In 2025 filings, use of the acronym DEI fell 68 percent among the largest listed American companies. The word racial fell 58 percent. The share of companies tying executive pay to diversity metrics was almost halved, from 68 percent to 35 percent. By any reading of the public record, this was a retreat.
In the same filings, the share of companies disclosing formal board or board-committee oversight of diversity rose from 72 percent to 79 percent. Companies said far less about it while formally supervising it more. Whether you read that as quiet persistence or as reputational management, it is the single most awkward fact in this debate and it does not fit either side's story cleanly.
The same pattern shows up in universities. The share of full-time faculty job advertisements requiring a diversity statement fell from about a quarter in 2024 to about 11 percent in 2025, and standalone statements fell further still. The practice at the centre of the campus fight had largely collapsed on its own before most of the laws aimed at it took effect.
This is where both sides cite real studies and reach opposite conclusions, because they are citing different things.
The largest review pooled 260 independent samples covering more than 29,000 participants and found an overall positive effect. Attitudes and reactions faded over time, but what people learned held steady. The effect was strongest when training ran over a sustained period, covered both awareness and skills, and sat alongside other measures rather than standing alone.
The most cited finding on the other side comes from a study of 805 private workplaces tracked over three decades. It found that mandatory manager diversity training was associated with falling numbers of Black and Asian women in management, while voluntary training tracked with gains. The difference was not the content. It was the compulsion.
Both findings can be true at once, and read together they say something fairly specific: training can teach people things, and requiring attendance can produce resentment that outweighs the teaching. That is a conclusion about how these programmes are run rather than whether they should exist, which is not a satisfying answer for anyone.
A great deal of corporate spending was justified by consulting research claiming diverse leadership teams outperform financially. Independent researchers went back and tested those findings against large listed companies and found no statistically significant relationship between executive racial diversity and earnings, sales growth, profitability or shareholder return.
There was also a sequencing problem in the original work: diversity was measured after the profits, which reverses the causal arrow. A profitable company can afford to hire more broadly, which is a different claim from diversity causing profit.
The reverse claim has not held up either. Researchers comparing companies that kept their commitments against those that abandoned them found no significant difference in returns or revenue. The honest summary is that there is no strong financial case in either direction, which means this is a question about values and law rather than about shareholder returns.
Two changes matter. Federal guidance now states that anti-discrimination law has no minimum threshold of harm, and that a general interest in diversity is not a defence against intentional discrimination. Then in June 2025 the Supreme Court ruled unanimously that majority-group plaintiffs face no extra evidentiary hurdle, removing a barrier that had existed in several appeals circuits.
Together these make a programme that restricts a mentoring scheme, a fellowship or an employee resource group by race or sex plainly exposed to challenge. Employers reading legal advice and scaling back are not imagining the risk.
The counter-argument is that the restrictions cut both ways. A federal appeals court struck down Florida's ban on mandatory workplace diversity training as viewpoint-based speech regulation, and that ruling did not depend on whether the training was any good. A government that can forbid an employer from running a session is exercising the same kind of power as one that can compel it.
Strip out the acronym and the fight is about what these programmes are for.
One view treats them as infrastructure. Much of the work is legally required anyway: harassment training states mandate, accommodation federal law requires, record keeping contractors owe. The same offices also run veterans hiring, caregiver policy, language access and disability accommodation, none of which most voters were asked about when the label was banned.
The other view treats them as an ideological programme that grew a bureaucracy and then a loyalty test. Large public universities averaged more than 45 staff with formal diversity responsibilities, roughly four for every one serving students with disabilities. Asking a job applicant to profess a political commitment before anyone reads their scholarship is not a neutral hiring practice.
Public opinion sits awkwardly between them. Support for promoting workplace diversity fell from 75 percent in 2019 to 64 percent in 2026, driven almost entirely by a 21 point drop among Republicans while Democratic support did not move at all. Forty-two percent say diversity efforts have made society more fair. Thirty-two percent say less fair. On this one, the country really is divided rather than persuaded.
Want the core arguments from both sides, side by side?
See both sides of the DEI Programs debate →