Every building and structure in Wikidata dated before AD 1100, 6,837 of them in 94 countries, on one map. What were they built for, what are they now, and what kept them up?
Most of what people built a thousand years ago has gone. Houses rotted, walls were robbed for their stone, towns were rebuilt on top of themselves. A few things survived. I wanted to know which, and why.
So I took every building and structure in Wikidata, the open database behind Wikipedia, that has a date of construction before AD 1100 and a place on the map. After removing statues, inscriptions, streets and other things that are not buildings, that leaves 6,837 churches, temples, castles, tombs, bridges, houses, walls and towers in 94 countries. For each I worked out what it was built for, what state it is in, and what it is used for today, and compared them with 40,200 structures built between 1100 and 1599.
The short answer is not stone, or not mainly. The oldest survivors are overwhelmingly places of worship that never stopped being used: a congregation is the best maintenance plan ever invented. The other survivors are things nobody needed, in places nobody wanted: tombs, mounds and ruins in dry country, far from any city. And many of the buildings that "survived" did so by being rebuilt, so that the name and the site are old but the walls are not.
A word on the data. Wikidata is uneven. Japan, Italy and the United Kingdom together hold 46% of the records, and much of the world is thin. So I compare shares, not counts, and every finding is about the buildings Wikidata records, not every old building on earth. Dates are Wikidata's: 45% are given only to the century or millennium, and many date the founding of a church or castle rather than the walls standing now (section 06).
Each dot is one building or structure. Colour shows what it was built for, or switch to what it is today. Filter by era, condition, material, form or country, or jump to a region. Tap a dot for its details and its Wikidata and Wikipedia links. The grey rings (off by default) are 442 archaeological sites with an early date that Wikidata does not class as a building.
Almost two thirds of the survivors were built for worship: 4,468 churches, monasteries, mosques, synagogues and Buddhist, Hindu and Shinto temples and shrines. Next come castles, forts and walls (14%), then tombs and burial mounds (7%). Houses and palaces, the buildings people actually lived in, are only 4%.
The mix changes sharply with age. Before 1000 BC, the survivors are mostly tombs and monuments: pyramids, mastabas, obelisks. In the last thousand years before Christ, houses, palaces and forts appear, most of them now ruins. After AD 500 the picture flips: three quarters or more of everything that survives from each century is religious.
For each building I asked what it is now: still doing its original job, turned to another use, a museum, a ruin, an earthwork, or gone. 58% of all the survivors, 3,938 buildings, still seem to do what they were built for, and 99% of those are places of worship. Most of this is inference: where Wikidata records no change, no ruin and no other use for a religious building, I count it as still in use. A hand check of a random sample (section 06) supports that for most, but not all.
The contrast is stark. 88% of religious buildings are still in religious use and only 8% are ruins. Castles and forts almost never still defend anything: 30% are ruins and most of the rest stand with no recorded use. Houses and palaces fare worst of all: 72% are ruins or excavated sites. Nobody has lived in a house built before 1100 for a very long time, as far as Wikidata records.
If buildings disappear at a steady rate, older centuries should leave fewer survivors. They do, steeply. Wikidata has 12,178 structures begun in the 16th century, 5,526 from the 12th, 1,866 from the 11th, and only 330 from the 1st century AD. Some of the slope is about records rather than buildings: people simply built more as populations grew, and later buildings are better documented. But the shape is the shape of attrition.
Comparing the oldest survivors with the later set shows what the filter of time has kept. Before 1100, 73% of survivors are religious, funerary or commemorative; for 1100 to 1599 it is 59%. Houses and palaces are 4% of the old set against 12% of the later one. Restricting both sets to Europe, where Wikidata is strongest, keeps the pattern: houses and palaces are 3% before 1100 and 14% after.
I can only see survivors. Nothing here proves why a building lasted; the reasons are inferred from comparing the buildings that did survive, across eras, kinds, materials, climates and places. With that warning, four ingredients stand out, and then there is a catch.
of religious buildings are still places of worship, and only 8% are ruins. A congregation repairs the roof. Compare 30% ruined for castles and forts and 72% for houses and palaces.
of survivors from before 1100 are religious, funerary or commemorative, against 59% for 1100 to 1599. Temples and tombs are protected by respect or taboo, and a mound of earth has nothing worth taking.
of the 469 early buildings with a recorded material are stone, against 62% of 2,150 later ones; brick is 21% against 40%. Only 7% of the old set record a material, so treat this as a hint.
is the median distance from a ruin to a city of 100,000 people today, against 20 km for buildings still in use. West of 65° E, in the dry belt between 25° and 35° north that runs from Morocco to Iraq, 67% of survivors are ruins or sites, against 14% between 45° and 55° north (41% without the excavated city of Dura-Europos).
The first route and the fourth are opposites. A building can last because people never stop needing it, or because they stop needing it completely and leave it where no one will reuse the stone. The dangerous place is in between: useful enough to be quarried, not useful enough to be repaired. That is where most of the houses, palaces, forts and public buildings of the ancient world went.
A building survives a thousand years if it is either indispensable or forgotten.
Now the catch. A Wikidata "inception" date is often the date a church, monastery or castle was founded, not the date of the walls you can touch today. Japanese shrines are rebuilt again and again; English parish churches were enlarged in every century; abbeys burned and rose again.
To see how much this matters, I drew a random sample of items with an English Wikipedia article and checked the first 32 by hand against their articles. 14 still have substantial building fabric from before 1100. 10 are mostly later buildings on an older foundation, and for 8 the article does not say. The split runs along the survival routes: of the 16 still in their original use, only 4 clearly keep pre-1100 walls, while 10 of the 12 ruins, earthworks and monuments do.
