After Hastings the Normans held England with earth and timber, then stone. I mapped every castle site Wikidata knows in Britain, dated the ones I could, and asked how fast the castle frontier moved, where it bunched, and what Domesday Book says it cost.
In October 1066 William of Normandy won a battle in Sussex. Over the next generations his followers built castles from Cornwall to the Clyde: mounds of earth with timber towers, ringed banks, and later great keeps of stone. Many still stand, as grassy humps in fields or as ruins on hills.
I wanted to see the conquest as a map that grows. Wikidata, the open knowledge base behind Wikipedia, records 3,288 castle sites of the Norman kinds in England, Wales and Scotland: 958 mottes and motte-and-bailey castles, 162 ringworks, 10 keeps and 2,158 sites classed simply as castles. The catch is the dates. Only 317 of them (10%) have any date of foundation at all, and only 113 fall between 1066 and 1199.
So this is a story told twice: once with the few castles that carry a date, and once with the many that do not. And it is checked against the one firm witness from the time, Domesday Book, which in 1086 names 41 castles in its text.
Drag the slider to move through the decades. Red dots are castles whose Wikidata date falls in that decade; blue ones were already standing. Gold diamonds are castles named in Domesday Book, so standing by 1086; they appear from the 1080s. Tick Show undated sites to see the much bigger layer of castles with no date: that is where most of the Norman castle landscape actually sits.
If the Normans had conquered like a slow tide, castle dates would grow later the further you go from the battlefield. They do not. The dated castles of the 1060s were already spread across the kingdom: 22 of them, with a median distance of 215 km from Battle. By 1069 the furthest, York Castle, was 354 km away. By 1072 Durham Castle stood 451 km north.
Across the 106 dated castles in England and Wales, the link between year and distance is weak (Spearman 0.12) and not statistically clear. In the first six years, 1066 to 1071, the median dated castle was 235 km from Battle; from 1072 to 1086 it was 327 km; from 1087 to 1135, 297 km; and from 1136 to 1199 back down to 202 km, as building filled in nearer home. Of the 32 castles dated 1066 to 1072, 8 were already more than 300 km from the battlefield.
The castles did not follow the conquest outwards. They were the conquest, planted at once in the county towns and river crossings of the whole kingdom.
The busiest decade in the dated record is the 1060s, with 22 castles in four years. The 1100s look as busy, with 22, but 17 of those carry the year 1100 or 1101, which in Wikidata can stand for a date known only to the century. Overall 17 of the 113 dated castles (15%) have such a round year.
Before trusting any of this, I tested whether the dated castles are a fair sample of the rest. They are not. In England and Wales, 227 castle sites have a date (of any period) and 2,242 do not.
60% of the dated sites are Grade I listed buildings, against 6% of the undated: about 10 times as likely. Only 15% of dated sites are earthworks (mottes or ringworks), against 43% of the undated ones. Dated sites are also far more often among the castles Domesday names (7.0% against 0.8%). In short, Wikidata dates the big stone and royal castles, and leaves 972 grassy mounds and ringworks undated.
The good news for the frontier question is that the bias is about kind, not place. The median dated site is 282 km from Battle and the median undated site 293 km. So the dated sample says something fair about how far and how fast, and little about how many.
Even Domesday's witnesses go undated. Of the 35 Domesday castles I could match by name to a Wikidata item, 19 have no date in Wikidata at all, among them Hastings Castle itself, Cambridge, Gloucester, Richard's Castle and Penwortham. Of the matched items, 2 carry a date before 1066 (Monmouth 1060 and Rockingham 1000), which I keep as recorded but leave out of the time series.
Domesday Book was compiled twenty years after Hastings, and it mentions castles as a matter of fact: who holds one, what land it sits on, which houses it replaced. Reading the entries the Hull Domesday Project flags as mentioning a castle, and keeping only those where the 1086 text speaks of a castle at that place (not just a modern name such as Castle Hedingham), gives 41 castles. Add the dated Wikidata castles of 1066 to 1086 and I count 69 castles known to be standing by 1086.
Town castles were dug into living streets. Across 9 towns, the Domesday text counts 395 houses destroyed, emptied or taken over for castles. That is a floor, not a total: most towns say nothing on the matter, and York's entries in the source carry no English text to check.
Did the countryside around castles suffer too? I took every located Domesday place (11,974) and compared those within 5 km of a castle known by 1086 (588 places) with the rest. Near castles, 13.8% of places have an entry calling land waste, against 10.2% elsewhere. But much of that gap is geography: castles and waste both crowd into the north and the Welsh border. Comparing places within the same county, the gap shrinks to 1.5 percentage points. The median place near a castle was worth 1.00 times its 1066 value in 1086, and the median place elsewhere 1.00 times. I read this as no clear local footprint: the damage Domesday records is regional, not a ring around each castle.
