The £400 billion that government departments and agencies attached to AI for almost two years traces back to a 2023 study that Google paid for and the consultancy Public First wrote, AI minister Kanishka Narayan confirmed in a written answer on 26 October. The same answer puts the government’s own central estimate at £120 billion, within a range running from £55 billion up to £140 billion of extra gross value added (GVA) in 2030, and calls those figures “hugely uncertain”. WIRED, which reported the gap on 4 November, notes that the official number is under a third of the one ministers kept repeating. The more useful question for anyone who writes AI figures into a strategy paper or board pack is how either number was built.
Strategic Insight: Both figures are modelled projections, and both publish their assumptions. The firm behind the larger one now says part of the gap comes from adoption being slower than its 2023 model allowed for, so the assumption worth checking first is how quickly organisations take AI up.
How did a commissioned estimate become official?
The route is documented. Public First’s 2023 report for Google put the potential from generative AI tools at “over £400 billion in value for the UK economy by 2030”. In January 2025 the AI Opportunities Action Plan, which the Science Secretary had commissioned Matt Clifford to draw up, said adoption of AI might expand the economy by “an additional £400 billion by 2030”, with footnote 19 pointing to Public First’s report. In October 2025 a Skills England press release headlined a “£400bn AI skills gap” and credited the figure to the Action Plan, not to the study behind it.
Each step was a citation of the step before. By the time the number reached a skills announcement, its origin sat two documents away.
The real story is the wording, not just the source
The figure changed meaning as it travelled. Public First described potential value from generative AI tools. Google’s own blog post on the report added that this was possible “under the right conditions”. The Action Plan turned it into additional growth from AI adoption. Skills England made it “up to £400 billion” of growth potential attached to a skills gap. None of those restatements is a new calculation; each is a fresh label on the same 2023 estimate.
| Figure | Who produced it | What it measures, as stated | Key stated assumption |
|---|---|---|---|
| Over £400bn by 2030 | Public First, commissioned by Google (2023) | Value created by generative AI tools, UK-wide | Full economic impact takes 10 to 20 years to arrive |
| £55bn to £140bn, central £120bn | Government analysis (written answer, 26 October 2026, and DSIT method note) | Added UK GVA in 2030 | Adoption rises from 5% in 2024; no AI breakthrough beyond software built on current models |
| Up to £47bn a year over a decade | UK government calculation | The Action Plan’s “previous” estimate, per the minister | Based on an IMF estimate from the April 2024 World Economic Outlook |
| Around 2% to over 10% of GDP by end of 2031 | BCC-commissioned modelling (November 2026) | GDP level against a baseline | Two adoption scenarios, researched by Anthropic |
Critical Context: Public First’s report states that Google commissioned it, that all estimates are Public First’s own, and that Google “has not supplied any additional data, nor does it endorse any estimates made in the report”. The disclosure was in the report. What WIRED could not find, before the minister’s answer, was any recorded instance of government saying that Google commissioned it.
What do the two models actually assume?
The methods behind the two headline numbers differ in kind.
The 2023 Public First method
According to the report’s methodology, Public First took the US O*Net database of about 17,000 work tasks across roughly 800 occupations, had a large language model rate each task for the likelihood that AI could cut the time it takes by at least half, weighted tasks by their importance to each occupation, and mapped the results to UK occupations through its own crosswalk. It then assumed, from the history of earlier general purpose technologies, that generative AI’s full effect would take between 10 and 20 years to be felt. The estimate therefore rests on US task data, a model’s judgement of each task, and an assumption about diffusion.
The government method
The DSIT method note, published on 28 January 2026 and linked from a later written answer, begins with an OECD working paper’s estimate that AI could add between 0.4 and 1.3 percentage points a year to the growth of UK labour productivity for ten years. The paper counts adoption as AI used in core business functions, and it has no comparable UK survey figure: the UK’s starting rate of 5.4% is predicted from internet use, data centres and STEM graduates, which the authors say leaves UK estimates with higher uncertainty. It models three adoption curves rising from 5% in 2024 to 30%, 47% or 60% in 2034, scales the productivity gain by how far adoption has got in each year, and applies the result to a 2030 GVA baseline built from Office for Budget Responsibility forecasts. The note says the OECD range is built on “no significant uplift in AI capabilities”. The OECD’s own medium and fast scenarios, which supply the UK’s 0.97 and 1.27 points, do assume “expanded capabilities”, meaning software built on top of current models; what the paper leaves out is a breakthrough scenario.
The worked medium scenario in that note gives adoption of 19.2% in 2030 and a GVA uplift of £117 billion. The written answer calls the central estimate £120 billion. The note itself does not state £120 billion, and a follow-up from Dr Ben Spencer, the Conservative MP who tabled the original question, asking how the central figure was derived, was answered on 4 November only by a reference back to the earlier answer.
Reality Check: The lower number is not a corrected version of the higher one. It is a different model with different inputs, and the department describes it as “highly uncertain” and says the analysis will be revised as fresh evidence arrives.
