Chicago building energy

• Color: Per square foot (the default) is site EUI, energy used on site per square foot of floor area, so a building does not go dark just for being big. Total is the fuel use each property reported, added up. Both use seven classes computed once from the 2022 data; the legend relabels when you switch, and the view stays put.
• What the other fills mean: hatched properties were required to report and did not. Gray ones reported but the City published no EUI for them. Square markers are properties with no outline in the City’s 2015 footprint layer, or whose outline match is unconfirmed, placed at a City coordinate that passed a location check or was accepted on review. Exempt properties are hidden until you check Show exempt.
• Density: choose quarter-mile hexagons or community areas. A dashed, unfilled cell has benchmarked properties but none that reported energy; a blank one has no covered building. Neither means low energy use. Most hexagons with a value hold a single building, so check the count in the tooltip.
• Hover and pin: hover for a property’s name or address, type, floor area and value. Click to pin it: the panel adds both measures, the fuel mix where it adds up, and how confidently it was matched to its footprint. Esc unpins.
• Getting around: drag and scroll, or use + and −. Go to jumps to the Loop, the medical district, Hyde Park or O’Hare.
• One year: the map shows 2022 only, and gains a year control when a later release passes the same checks.
• By keyboard: once the map has focus, the arrow keys pan it and + and − zoom. In the buildings view, the list Properties in view, by address names the properties in the current view: the arrow keys move through it, Enter pins one and Esc unpins.
• Not built yet: a table view, and a keyboard path to the density cells. Every figure on the map is in the downloads below.

Two measures per property: site EUI, the energy used on site per square foot of gross floor area, and total site energy, the electricity, gas, district steam and chilled water, and other fuel it reported, added up. Both are what owners reported through ENERGY STAR Portfolio Manager, verified by a third party once every three years. Nothing here is metered by the City, and nothing covers a building under 50,000 square feet.

Of 3,613 properties covered for 2022, 2,645 reported, 732 did not, and 236 were exempt. The non-reporters are hatched rather than left off, because a compliance gap is a finding, not a blank; the exempt are outlined when you ask for them. 83 properties reported but have no EUI in the City’s release, so no total is formed for them. At least one building is listed twice, once as reporting and once as not.

The map was built on data year 2022, and 2023 has not yet been reviewed for it. The City’s 2023 release does fail one basic test: only about 9% of its rows put the building inside the community area the same row names, and Rush’s campus on the Near West Side lands in North Center, almost five miles north. But this map uses a release’s own coordinate only where it passes that test or was accepted on review, and uses no 2023 coordinate at all. Located the way it locates everything, by property id and address, 87% of 2023’s reporting properties are matched at high or medium confidence, about the same as 2022; 2023’s matches have not been checked for precision. And 2023 is the release whose floor areas are consistent (see the next item). It is the next year to add. Data years 2024 and 2025 had not been published at the snapshot date.

Total site energy is the fuel use each property reported, added up. It is not site EUI times gross floor area, although the two should be equal: for 989 of the 2,562 reporting properties with an EUI, the floor area the City published for 2022 is inflated, by about a fifth at the median, so that product overstates them and the city as a whole by about 12%. The City’s own greenhouse-gas figures agree with the fuel totals and disagree with the floor area by exactly the same factor, and its 2023 release lists the same buildings with smaller floor areas that fit. The per-square-foot figure is the City’s own and is not affected. The same problem runs through the 2020 and 2021 releases.

