£120 billion is 30% of £400 billion. It is tempting to read the gap as a verdict on the technology — the AI got less impressive, so the benefit shrank. That is not what PF Global, the consultancy formerly called Public First, says happened. Its head of economics, Jonathan Dupont, told Wired the disparity is partly down to slower take-up: the government's numbers assume "a much slower rate of adoption than looked possible back in our model that was created in 2023". Dupont adds that on that slower path its central estimate is now much closer to the government's.

So the downgrade is an indictment of British adoption, not of AI. Every billion pounds of the gap is a firm that never trained its staff, a workflow nobody rewired, a back office still doing by hand what a model could draft. That is precisely the failure a national AI plan exists to fix — and the plan's headline number was the thing telling everyone the plan was working.

Wired's report on the exchange has a second finding that should annoy people. Ministers have had the lower estimate since at least November 2025, almost a year before the written answer, and Skills England — the agency responsible for apprenticeships and skills training, which is to say the exact lever the adoption argument depends on — was still publishing the £400 billion figure four months later. The big number did not die when it was superseded; it kept circulating because nobody with a press release wanted to be the one who shrank it.

Why the hop matters more than the number

The mechanism is ordinary, which is why it's worth writing down. A company commissions research about its own sector. A consultancy produces a defensible scenario with conditions attached. A government adopts the top-line figure, drops the conditions and builds a plan around it. Ministers repeat it, agencies repeat it, industry repeats it, and by the third hop the only trace of the interest behind the number is footnote 19.

Google's interest here is obvious and not sinister on its own terms: it paid for a study arguing that Britain should adopt AI faster. PF Global says its research is independent and its judgments its own, and that it continues to work for Google. What is not defensible is the state presenting a vendor's optimistic scenario as its own forecast — and doing it for two years, across a change of government, while its own analysts had a much lower figure in a drawer.

That mattered because the number was load-bearing. Lord Holmes of Richmond, who asked in January 2025 where it came from and got no answer, told Wired that "billions of pounds of our money, and how the government chooses to deploy it, depend on it". Dame Chi Onwurah, who chairs the Commons science committee, put the failure mode more sharply: the central barrier to digitising public services is hype, and overstated benefits "undermine the case for change".

The winners are whoever sells adoption. A government that believes AI is worth £400 billion will subsidise compute, tolerate the power bills, wave the planning applications through and defend the buildout when voters object. The losers are the next set of numbers, including the honest ones — the same credibility problem we flagged in Gartner's $2.7 trillion AI spending forecast and in the claim that the AI buildout is running at 3.6% of GDP a year. Forecasts of this kind are only as good as the assumptions a reader is allowed to inspect.

The skeptical case, which is not weak

Three objections deserve air. First, the footnote was public for 21 months. Anyone could have read it — including the MPs who are now indignant, and including every journalist who repeated the figure. Second, revising a number down is what competent government looks like, and burying the revision in a written answer to a shadow minister is the least performative way to do it. If the criticism is "they corrected it quietly", the alternative was correcting it loudly, which almost no government chooses.

Third, and most uncomfortable for everyone involved: £120 billion is not more auditable than £400 billion. Nobody can forecast a general-purpose technology's macro contribution seven years out, and the gap between the two figures is the gap between two guesses about adoption speed. Neither has been tested against firm-level evidence. If you want a number you can defend, you want take-up data — what share of British firms used AI in a production workflow this quarter, and did output per hour move — not a forecast.

What to watch

Whether the Action Plan's published text gets corrected, or the footnote simply stops being quoted — Matt Clifford wrote this plan and has since left public life for Anthropic, so nobody senior owns its economics. Whether the £120 billion shows up in the next fiscal document while ministers keep giving speeches about £400 billion. Whether Skills England updates. And whether anyone asks the same of the other headline figures in circulation: the EU, Japan and a dozen US states all run on consultancy-supplied projections, and almost none has been audited as cheaply as Britain's just was, by one backbencher with one written question.

The test is not the announcement. It is the adoption curve. If British firms take up AI at the pace the 2023 model assumed, £400 billion comes back into play and the correction looks like a blip. If they do not, £120 billion is the optimistic case — and the number to watch is the one nobody publishes: how much of the promised productivity shows up in the output statistics by 2028.

Should a government be allowed to cite a vendor-commissioned report as its own forecast? Write to us via our contact page.