Your Household Income Went Up, Your Share of the Economy Went Down
Sign Up For Daily Job Market News
Two Census Bureau numbers came out this week that don’t agree with each other, and both are true at the same time. Real median household income hit $87,460 in 2025, up 2.6% and the highest figure since the Census Bureau started tracking it in 1967. Median post-tax income climbed 3.1% to $76,060. The official poverty rate fell for the third straight year to 10.2%, about 34.5 million people, and child poverty hit a record low of 13.4%.
None of that reached the bottom of the income ladder. Households in the top 10% saw their income rise 1.7% last year. Households in the bottom 10% saw no measurable change at all. A household at the 90th percentile brought in about $261,300 in 2025. A household at the 10th percentile brought in about $20,010, roughly what full-time work at $10 an hour produces. The top fifth of households took in 52.4% of all household income in the country. The bottom fifth took in 3%.
The racial breakdown shows the same unevenness. Asian American households had the highest median income at $126,300. White households came in at $91,930, up 3% for the year. Black household income rose 4.8% to $59,980, the fastest growth of any group Census tracks, though it still trails every other group in dollar terms. Hispanic households posted a median of $73,260, statistically flat for the year, while Hispanic poverty fell to 13.9%, a record low since Census started tracking it in 1973. This progress on poverty comes alongside a 21.8% uninsured rate among Hispanic adults age 19 to 64, still the highest rate among the major groups Census reports.
A broader measure of hardship tells a less optimistic story than the headline numbers. The Supplemental Poverty Measure, which factors in taxes, benefits, housing, work costs, and medical expenses, held at 13.1%, nearly three points above the official rate. Medical costs alone pushed an estimated 7.7 million more people below that line. The uninsured rate stayed flat at 7.9%, about 26.7 million people, and Medicaid coverage slipped to 17.1% of the population, roughly 1.4 million fewer people than the year before.
Robert Greenstein of the Brookings Institution told Axios that 2025 mostly predates the safety-net cuts in the federal budget package passed under the current administration, and that this release may be the last clean snapshot before those cuts start showing up in the data.
Put this next to a second number that came out the day before, and the picture sharpens considerably. I cover that one below, but the short version: workers’ overall share of what the economy produces just hit a record low, even as household income hit a record high. Both are true because the gains are landing near the top of the distribution, not the middle or bottom.
The $20,010 figure at the 10th percentile is the one I’d sit with the longest. A full-time, minimum-ish wage job produces that in a year the economy is calling historic, and it moved by nothing. If your entry-level and hourly pay scales haven’t kept pace with your management and specialist tiers, this data says you’re not imagining the gap. The medical cost figure matters just as much for hiring: 7.7 million people fell into poverty because of health expenses alone, which means for a lot of candidates, the benefits package is doing more work in their decision than the base salary. As Medicaid coverage recedes, that’s only going to matter more, especially for entry-level and hourly roles where employer coverage was never guaranteed to begin with.
Staffing your team doesn’t have to be hard.
Reach out and see how we can help.
Workers’ Share of the Economy Just Hit an All-Time Low
A day before the Census release, BLS published a smaller number that explains a lot about how a record income year and a stalled bottom 10% can happen at once. Labor share, the fraction of everything the economy produces that goes to workers as pay, fell to 52.8% in the second quarter of 2026, the lowest reading in the 79 years BLS has tracked it.
Labor share counts wages, salaries, benefits, pensions, and employer payments toward government social insurance. Whatever doesn’t go to workers goes to capital: profits, interest, rent, and returns to owners. For most of the post-World War II era, the number held in a narrow band between 60% and 66%, peaking at 66.2% in the fourth quarter of 1960. It started sliding after 2000, dropped sharply after the 2001 and 2007-2009 recessions, and mostly leveled off through the 2010s. The pandemic briefly reversed the trend, pushing labor share up to 58.9% in the second quarter of 2020 because output collapsed faster than payrolls did, before the decline resumed and eventually broke every prior low on record.