Reported addresses are matched to City building footprints where a match can be found, with the release’s own coordinate used only as a check or a fallback, and only if it falls inside its row’s stated community area or was accepted on review. 2,304 of the 2,645 properties that reported (87.1%) are matched at high or medium confidence, short of the 90% the project set before matching; the first version of this page compared an earlier figure with a threshold lowered to 85% the day that figure came in. A check of 406 sampled matches, each against evidence the matcher did not use, puts about 94% of the drawn outlines on the right building: about 99% of high-confidence matches and 74% of medium ones, the weakest being an outline chosen only because it is nearest the City’s coordinate. The check was made by AI reviewers working to a written rubric; no person has re-checked it, and no aerial imagery or site visit was used. Every verdict and its reason is in the download. A match that fails a basic size check is rejected or downgraded, and a property whose outline is unconfirmed is drawn as a marker at its City coordinate instead. The footprint layer is a 2015 snapshot, so a matched outline is whatever stood at the address in 2015, and 181 reporting properties built since have no outline at all. A property with several buildings reports one figure; it is drawn on the outlines its address or coordinate ties to it, often just one of its buildings, and never divided among them. Sixteen properties the tiers could not settle, UChicago Medicine and McCormick Place among them, were resolved one at a time against named evidence — fourteen of them placed, two deliberately left unplaced. Those reviews were made the same way as the precision check, by AI reviewers rather than by a person, and the evidence for each is in the download.

Each measure has seven quantile classes, computed once from the 2022 values and published in classes.json; the map reads them rather than computing its own. Two reporting properties exceed 1,000 kBtu per square foot: a data center, which is plausible, and a worship facility, which almost certainly is not. Both are drawn as reported.

A cell’s value is the reported energy of the buildings in it divided by the cell’s whole area, parks, rail yards and lakefront water included. 591 of the 964 hexagons with a value hold a single reporting building, and 14 community areas hold fewer than five, so a dark cell is often one large building rather than a dense neighborhood. Every reporting property with a total is placed, so none is left out of the sums.

Commercial, institutional, education and healthcare properties carry the name the City publishes. Residential properties are labeled by address and type, and by name only where the City’s name is a building name rather than an owner or LLC. A property with no stated type, as most non-reporters are, loses its name if it carries an owner, LLC or association token.

density_ca.csv and density_hex.csv print energy per square mile as the shortest decimal that reads back as the same number, which drops the floating-point residue some values carried. facts.json labels the match-precision figures and the count of properties resolved on review “(AI review)” and says that no person re-checked those verdicts. The data dictionary no longer says a sample of the AI verdicts was re-checked, since no record of such a re-check was kept, and it gives the greenhouse-gas agreement as measured, within 3% for 1,566 of the 1,569 records whose floor area is consistent, rather than “exactly”. It also says what the location test of 2023 found, rather than pointing to its report, which is not published. match_review.csv names the Goodman Theatre in one reviewer’s note; no verdict changed. The map’s caveats no longer say it never uses a release’s own coordinate: it uses one only where it passes the location check or was accepted on review, and never a 2023 coordinate. No figure changed; energy.csv, buildings.csv, classes.json and footprint_overrides.csv are byte for byte the 2026-09-24 release, and so are the map’s geometry and value files. The full list is in the release notes.

density_ca.csv gains a display name for each community area, facts.json states the hexagon width in miles, and the map’s own files are published and checksummed under site/. No figure changed; the other data files are byte for byte the 2026-09-23 release.

The release adds facts.json: every figure this page states, with its definition and source file. No figure changed; every other data file is byte for byte the 2026-09-22 release.

An audit of the data, the repository and this page found no error in the figures, and none of them changed. What changed is how the work is described. Sixteen properties whose location the matcher could not settle were said here to have been placed “by hand”: they were resolved the same way the match check was, by AI reviewers working to a written rubric, which this page now says. The map’s own caveats say it too, rather than leaving it to the repository. The code that builds all of this is now published alongside the data, under an MIT license that previously covered nothing a reader could see. Rush’s 2023 coordinate lands in North Center, not Lakeview. The full list is in the release notes.

The first version of this page and its data, from 2026-09-18, had errors that are now corrected. Total energy was site EUI times floor area, which the inflated floor areas pushed about 12% too high. UChicago Medicine was not placed on the map, and so was missing from the density view without notice. McCormick Place was placed in Armour Square, and Mount Sinai and Hilliard Homes were matched to the same addresses on the North Side. A check on construction years had moved 224 properties onto neighboring buildings, including the largest energy user, and some matches put large buildings on sheds. The match rate was reported against a threshold lowered the day the result came in, and no error rate had been measured. Names of owners and condominium associations showed on some properties, and the counts on the home page did not add up. The full list is in the release notes.