In the 2009-2010 school year, the Bavarian State Office for Health ran an experiment in six German schools. Twardella and colleagues tracked carbon dioxide levels across 20 classrooms and matched them against pupils’ test performance on the same days. In classrooms with good air, CO2 sat around 1,045 ppm. In classrooms with poor air, it climbed to 2,115 ppm. The children in the worse air were not slower overall. They made significantly more errors on the tests (Twardella et al., 2012, Indoor Air). Neither the pupils nor their teachers noticed a difference in the room.

That is the pattern I keep finding in this literature: room conditions change measurable performance, and the people in the room rarely notice.

The Charité hospital in Berlin shows the same pattern where the stakes are higher. In 2013 the architecture practice GRAFT rebuilt two intensive care rooms. They added a lighting system that simulated day and night, damped the ambient noise and moved medical equipment behind wall panels so patients no longer woke to the sight of it. Spies and colleagues later measured the result: delirium in the redesigned rooms ran at 46 percent, against 76 percent in the standard rooms nearby (Spies et al., 2024, Critical Care Medicine). ICU delirium extends hospital stays and is expensive to treat. A room redesign cut it by 30 percentage points.

The same question moves into offices too. Joseph Allen and colleagues ran the COGfx study on how ventilation shapes cognition at work. In well-ventilated buildings, cognitive test scores came out roughly double those measured in conventionally ventilated ones. Crisis response scored 131 percent higher, strategic thinking 288 percent higher (Allen et al., 2016, Environmental Health Perspectives). The air in a building shaped how people think more than most of what architects and clients debate in a design review.

The pattern recurs well beyond schools, hospitals and offices, across decades and hundreds of studies in air, light, noise and materials. In 1984, Roger Ulrich published a study in Science showing that surgical patients with a view of trees needed less pain medication and were discharged earlier than patients who looked at a wall. Successive teams have reproduced the effect in the decades since. The World Health Organization estimates that indoor dampness accounts for about 15 percent of new childhood asthma in Europe (WHO Housing and Health Guidelines, 2018).

Drawing on my experience with AI-based analysis in marketing and retail, I built a system that evaluates this research by building type and user group. Moving into an unfamiliar field was a fairly big stretch. AI tools and the coding tools built around them made the path less rocky than expected. The data itself was extensive, and because most of it was scientific, it was very consistent and easy to work with.

Working through it changed what I thought the actual problem was. I expected to find that the research was thin, or that nobody had worked out how to put it into terms a planner could use. Neither is true. What I found instead is that the research sits in separate disciplines that rarely meet: acoustics, lighting research, indoor climate and materials science each run their own studies, journals and conferences. Almost nobody in daily practice works through all four in parallel, so the picture stays split even when each piece of it is solid.

Architects and engineers set a room’s air, light, noise and outlook in the floor plan and the building’s mechanical systems, mostly at no extra cost. Whether an ICU patient’s window frames trees or a firewall is fixed early, at the same price either way. A classroom’s ventilation running at 1,000 ppm instead of 2,000 ppm is a question of how the equipment is set, and costs nothing to change. Architects and engineers make these decisions anyway, on some basis. That basis is rarely the evidence that already exists.