The gap between young and old in Britain's job market is the widest on record, and the survey that found it has been losing young people

On 29 July this paper published a report built on one calculation: a country's youth unemployment rate divided by its rate for all ages. We called it the youth penalty and said plainly that the sum was ours, not the International Labour Organization's.

The inputs are two indicators from the World Bank's World Development Indicators, both drawn from the ILO Modelled Estimates database, which the World Bank last accessed on 17 January 2026. The series runs annually from 1991 to 2025. That is still the current release: the World Bank last updated these indicators on 27 January 2026 and the access date on both live pages has not moved.

That report ranked Britain against the world and found its 2025 figure of 3.09 sat 33rd of 182 countries against a median of 2.34. This is the same dataset with a sharper question: what has Britain's own line been doing?

The line

Between 2005 and 2013 the ratio stayed between 2.54 and 2.71. It did not break that band during the financial crisis. In 2019 it was 2.92. In 2022, 2.81. In 2024 it reached 3.28, the highest reading in the whole thirty-five year series.

The two years in between are what matter. Youth unemployment rose 3.7 percentage points, from 10.6 to 14.3. The rate for everybody rose 0.6, from 3.8 to 4.4. The young end moved about six times as far.

One thing about 2025 first, because it reads the wrong way round. The ratio fell, to 3.09. Youth unemployment did not fall. It rose again, to 14.6. The gap closed because unemployment for everybody else rose faster.

Why a move here is unusual

Our July report found something that makes this harder to wave away. The ratio does not respond to recessions. In 2009 the average adult unemployment rate across rich countries rose 1.85 points, the largest single-year move in the data, and the gap between young and old moved about a hundredth of a point. In 2020 adult unemployment rose 1.33 points and the gap moved a tenth, less than it moved in 2019 or in 2023, neither a recession year. Downturns hit both age groups and the gap holds still.

Britain's moved 0.47 points in two years with no recession, four times what the middle of the pack managed in the 2020 crash.

The two explanations on offer

Artificial intelligence first. Ranked by change in the ratio between 2022 and 2025, across twenty-four advanced economies: Italy plus 0.277, the United Kingdom plus 0.276, Switzerland plus 0.197, Korea plus 0.172, Iceland plus 0.170, Finland plus 0.166, Ireland plus 0.163, Portugal plus 0.155. The United States sits at plus 0.004, fifteenth of twenty-four. Germany, Japan, Sweden, Norway, Denmark and Luxembourg all fell.

The country with the highest adoption and the frontier labs barely moved, and the top of that table is not a ranking of AI adoption.

The employer National Insurance change second. The rise in Britain runs 2022 to 2024. The change took effect in April 2025. A cost arriving in 2025 does not explain a rise that had already finished.

One survey, and what it was going through

Which leaves the question of whether this number is telling us about young people or about the survey.

Britain has one source for unemployment. The ONS says so: the tax records and the survey do not agree with each other, and even so, the Labour Force Survey remains the only source for unemployment, economic inactivity and self-employment. Payrolled employment can be counted from tax submissions. Unemployment cannot. There is nothing to check it against.

And in the years the gap widens, that survey was in trouble. Responses fell to their lowest point in July to September 2023, at a 12.7 per cent response rate. Detailed figures were not published between October 2023 and January 2024 on quality grounds. When publication resumed in February 2024 the ONS relabelled its own output as statistics still in development, which is how a statistics office tells you not to lean on them. It stayed that way until 11 August this year, when the Office for Statistics Regulation agreed the figures could go back to being ordinary official statistics.

The specific problem is worse than the general one. The ONS titles Figure 5 of its own quality report with the finding that the responding sample now holds more people aged over 55 and fewer aged under 24 than before the pandemic. Its attrition table shows 20 to 29 year olds making up 9.5 per cent of first-wave responders and 7.2 per cent of those who answer at every wave, against 11.3 per cent of non-responders. The total response rate for Great Britain in the first quarter of this year was 25.1 per cent.

And for the youngest part of the band, the answers are not theirs. Among 16 and 17 year olds, 95.3 per cent of responses come by proxy, meaning another adult in the household answers for them. For 18 and 19 year olds it is 87.5 per cent. For everyone aged 20 and over it is 33.5 per cent. Whether a British teenager has a job is, nine times in ten, something a parent told a survey.

What does check out

One source is not a survey at all. PAYE Real Time Information counts people on payrolls from tax submissions, so there is no response rate to collapse.

Between March 2024 and March 2025 the number of payrolled employees aged under 25 fell by 115,000, while the number aged 35 to 49 rose by 87,000. The July bulletin has the same shape: down 102,000, up 53,000. November: down 60,000, up 58,000. December: down 65,000, up 51,000.

That is a real divergence in a source nobody has to answer, and it is the reason this cannot be written off as a broken survey. It does not confirm the 3.28, for two reasons. It runs 2024 to 2025 while the widening runs 2022 to 2024, so it matches the direction and not the timing. And it counts people on payrolls, which is not a measure of unemployment, so it cannot confirm a peak in a figure it does not produce.

Two things would settle this, and neither is expensive

The first is the employment record for 21 and 22 year olds. If the widening is concentrated in that group it points one way. If it is spread evenly across everyone aged 15 to 24 it points somewhere else entirely. Those figures exist and nobody has looked.

The second is a paragraph from the ONS. It reweighted its estimates from 2019 onwards using 2022 population figures, and whether that is enough to correct the 15 to 24 group has never been published. The people who did the reweighting know the answer.

There is a candidate explanation that fits the dates and this paper is not printing it, because a date that lines up is not a cause and the work has not been done.

What can be said now is firmer than either of the two confident stories in circulation. Something real is happening to young workers in Britain. The tax records prove that much on their own. What nobody can yet separate is how much of the 2024 peak is those workers and how much is the survey watching them. A country that cannot answer that about its own young people is not short of data. It is short of a second source, and that is a choice rather than an accident.