Britain and Ireland have 4,147 hillforts. A handful take almost all of the attention. I measured ten years of reading to find out which ones, and why.
My usual work is on attention on social media: which parties, posts and politicians get seen and which disappear. The same question works for the past. Britain and Ireland's hillforts are a perfect test. There are thousands of them, they were all built for broadly the same reasons, and archaeologists have measured every one of them in the same way. So if attention followed importance, we would know what to expect.
It does not. Between 2016 and 2025 the English, Welsh, Irish and Scottish Gaelic Wikipedias served about 26.5 million reads of articles about these sites. Half of that went to just eleven of them. Most hillforts get no readers at all, because there is no article to read.
Every dot below is a hillfort in the Atlas. Grey dots have no English Wikipedia article. Orange dots do, sized by how many people read about them each year. Switch to size by area to see the same map drawn by how big the forts actually are. The two maps are not the same map.
Line every hillfort up from most-read to least-read and add up their share. The curve below shoots straight up. The single most-read site, Edinburgh Castle, takes 12% of everything on its own. The top four take 34%. Eleven sites take half, and 161 take nine-tenths. The other 3,986 share what is left.
If attention followed importance, size would be the obvious place to look. The largest hillforts were the regional centres of the Iron Age. Size does help. In the largest tenth of sites (4 hectares and up), 39% have an English article. In the smallest tenth it is 5%. But that still means 61% of the biggest hillforts in these islands have no English article at all.
And once a site has an article, size barely predicts how many people read it. The scatter below plots every hillfort with an English article. Hover or tap for names. The sites floating at the top are not the biggest. They are mostly the ones with a castle on top, or at least a castle in the name.
The most-read "hillforts" are mostly famous for what came later. Edinburgh, Bamburgh, Stirling, Dover, Dunnottar and Urquhart are all in the Atlas because a prehistoric or early medieval fort sits underneath the castle. Readers come for the castle. The next site on the list, Old Sarum, is read mostly as the site of a Norman castle and the first Salisbury Cathedral, even though the Atlas does not record it as reused, so the true share is higher still. Across the whole dataset, the 445 sites that were reused after AD 400 or carry "castle" in their name are 11% of hillforts and take 64% of the reading.
The most-read Iron Age sites are mostly read for their medieval stories.
Among the mainly prehistoric sites, the leaders are a familiar list: Mam Tor in the Peak District, one of England's best-known hill walks; Maiden Castle in Dorset; Solsbury Hill, helped by a Peter Gabriel song; Dún Aonghasa on the Aran Islands; and South Cadbury, which John Leland linked to Camelot in 1542. A good story, or a good view, beats a big rampart.
To separate these effects I modelled two things together for every site with a recorded area: whether it has an English article at all, and, if it does, how much it is read. The predictors fall into three groups. Importance: size, ramparts, legal protection, excavations and dating. Fame and reuse: Roman or medieval reuse, a castle name, a storied name like Arthur, Caesar or "the giant's". Access: the coast, the distance to the nearest town, and the population within 25 km. Each model also controls for the nation.
Two findings stand out. First, attention follows archaeological attention. Every doubling of the number of excavations raises the odds of an article by 65% and the reading of an existing article by 58%. Legal scheduling nearly quadruples the odds. The public reads about the places archaeologists have already written about, much as citations pile up on papers that are already cited. It is probably a loop. Second, the castle effect survives all the controls: a "castle" name raises reading by 82%. Access barely matters. Distance to a town and nearby population explain almost nothing. Wikipedia readers are not mostly visitors planning a walk.
That last point shows up in time too. Visitor numbers at heritage sites peak in summer. Wikipedia reading of hillforts does not: the median prehistoric site gets almost exactly the same reading in July as in March. It dips a little in November and December.
The yearly averages hide the way attention actually arrives. It comes in bursts, and the biggest bursts are handed out by editors. When Cefnllys Castle in Powys was Wikipedia's front-page featured article on Christmas Eve 2020, it got 280 times its usual monthly reading in a single month. The Trundle in Sussex had the same treatment in January 2022. News does it too, on a smaller scale: Tap o' Noth when archaeologists announced in May 2020 that it was the largest Pictish settlement ever found, and Stanwick when the Melsonby hoard of Iron Age chariots was announced in March 2025.
This is the same pattern I find on Facebook: a long, quiet tail and a few moments when an algorithm or an editor points everyone at the same thing. The difference is that on Wikipedia, the gatekeepers are volunteers, and the bursts fade almost completely within a month.
One nation breaks the pattern. English Wikipedia covers 29% of England's hillforts, 11% of Wales's, 7% of Scotland's and only 3.6% of those in the Republic of Ireland. But Welsh Wikipedia has its own articles on 38% of Welsh hillforts. In Wales, a hillfort is more likely to be written about in Welsh than in English. Ireland's forts, including some of the largest in the Atlas, are almost entirely unwritten in any language.
Here are the largest sites in the Atlas with no article in any of the four languages I checked. Half are in Ireland, including the four largest. Spinans Hill in County Wicklow, the largest enclosure in the whole Atlas at 130 hectares, is part of the Baltinglass hillfort complex. Some very large sites, such as Ham Hill in Somerset and Hengistbury Head in Dorset, do have articles, filed under the hill or the headland rather than the fort. The sites in this table have nothing in any language.
Sites. All 4,147 hillforts in the Atlas of Hillforts of Britain and Ireland (Lock and Ralston 2017), via the copy of the Atlas source data released with Mike Middleton's Hillforts Primer. Area, ramparts, scheduling, dating, excavation counts and later reuse all come from the Atlas.
Articles. Each Atlas record has a Wikidata item. I took the English, Welsh, Irish and Scottish Gaelic articles linked to it. For sites with no linked article I matched nearby Wikipedia articles by location and name, which catches pages filed under a castle, a hill or a headland. These matches are conservative, so a few real articles will still be missed.
Reading. Monthly human pageviews from the Wikimedia API, January 2016 to December 2025, excluding bots and crawlers.
Why not Google reviews? I planned to add Google Maps review counts as a second attention measure. Google's terms do not allow reviews or place data to be scraped or stored for analysis, so I left them out. Wikimedia Commons photo counts are in the data release as an extra measure.
Car parks. I also planned to count car parks near each site from OpenStreetMap, but the public query servers were unavailable while I collected the data. Access is measured by distance to the nearest town of 10,000 people and the population within 25 km instead.
Models. Logistic regression for having an article and OLS on log yearly reading for sites with one, with robust standard errors and nation controls. Full tables and code are in the paper and data release.