What language do museum visitors actually listen in?

Museums translate their interpretation into the languages they believe visitors need. The belief is rarely tested, because once a visitor walks away from the desk there is usually no way to see what they chose.

We can see it. Between 8 September 2025 and 7 September 2026, visitors at venues running Walkie Talkie started 34,560 audio guide chapters and listened 30,540 of them to the end. Each play records the chapter, the language the visitor selected and whether playback reached the end. Nothing else.

What follows is the whole picture, including the parts that make our own dataset look smaller than we would like it to.

Four venues account for 96 percent of everything measured here. One large art museum in a Nordic capital carried 54.7 percent of all plays on its own. A heritage site elsewhere in Europe carried 24.1, a regional art museum 9.3, and a second museum in that same capital 8.0. The remaining five venues share what is left.

So this is not a survey of the sector. It is a detailed look at a handful of real places over a full season, which is a different and in some ways more useful thing. Every number below is reported per venue rather than averaged, so you can find the site that resembles yours instead of trusting a mean that describes nobody.

At the large art museum in a Nordic capital, the local language accounted for 6.7 percent of listening. English took 52.9, German 16.4, French 9.2, Dutch 5.3 and Italian 4.3. The house language came fourth in its own building.

At a second museum in the same city the ranking repeated exactly. English 49.5 percent, French 23.0, German 19.5, then the local language on 7.8.

At an open air museum in a different country, the local language also came fourth, on 7.8 percent. German took 59.1.

That last figure is worth sitting with. Six out of ten people who listened there did so in German, and fewer than one in twelve did so in the language of the country they were standing in.

The last two entries are the ones that keep this honest. At a rural district museum the local language was 61.5 percent of listening, and at a parish church it was 62.7. Neither place sits on an international coach route.

So the rule is not that a museum's own language is always marginal. Its share collapses the moment the audience becomes international, and it collapses much further than most interpretation budgets assume. A site that counts a lot of foreign visitors at the door is probably already below ten percent and has never had a way to find out.

The common European compromise is to produce the local language plus English and call it multilingual. In this dataset English carried 35.6 percent of all listening, so that compromise reaches roughly a third of the people who actually press play.

German carried 20.4 percent and French 16.2. At the regional art museum, German alone reached 50.1 percent of listening, more than double the local language at 23.8. At the open air museum it was 59.1. Two very different institutions, in two different countries, with the same answer.

The long tail matters less than the headline suggests. Chinese, Polish and Japanese came to 2.8 percent between them. What moves the number is the set of large European languages that most museums could name off the top of their head.

Across all 34,560 plays, 88.4 percent ran to the end of the chapter.

Some of that is selection rather than brilliance. Nobody scans a code and presses play by accident, and the chapters here mostly run two to four minutes. A visitor who has already decided to listen is a visitor who tends to finish.

The breakdown is where it gets interesting. Dutch listeners completed 92.8 percent of chapters, French 90.8, Italian 90.7, English and Spanish 89.9, German 86.5. Listeners using the venue's own language finished 79.5 percent, the lowest of any language with real volume behind it.

The local visitor dips in, recognises the story and moves on. The visitor with no other way into the room stays until the end. Read that alongside the language shares and the audio guide starts to look less like an amenity and more like the only interpretation a large part of the audience actually receives.

Grouping unique listeners by the country they connected from gives a rough map of who was standing in these rooms.

Norway 22.2 percent, Germany 14.4, France 10.6, Italy 7.4, the United States 6.7, the United Kingdom 5.6, the Netherlands 5.2, Ireland 4.8, Spain 4.8, Denmark 3.8, Sweden 2.7.

German and French visitors together outnumbered every Nordic visitor in the set. Front of house staff know this. Interpretation budgets tend not to.

98.2 percent of plays came from a mobile phone. Laptops accounted for 1.6 percent and desktops for 0.2. Tablets managed exactly one play in a year.

Average listening ran to 6.9 chapters per visitor, which is a real visit rather than a curious tap. Anything that fails on an old handset with a weak signal therefore fails altogether, and anything requiring an app install or a queue at a desk loses people at the moment they were ready to listen.

A number without its caveats is just a claim, so here are ours.

One venue is more than half the dataset, and four venues are 96 percent of it. These are also places that chose a QR based audio guide in the first place, so they are self selected rather than representative.

The heritage site is left out of the language findings entirely. Its English guide was not published for most of the period, which pushed its listening towards French, Italian and German for a reason that says more about our delivery schedule than about its visitors. Its 8,333 plays still count towards completion, device and country figures.

We see which language a visitor selected. What they speak at home is invisible to us, so a Dutch visitor who picks English is counted as English. The Swedish completion figure of 51.5 percent rests on 167 plays and should not be treated as a finding. One small site recorded more completions than plays. That points at a tracking fault we are still chasing, so it has been excluded throughout.

Count the languages you offer against the nationalities arriving at your door. The list already in print is the wrong baseline. The gap is almost always German, then French.

Watch completion rather than downloads. A download tells you somebody was willing. A completion tells you the content held them all the way through, and the two numbers point in different directions more often than you would expect.

Assume the phone is the only device that matters, because for 98 percent of these listeners it was.

Treat your own language as one language among several rather than as the original that the others are translated from. At the three most international venues here it ranked fourth.

The period runs from 8 September 2025 to 7 September 2026, and the figures cover nine venues in northern and southern Europe and in North America that recorded at least fifty plays. Our own demonstration and test sites have been removed, along with plays on chapters that have since been deleted.

Events are collected through a self hosted Umami instance on servers in the EU, with no cookies, no cross site identifiers and no personal data. Venue names and identifying details are withheld, because the underlying figures belong to the museums rather than to us.

We will share the aggregate tables behind every number here with researchers, journalists and museum associations, and we will run the same breakdown for any museum that wants to see its own. Write to hello@walkietalk.ie.