PF Global, as Public First is now known, told WIRED the gap is partly about adoption. Jonathan Dupont, who leads economics there, said the government’s numbers “assume a much slower rate of adoption than looked possible back in our model that was created in 2023”, and that its own central estimate for 2030 is now “much closer to the Government’s number”. The method note bears out why adoption matters: in the government’s model the 2030 uplift is scaled by how far adoption has reached that year, 40.8% of its 2034 level in the medium case.
Who has been relying on the number?
WIRED reports that departments and agencies relied on the £400 billion for close to two years, across two governments, and that Peter Kyle, a former technology minister, also used it to promote AI. It also says ministers held the lower estimate from November 2025 at the latest, yet Skills England was still putting out £400 billion four months on. Lord Holmes of Richmond, a Conservative peer, told WIRED he asked about the figure’s origins in January 2025 and did not get a response, and that the figure was then used “by multiple other organizations”.
That last point is where the exposure sits for business readers. A number that appears in an official action plan is easy to treat as an official estimate. In the Action Plan, the only pointer to its origin is footnote 19, which names Public First’s “Google’s Impact in the UK 2023”.
| Group | What the change means for them | What is still unknown |
|---|---|---|
| Organisations that cited £400bn in plans or bids | The figure now has a disclosed commercial origin and a lower official counterpart | Whether funders or boards will expect restatement |
| Policy teams in government | The minister’s answer neither withdraws the £400bn nor says it was wrong | Whether departments will stop using it |
| Readers of new AI growth claims | Fresh scenario-based figures are still arriving, such as the BCC’s | How each will be reconciled with the official range |
⚠️ Warning: WIRED points out that Narayan neither withdrew the £400 billion nor said it was wrong. The AI Opportunities Action Plan page on gov.uk still carries it, so expect to keep meeting it in documents written before 26 October.
What this means for anyone citing an AI growth figure
The Ada Lovelace Institute’s Sumedha Deshmukh gave WIRED three questions the government should ask “before it uses any number around economic or productivity gains from AI”: “where did it come from, what does it assume, and who actually benefits from a given figure?” The documents behind the £400 billion show how each question can be answered from the public record, and they apply just as well to a board paper.
Where it came from means following the footnotes to the first calculation, not the most recent citation. In this case that took two hops, from Skills England to the Action Plan to Public First.
What it assumes is usually printed in the methodology. For the 2023 figure, that meant US task data, a language model’s classifications and a 10 to 20 year diffusion window. For the government figure, it meant OECD productivity estimates, a modelled UK adoption rate, adoption curves and no breakthrough beyond current models.
Who benefits is a question about incentives, not honesty. Deshmukh told WIRED that figures about AI “that look precise often have assumptions and incentives built into them, and industry often funds or shapes the research behind them”. Public First’s disclaimer and PF Global’s statement that its research is independent are part of the answer; the funder’s name belongs next to the number either way.
Strategic Reality: This week’s BCC modelling is a live test. Its range runs from around 2% to more than 10% of GDP by the end of 2031, and the BCC says Anthropic researched both scenarios. The three questions apply before it goes into anyone’s slides.
The open challenges
Two units that do not line up
The government states its estimate as added GVA in 2030. Public First speaks of value created across the economy by 2030, which its report likens to yearly growth of 2.6%. Without a common unit, the “less than a third” comparison tells you the gap is large, not exactly how large.
The central figure is softer than it looks
Both the minister and the method note call the official estimate uncertain. The £55-140 billion range is wide, and the published worked example does not reach the stated central figure. Anyone citing £120 billion is citing the central case of a model its own authors call highly uncertain.
Hype as a delivery risk
Dame Chi Onwurah, who chairs the Commons Science, Innovation and Technology Committee, said in comments to WIRED that her committee has found “hype” to be a major obstacle to modernising public services with digital tools, and that “overstating the benefits of technology risks undermining the case for change”.
What to watch
DSIT has said it will revisit its numbers as evidence builds. An update that publishes the adoption rates it has observed, not just the assumed curves, would show whether the 2030 range is tracking reality.
The Action Plan page on gov.uk still carries the £400 billion line, citing Public First. A correction or an added note would be the clearest sign that the government regards the figure as superseded. Spencer’s follow-up questions are a second signal: a fuller answer on how the £120 billion was derived would close the gap between the minister’s figure and the published worked example.
The BCC’s London adoption summit on 8 December is the next stage for its scenario numbers. Whether ministers or departments begin quoting the 10% figure, and whether they name who researched the scenarios when they do, will show whether this episode changed how commissioned numbers are handled.
The £400 billion travelled by citation: each document pointed to the one before, and by the time it reached a skills announcement the release named only the Action Plan, not Public First or its assumptions. That habit is cheap to fix, and it is the same in a ministry as in a boardroom.
Strategic Insight: Our news report covers the minister’s answer and the reaction. The figure in your own documents is the one worth checking next.
Source: UK Officials Said AI Would Add £400 Billion to the Economy. That Figure Mainly Came From Google (WIRED, by Ben Gartside, 2026). Additional sources: UK Parliament written answers 32092, 30469 and 38587; DSIT, AI Skills Boost: explainer (2026); AI Opportunities Action Plan (2025); Public First’s 2023 UK economic impact report for Google; British Chambers of Commerce (2026).
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching whether DSIT’s next update reports observed adoption alongside its assumptions. Read more analysis at Insights, or get in touch.