Recognition is becoming a habit of fewer people

Workplaces have changed. Has the everyday act of recognizing a colleague changed with them? We examined recognition records across several years, then asked what research on collaboration, attention and appreciation can explain.

In the same five organizations, among employees with comparable time on the platform, monthly giving participation fell from 60.5% in the first half of 2021 to 35.0% in the first half of 2026. People who continued to give sent nearly as many recognitions. Fewer people shared in the practice.

The decline is observable in these records. Its causes remain open. Literature helps identify explanations worth testing, but cannot establish what happened inside these organizations.

The decline persists within the same organizations

All five organizations declined. The pooled decline is 25.4 percentage points using unrounded estimates. Giving each company equal weight yields a 33.6-point decline, with a descriptive bootstrap interval of 21.1 to 46.1 points. Five surviving organizations cannot establish the size of a global trend.

The same 417 people enrolled in all six months at both endpoints also declined, from 65.3% to 39.7%. Excluding the one organization no longer currently live leaves a decline from 60.5% to 35.6%. These checks reduce some alternative explanations, while survivor selection remains.

Participation changed more than giving frequency

Among months when someone gave recognition, the average number of events changed relatively little. Recipient reach narrowed, however. The remaining recognition is not completely unchanged.

Monthly persistence among previous givers fell from 83.1% to 75.2%; activation among previous non-givers fell from 24.1% to 12.7%. These transition comparisons include all tenures in the five-company panel and require enrollment in consecutive months. Maintaining participation and restarting it both deserve attention.

The longer trend includes a recent recovery

The broad population shows lower participation than earlier years, alongside recent improvement. Because the companies and employees contributing to these lines change, they describe observed activity rather than a controlled time effect.

A broader company-matched comparison of 2025 and 2026 finds declines in 12 of 18 organizations with sufficient tenure-matched observations. The equal-company average change is −6.3 percentage points (descriptive 95% interval: −10.5 to −2.4). Decline is neither universal nor uninterrupted.

What changes in work could explain this?

The following mechanisms are hypotheses about these customers. External studies measure different people and outcomes; their effect sizes cannot be assigned to this recognition decline.

Changed collaboration could reduce opportunities to notice contributions

There is credible evidence that working arrangements can change who interacts with whom. Yang and colleagues studied communication among 61,182 US Microsoft employees during early 2020. Their natural-experiment analysis found that firmwide remote work made collaboration networks more siloed and static. It studied communication networks, not recognition, and forced full-time remote work at one technology company does not represent every workplace. Yang et al., Nature Human Behaviour, 2022.

Related evidence from software engineers links proximity to more coding feedback. That supports investigating visibility and interaction, but coding feedback is not appreciation and its estimated effect should not be transferred to recognition. Emanuel, Harrington and Pallais, The Power of Proximity to Coworkers.

Our inference: if employees encounter fewer contributions outside their immediate circle, recognition could reach fewer people and become concentrated. Our recipient-reach finding fits this hypothesis, but does not distinguish it from smaller teams, changed incentives, or platform use.

Test: obtain historical team membership and actual working arrangements. Measure recognition across teams and to previously unrecognized colleagues before and after a documented change. Compare with teams whose arrangements did not change. A similar decline in unchanged teams, or preserved cross-team reach, would weaken this explanation. Current reporting lines cannot safely stand in for historical teams.

Attention and manager reinforcement deserve separate tests

Attention pressure is plausible context, not a demonstrated historical cause. In Microsoft's 2023 survey of 31,000 workers across 31 markets, 68% reported insufficient uninterrupted focus time. This is a vendor-sponsored snapshot; it neither measures recognition nor establishes that focus deteriorated in our customers. Microsoft Work Trend Index, 2023.

Our inference: when recognition requires a separate voluntary action, competing demands may reduce the number of people starting it while leaving established givers relatively active. Our participation-versus-frequency pattern fits, but could also result from a less visible entry point or fewer reminders.

Test: distinguish people who never see the opportunity from those who open the flow and abandon it. Recover notification delivery, entry-point exposure and completion history where available. Test a relevant, low-burden opportunity to recognize a contribution, measuring new givers and next-month persistence rather than clicks alone.