For 50 years, roughly six of every ten dollars the private sector produced went to the people doing the work. Now it’s a little more than five. This gap doesn’t show up as a pay cut on anyone’s stub. It shows up as wages that grow in dollar terms while growing slower than the output they’re attached to, which is a much quieter story to tell than a layoff and a much harder one for workers to point to.
I’d put this in the same category as the automation shift I described after Lynda Gratton’s four-box framework earlier this week: when software does work that used to take headcount, output holds or grows while the compensation line doesn’t move with it, and the ratio drifts toward capital by definition. The last five years of this data already show the pattern, and AI is the newest and cheapest version of the same force. For workers, the practical response is finding something that doesn’t depend on a paycheck alone: equity, a scarce skill, ownership of some kind. For employers, the warning is that a workforce watching record profits pass them by doesn’t stay quiet forever. It shows up in turnover, in unionization drives, and in how hard your best people push back the next time you ask for more with the same headcount.
What a Recorded Interview Means for What You Say in It
The push for wage transparency and pay equity isn’t the only place technology is reshaping the hiring conversation this month. Recording, transcription, and AI summaries have moved out of business meetings and into job interviews, and Challenger, Gray & Christmas flagged just how far that shift has already gone. 70% of U.S. employers now use AI somewhere in their hiring process, according to an April 2025 TestGorilla survey, and 75% of professionals use AI notetakers in at least some meetings, per a 2025 Fellow survey. Zoom, Teams, and Google Meet all build in recording and AI summary features now, and dedicated recruiting platforms go further, generating interview notes, filling scorecards, and comparing candidate answers to job requirements automatically.
John Challenger’s read is direct: “The important change is not that more meetings are being recorded. Conversations that once disappeared after a meeting can now become permanent, searchable records that are stored, shared, and incorporated into other systems.” The practical consequence for candidates is that a hiring manager may read an AI-generated summary and never watch the actual interview. In Challenger’s words, “that summary can become the initial evaluation of the candidate, even when the software is described as a notetaker. And sometimes those notetakers make mistakes.”
The bias data is the part of this story worth slowing down for. Research published in the ACM FAccT Proceedings and covered by Stanford HAI examined 4 million applications to more than 1,700 positions at over 150 companies, all routed through a single hiring vendor. It found that 26% of Black applicants and 15% of Asian applicants applied to positions where the AI system discriminated against their racial group, and the software recommended only 58.2% of applicants to hiring managers at all. Challenger put the pattern plainly: “Typically, people hire people they like, who they click with. This can cause, and in fact, has caused, bias in hiring practices, as hiring managers tend to hire people who are very similar to themselves. In theory, this is something an AI hiring tool can mitigate, but in practice, automated and AI-screening tools show bias, as well.” On privacy, 47% of active AI notetaker users told Fellow that a notetaker had recorded or shared something they didn’t intend to capture.
My read on this lines up with what I’ve watched happen on the recruiting side over the past couple of years: a recruiter’s notes used to be the record of a screen, and now the recording is the record, with the AI summary as the first and sometimes only thing a hiring manager reads. For candidates, that changes what a good answer sounds like. Pronouns like “it” and “that” work fine out loud and fall apart in a transcript, so name the project, the number, and the part you personally owned. In Challenger’s phrasing, “Candidates should be clear enough for the transcript, specific enough for the evaluation, and natural enough for the person conducting the interview.” For employers, disclosure is the floor, not a courtesy: “Candidates should not have to interrupt an interview to determine how their information will be used, stored, or shared.” If your recording and retention policy is a mystery even to your own recruiters, fix that before the next candidate finds out the hard way.
Challenger’s closing point is one I’d underline even though the firm has an interview-coaching product to sell: “Hiring is still fundamentally a human decision. The best interviews reveal whether a candidate can work together with their teams to solve problems, and no algorithm can make that judgment for you.” A transcript can tell you what someone said. It still takes a person to decide whether you’d want to work next to them.