The regional pattern is stark. In 1069 and 1070 William crushed a northern rising by laying the land waste. Domesday, sixteen years later, still shows it. Of 1,780 located Yorkshire places, 27% have an entry calling land waste, against 7.4% in the rest of the surveyed counties. Yorkshire's recorded value in 1086 was 39% of its 1066 value, where the rest of England, taken together, stood at 108%.
Yorkshire is not top of the list. Cheshire is, with 57% of places touched by waste, followed by Shropshire (33%) and Herefordshire (31%). Cheshire was harried in the same campaign, and the other two are the Welsh border counties. Wigmore and Clifford castles were, in Domesday's words, built "on waste land". The four most wasted counties of Domesday are the two great castle frontiers: the north and the March of Wales.
Count every castle site, dated or not, and divide by land area. A band 40 km either side of the England and Wales border covers 15% of the land of the two countries but holds 35% of their castle sites, and 36% of their mottes and ringworks. That is 39.5 castle sites per 1,000 km², against 12.4 in the rest of England and Wales: 3.2 times the density.
Herefordshire tops the table with 72.9 sites per 1,000 km², and Shropshire has the second highest density of earthwork castles. Wales as a whole has 43.5 castle sites per 1,000 km², England 12.0. One caution: Wikidata's Welsh records may simply be fuller, since they draw on a different national heritage list, and its Welsh "castles" include the castles of the Welsh princes as well as the Normans'. Even counting only mottes and ringworks, the border band is 3.3 times as dense as the rest.
The land between the Ribble and the Mersey appears in Domesday as an appendix to Cheshire, and most of the far north-west not at all. The dated castles there begin in the 1090s: Brough Castle (1092), Carlisle Castle and Lancaster Castle (1093). The median dated castle in Lancashire, Cumbria, Merseyside and Greater Manchester is about 1100, against 1091 in the south-east.
But Domesday puts a castle at Penwortham, on the Ribble, by 1086: "It paid 10d. Now there is a castle." So the first known castle in Lancashire is a few years older than the dated record shows. The north-west is also thin castle country: Lancashire has 6.2 castle sites per 1,000 km², against 72.9 in Herefordshire.
Scotland was never conquered by the Normans, but Anglo-Norman lords settled there in the twelfth century and built mottes. Wikidata holds 819 Scottish castle sites of these types, only 113 of them mottes or ringworks. Just 7 have a date before 1200, from Jedburgh Castle (1100, a round year) to Ayr Castle (1197). The earliest, Dunbar Castle at 1070, may not be a Norman castle at all. Most of Scotland's "castles" in Wikidata are later tower castles, so I treat Scotland as context, not evidence.
As a coverage check I took OpenStreetMap's castle features in Britain (1,310). 64% of them lie within 1 km of a Wikidata castle site in my set. The other way round, only 27% of Wikidata sites have an OpenStreetMap castle within 1 km, and only 14% of the mottes and ringworks. OpenStreetMap maps the standing castles; the earthworks live mostly in heritage lists, which is what Wikidata draws on. Neither is complete, and the gaps differ by country.
Castles. From a Wikidata extract (retrieved 27 October 2026) of UK items with coordinates, I kept items classed as castle, motte-and-bailey castle, motte, ringwork castle, shell keep, keep, castle ruin, enclosure castle or siege castle, deduplicated by item, and placed each in a county with ONS boundaries. Items classed only as tower houses, fortified manor houses, quadrangular castles, Welsh princes' castles or other later forms were left out; 107 items carry one of those classes as well as a castle class and stay in. Northern Ireland was left out. That gives 3,288 sites. Many "castle" items are later than the Normans; the type filter on the map lets you look at the earthworks alone.
Dates. A castle's date is its earliest Wikidata inception (P571). I never add dates of my own. Wikidata stores a year for dates of all precisions, so a castle "of the 12th century" can appear as 1100 or 1101; I flag years ending in 00 or 01 and test the results without them. In all, 12 items carry dates before 1066 and 192 after 1199; both are kept in the data but left out of the time series.
Distance. Straight-line distance in British National Grid from Battle, East Sussex (OS Open Names), the site of the battle.
Domesday. Entries come from the Hull Domesday Project's coded Domesday via the realm-domesday repository. I kept 41 castles where the translated 1086 text speaks of a castle at or of a place, placed at the Domesday place's grid reference (Penwortham, which has none in the source, at its Wikidata coordinates). I matched them to Wikidata only where the item's name is the same castle and it lies within 8 km (35 matched, 6 left unmatched as ambiguous). Waste is the source's flag for entries that call land waste; values are the 1066 and 1086 values of entries that give both, shared equally among the places an entry describes.
Density. Sites per 1,000 km² of land, by county (ceremonial counties in England, preserved counties in Wales, council areas in Scotland) and by hexagon. The March band is all land within 40 km of the England and Wales border.
Known problems. Wikidata coverage is uneven: the Welsh and English heritage lists feed it differently, and Scotland's castle items are mostly later. The dated sample favours big, listed castles. Domesday is silent on most castles, says nothing of the far north, and its Yorkshire entries in the source have no English text. Distances from Battle are a rough measure of a conquest that moved by road and river.