Manager engagement has also declined recently in an external survey. Gallup reports 31% of managers engaged in 2022 versus 22% in 2025. That warrants investigating manager reinforcement, but engagement is not a direct measure of available managerial time or recognition behavior. Gallup's engagement construct itself includes recognition, so it is not an independent causal explanation. Gallup, State of the Global Workplace 2026.

Test: establish whether a manager's reduction in giving precedes a team's decline, controlling for common company-month changes. Recover manager transitions and team size over time. Even temporal ordering could reflect shared workload or HR campaigns; a randomized manager practice would provide stronger evidence. The inherited-versus-later-joiner study cannot resolve this because actual hiring ownership and historical manager assignments are missing.

Expressing appreciation can require more effort than people expect

A longstanding expression barrier could become more consequential when recognition needs deliberate initiation. In three gratitude-letter experiments, people underestimated recipients' positive reactions and overestimated awkwardness. These experiments establish a possible barrier to expression, not a decline in gratitude over time or a workplace trend. Kumar and Epley, Psychological Science, 2018.

Our inference: if informal cues or established routines disappeared, employees may need more help noticing an occasion and deciding what to say. Lower persistence and reactivation are consistent with this, but cannot establish that routines changed.

Our recognition descriptions became longer, with a median of 26 versus 74 characters in the same-company event sample. That does not show that expression became harder, more authentic, or AI-generated. Language, interface requirements and the selection of remaining givers could all affect length.

Test: ask quiet employees about the last contribution they appreciated but did not recognize in Happily. Separate lack of occasion, discomfort, forgetting, effort, and a decision to use another channel. Compare a simple optional writing aid with a reminder alone and a holdout. Measure whether recognition reaches more people and continues without repeated prompting.

The evidence also limits the story we can tell

Hybrid work is not equivalent to social or organizational deterioration. A randomized trial of 1,612 Trip.com employees in China found that two home-working days per week improved satisfaction and reduced departures without damaging measured performance. It did not measure recognition, but it rules out treating all hybrid arrangements as inherently harmful or prescribing office attendance from our findings. Bloom, Han and Liang, Nature, 2024.

Remote work has not kept rising continuously. A multicountry study found work-from-home rates among college-educated employees broadly stabilized between 2023 and early 2025. Geography and occupation matter. A continuing recognition decline would need a more specific mechanism than simply increasing remote work. Aksoy et al., PNAS, 2025.

External recognition measures do not all show decline. Gallup and Workhuman reported that 22% of employees felt they received the right amount of recognition in 2024, unchanged from 2022. Perceived adequacy is different from our recorded giving rate, and the research has a recognition-industry partner. Still, it does not support claiming that recognition universally fell. Gallup–Workhuman, 2024.

These differences matter for timing. The 2020 disruption precedes our high 2021 baseline; it cannot by itself explain the subsequent slope. Local declines at several points, including early 2022 and early 2025, need customer-specific dates. Our records do not support a clean ten-year giving trend: sender identity is missing in 2018 and largely missing in 2019. Nor does longer text identify an AI effect. The literature provides historical context; it cannot fill those measurement gaps.

What this means

Prioritize evidence that separates competing explanations. Recognition could have moved to another channel, become harder to initiate, or lost a routine that once supported it. These possibilities call for different actions.

Start with 12–18 exploratory interviews across at least six organizations, including stable or improving customers. Ask quiet employees about the last contribution they appreciated and what they did next. Combine these accounts with dated records, then compare behavior before and after changes against credible comparison groups. A successful intervention today would not, by itself, prove why the historical decline occurred.

Limitations

• The long-run panel contains only five surviving organizations. Enrollment is assigned-question presence, not historical HR headcount or confirmed active use.
• Recognition outside Happily is unobserved. Missing sender identities and event-definition changes require care; wholly unlogged events cannot be bounded.
• Historical working arrangements, manager assignments, product exposure and incentives are unavailable. Observational associations do not establish cause.
• The literature review is focused, not exhaustive. Several studies concern technology workers; industry surveys have commercial interests. Geography and occupation may differ from these customers.
• Longer descriptions do not establish higher quality or AI authorship. No claim is made that people became less generous.