Architectural styles in Wikidata tell the same story at scale. Of the 1,255 survivors with a dated style recorded, 248 (20%) list only styles that began after 1100, such as Gothic or Baroque. So the honest version of the recipe is this: use keeps a place alive, but it often does so by replacing the building. Abandonment keeps the original stones, but as a ruin.
Tap a card to fly to the building on the map.
The table lists every country with at least one record, its count, and its three oldest surviving entries (lost buildings left out). Use it with care: many "oldest" entries carry founding dates or broad century dates, and small countries may have only one or two records. Search or sort by clicking a column heading.
Two regional patterns stand out. East Asia's survivors are almost all religious and almost none are ruins: Japan alone has 1,380 records, 478 of them Shinto shrines and 595 Buddhist temples, many of them founded in the first millennium and repeatedly rebuilt in timber, plus 237 burial mounds (kofun). The Middle East and North Africa are the reverse: about half of their survivors are ruins, many in desert sites. Syria's count is dominated by one excavated city, Dura-Europos on the Euphrates, whose individual houses and temples are separate Wikidata items (260 records within a few kilometres).
The United Kingdom has 785 records, 11% of the whole map. 661 are religious, overwhelmingly parish churches, and 483 (62%) are dated to the 11th century, most only to the century. Wikidata's oldest UK "churches" are often the founding dates of monasteries and cathedrals whose present buildings are much later, and some are ruins, so I set its lists beside Wikipedia's curated list of the oldest buildings in the UK, which dates the oldest standing fabric.
Wikipedia's list of the oldest UK buildings has 99 entries dated before 1100, from Orkney's Neolithic farmsteads and tombs to Norman cathedrals. I could match 45 of them to my catalogue. Many of the misses are prehistoric: all 12 prehistoric entries, such as Skara Brae and the Knap of Howar, are archaeological sites in Wikidata rather than buildings, so they sit outside my main map. On that list the oldest church is St Martin's, Canterbury (597), followed by St Peter-on-the-Wall at Bradwell (654) and Escomb (670). Its houses are prehistoric and Roman remains plus Icomb Place (1085) and Westminster Hall (1097), and it has no bridge at all.
The north-west is thin on early buildings. My catalogue has 18 records in Lancashire, Cumbria, Merseyside, Greater Manchester and Cheshire, 5 of them in Lancashire. The oldest Lancashire entries are the Roman fort at Ribchester (Bremetennacum, AD 72, a ruin) and the two Heysham churches on the headland above Morecambe Bay: St Peter's, dated to the 7th century, and St Patrick's Chapel, 8th century. Wikidata gives no condition for St Patrick's, so my rules count it as standing, but it is a ruin, "built of sandstone rubble" in the 8th or 9th century according to its Wikipedia article: a good example of the gaps in the data. Lancaster Castle (1093) completes the county's list, with Penwortham Priory. Further north, Cumbria adds a run of 11th-century churches and the castles at Carlisle and Brough; Wikipedia's list adds Hardknott Roman Fort and the Roman bathhouse at Ravenglass.
Coverage is uneven. Wikidata records what volunteers and imports have put into it. Japan, Italy and the UK are rich; most of Africa, the Americas and South Asia are thin. Shares within the data are more trustworthy than counts across countries.
Dates are often founding dates. 45% of dates are only to the century or millennium, and many describe the foundation of an institution. My hand check found later fabric in 10 of 32 cases.
Condition and use are inferred. Wikidata records a conservation state for few items and a current use for fewer. I combined those with classes such as "castle ruin" or "former church", with labels, and with two rules of thumb: classical public buildings and buildings dated before Christ count as ruins unless a use is recorded, and religious buildings with no sign of change count as still in use. Section 06 shows how often that is right in a random check. Some errors remain: St Patrick's Chapel at Heysham, for example, and Chepstow Castle, which Wikidata marks as demolished.
Survivors only. Everything here survived at least to the point of being recorded. The "recipe" is a comparison of survivors, not a measurement of what was lost.
Data. From the QLever Wikidata endpoint (27 October 2026): all items that are instances of a subclass of architectural structure (Q811979) with an inception (P571) before 1100 and coordinates (12,794 result rows, 7,058 items), with their classes, country, material (P186), conservation state (P5816), current use (P366), heritage designation (P1435), architectural style (P149), number of Wikipedia sitelinks and English Wikipedia article. The same query for inceptions from 1100 to 1599 gives the comparison set (40,200 items after cleaning), and a query for archaeological sites (Q839954) with early dates gives the separate grey layer.
Cleaning. I removed 221 items: 211 that are only objects, places or routes (stelae, statues, reliefs, runestones, mosaics, streets, squares, gardens, settlement mounds), 6 with dates before 10,000 BC, and 4 without usable coordinates or dates. Where an item has several inception dates I keep the earliest. Countries come from the Natural Earth 10 m boundaries, so ancient states are placed in modern countries.
Classification. Original use, form (building, structure, earthwork) and condition come from regular-expression rules on the English class labels, listed in code/rules.py; every item's rule basis is in the data file. Material is P186 grouped into stone, brick, timber and earth, plus a few classes that imply a material (stave church, stone bridge). Distance to a city uses GeoNames places with 100,000 or more people. Latitude bands are a crude stand-in for climate; I did not use a climate dataset.
Comparison. For fair comparisons with the 1100 to 1599 set, which has no labels, ruin shares in section 04 use only recorded conservation states and ruin classes.