What Office Occupancy Data Shows About Return-to-Office in 2026

One widely followed weekly measure of US office attendance comes from Kastle Systems, a security-access company whose "Back to Work Barometer" tracks how often badge and fob holders actually walk into office buildings across ten major metro areas. For the week ending Wednesday, September 9, 2026, average occupancy across all ten metros and every building class was 52.6%, down 0.6 percentage points from 53.2% the previous week. That's a real, specific figure — but it describes one vendor's own tracked building portfolio in ten cities, not every office in the United States, and "occupancy" here means badge-entry activity measured against a February 2020 baseline, not a literal count of how full any building is at any given moment.

Kastle publishes its Barometer on a rolling weekly basis, typically on Tuesdays, using anonymized access data from more than 300,000 users in its tracked buildings. The current Barometer page was checked directly on September 23, 2026; the reporting week itself runs Thursday through the following Wednesday, and the figures below are a dated snapshot, not a live reading or a measure of the full year. The headline stats and city breakdown that follow all come from that same page.

What the barometer actually measures

Kastle's methodology, described on its own data-science page and FAQ, counts only the first building entry each cardholder makes on a given day, specifically to avoid overcounting from people badging in and out several times. That daily first-swipe count is then compared against a fixed reference point: average weekly first-swipe activity during a three-week window in February 2020, before the pandemic changed office attendance. A reading of 52.6% means entry activity across the ten metros ran a little above half of what it was in that February 2020 window — it doesn't mean roughly half of all desks were occupied at once, since Kastle isn't measuring desk occupancy directly.

Kastle describes its own dataset as a complete transaction record rather than a sample: "The Barometer reflects the universe of all transactions rather than a sample and therefore has no margin of error." That's a real claim about internal consistency, but a transaction record with no margin of error is still only a record of Kastle's own buildings. The ten tracked metros are New York City, Houston, Chicago, Los Angeles, Austin, Washington, D.C., San Jose, Dallas, San Francisco and Philadelphia. The data says nothing directly about other metros, about buildings outside Kastle's client base, or about workers who never badge into a Kastle-equipped building at all — including people working from home or from a workspace using a different access system.

Occupancy by metro, week ending September 9

Occupancy varied widely across the ten tracked metros this week. Kastle reported week-over-week changes for all ten metros. The chart shows the current-week readings; the paragraph below reports the corresponding changes. Because this reporting week included Labor Day, its week-over-week movement should not be treated as evidence of a lasting attendance trend.

Among the ten cities, New York City posted the largest reported increase, up 3.9 percentage points to 57.8%. Houston rose 1.3 points to 59.2%. The following four also declined: Austin down 1.7 points to 69.2%, Washington D.C. down 1.4 points to 48.8%, Los Angeles down 1.9 points to 44.8%, and Chicago down 2.5 points to 54.0%. Kastle also reported declines for San Jose (down 1.2 points to 44.8%), Dallas (down 1.0 point to 63.9%), San Francisco (down 0.7 point to 44.7%) and Philadelphia (down 0.7 point to 38.9%). These are single week-over-week comparisons, not evidence of a sustained trend.

How to use the numbers in a workspace decision

For a team planning office capacity, the 52.6% reading is not a seat-planning ratio. It compares one vendor's current entry activity with its own February 2020 activity, rather than showing how many desks a particular team needs. A team with 40 employees should not multiply 40 by 52.6% and assume that 22 desks will be enough. Instead, examine actual team attendance by day, meeting-room use, simultaneous peak demand and the consequences of having too few seats.

The difference between the 52.6% weekly average and the 64.0% peak-day reading is 11.4 percentage points. That arithmetic illustrates why an average and a busy day answer different questions, but neither figure predicts a particular employer's peak occupancy. Employers should use their own attendance records and policies when sizing a workspace.

City comparisons also need care: Austin's 69.2% and Philadelphia's 38.9% are readings for Kastle-tracked properties in those markets, not marketwide vacancy rates, coworking utilization rates or measures of the share of workers who follow an employer's return-to-office policy.

What this snapshot doesn't show

None of the above should be read as a verdict on return-to-office across the tracked metros, on hybrid-work adoption broadly, or on demand for coworking or flexible space specifically. The Barometer is a proxy built from one company's own client buildings, and this piece uses one week's absolute figures plus one week-over-week comparison for the 10-metro average, the Class A+ average, the peak day and all ten metros.

None of that makes the Barometer useless. It provides regularly published metro-level office-entry data with a disclosed methodology, which makes it useful for a carefully bounded comparison. It's simply a proxy from one vendor's badge system in ten cities, not evidence about all US offices as a whole. Readers weighing their own workspace strategy, including whether a flexible arrangement makes sense alongside or instead of a traditional lease, can browse coworking spaces, private offices and meeting rooms across the US on BestCoworking and compare current options directly in their own metro rather than relying on any single national data point.