Summary

• The £400bn estimate cited by UK officials came from a 2023 Public First study commissioned by Google.
• The government's more recent assessment estimates £55bn–£140bn in additional UK gross value added in 2030, with a central adoption scenario of £120bn.
• Differences in adoption assumptions and modelling approaches have prompted scrutiny of how commercially commissioned research informs public policy.

A £400 billion estimate repeatedly used by British officials to describe artificial intelligence’s potential economic contribution originated in a study commissioned by Google, while the government’s more recent central projection puts the potential benefit at £120 billion in 2030.

The distinction was confirmed by AI minister Kanishka Narayan in a parliamentary answer on 26 October and examined in reporting published on 4 November. The larger estimate came from research produced in 2023 by consultancy Public First, now known as PF Global, which assessed the potential contribution of generative AI to the UK economy.

The UK government has since adopted a separate assessment suggesting that AI adoption and data investment could add between £55 billion and £140 billion to annual gross value added in 2030. Its central scenario is £120 billion, although ministers emphasise that these figures remain highly uncertain.

The difference does not establish that £280 billion in economic activity has disappeared. Both figures are projections, and their results depend on assumptions about how quickly organisations adopt AI, which activities become more productive, and how efficiency improvements translate into additional output.

A commercial forecast became an official reference

Public First prepared the original research for Google in 2023 as part of an assessment of the company’s economic impact and the wider opportunities associated with generative AI.

The consultancy estimated that the technology could generate more than £400 billion in value for Britain by 2030. The figure subsequently appeared in government communications, including material promoting AI adoption and the economic possibilities associated with the technology.

Commissioned economic research is not inherently unreliable, and a commercial sponsor’s involvement does not invalidate the resulting analysis. However, sponsorship, modelling assumptions, and the intended scope of a forecast are relevant when its conclusions become part of the public case for policy decisions.

According to an investigation by WIRED, officials continued citing the larger figure after the government had access to a lower assessment. The publication established that the alternative estimate had been available to ministers since at least November 2025.

Public First has maintained that its research was independent and that the consultancy retained responsibility for its economic judgements. Its head of economics, Jonathan Dupont, told WIRED that the earlier analysis assumed faster AI adoption than subsequently appeared likely.

The consultancy now considers the economic gains likely to be achieved by 2030 to be closer to the government’s projection, while maintaining that AI could deliver substantially greater benefits over a longer period.

Narayan’s parliamentary answer identified the original funding arrangement and placed the two projections alongside one another. It did not formally withdraw the £400 billion figure or declare the earlier research invalid.

Productivity estimates depend on adoption

The government’s newer assessment draws on Organisation for Economic Co-operation and Development research suggesting that AI adoption could increase UK productivity growth by between 0.4 and 1.3 percentage points.

Those estimates have been translated into a potential increase of £55 billion to £140 billion in gross value added in 2030. The central adoption scenario produces a figure of £120 billion.

Gross value added measures the economic output generated after deducting the cost of intermediate goods and services. It differs from the revenues earned by AI suppliers, the value of investment in computing infrastructure, or the amount spent purchasing AI products.

Because the estimate concerns additional annual output in 2030, it should not be interpreted as a cumulative total of benefits already received between now and that year.

Forecasts also depend heavily on how businesses introduce the technology. An employee may complete a particular task more quickly with AI assistance, but the time saved does not automatically produce equivalent growth in the organisation’s output.

Managers may need to redesign processes, update software, improve the quality of operational data, and train employees before improvements can be sustained. Generated material may also require checking, especially where errors carry financial, legal, or safety consequences.

Those costs and constraints differ between industries. A professional services firm using AI for document preparation faces different implementation requirements from a manufacturer introducing automated quality control or a public authority redesigning administrative decisions.

Adoption rates therefore influence the estimates in two ways. They determine how many organisations use AI, but the depth of adoption also affects whether the technology changes isolated tasks or the wider processes through which economic value is produced.

The government’s June 2026 response to industry AI adoption plans acknowledged that many businesses were still using the technology for relatively limited activities, including drafting documents, conducting research, and assisting with software development.

More extensive gains may require changes in how organisations operate rather than simply purchasing additional AI licences.

Forecasts influence spending decisions

Large estimates of potential economic benefit can influence public debate about investment in computing infrastructure, skills, data systems, and technology adoption programmes.

National economic projections provide useful context, but they cannot establish whether an individual government technology project offers value for money. That requires a more specific assessment of implementation costs, expected operational changes, risks, and measurable outcomes.

An AI application may be economically attractive when it reduces processing time without weakening service quality. Another could increase expenditure because staff must repeatedly correct its output or maintain systems that duplicate existing functions.

The overall economic forecast cannot distinguish between those outcomes. Organisations need evidence from deployments to establish whether promised productivity improvements are being realised.

Researchers have also questioned the way industry-funded projections are incorporated into public policy. The Ada Lovelace Institute has urged greater scrutiny of underlying assumptions, independent economic research, and systematic measurement of actual benefits.

Such scrutiny does not require rejecting commercially commissioned research. It requires clear disclosure of where estimates originate, how their assumptions were developed, and which limitations apply when the findings are used outside their original context.

Official reporting already distinguishes between the economic contribution of the AI industry itself and the wider potential effects of AI adoption across the economy. Mixing those measures can create further confusion about the scale and timing of expected benefits.

The £400 billion and £120 billion projections both describe possible future outcomes. Neither provides a direct measurement of national productivity gains already delivered by AI.

Evidence will develop as deployment increases

Economic models can become more reliable as researchers gather information about actual adoption, changes in working practices, expenditure, and output across different sectors.

The UK has established an AI Economics Institute involving the Treasury and technology policy departments to examine how AI affects productivity, employment, and economic growth. Its work could contribute additional evidence to future assessments.

However, uncertainty will remain because technological capability, costs, regulation, and business behaviour may all change during the period being forecast.

The parliamentary answer establishes the source of a widely repeated estimate and sets out the government’s more recent range. Future public assessments will need to show whether the assumed rates of adoption are being achieved and whether improvements in individual activities translate into sustained economic growth.

Until comparable evidence is available, the figures are best treated as conditional economic scenarios, with their underlying assumptions disclosed whenever they are used to justify spending or policy choices.