Airbnb prices can partly be explained by AI-generated neighbourhood descriptions
Airbnb listings in Manhattan, New York.
With the advent of AI, classic housing variables such as size, location and standard can be complemented by digitally created attributes that provide vivid descriptions of neighbourhoods – for example “vibrant”, “historic” and “affluent”. A new study shows that the researchers’ AI-generated neighbourhood descriptions help to explain accommodation prices in much the same way as physical characteristics.
Over the past decade, Airbnb has grown into a global platform that has not only transformed tourism and accommodation, but has also become a rich source of data for research on cities. The new study, published in Computers, Environment and Urban Systems, demonstrates how Airbnb prices can be explained not only by a property’s location and characteristics, but also by neighbourhood descriptions generated by AI tools such as ChatGPT, Grok and Copilot at the researchers’ request.
The study was conducted by IBF Visiting Professor John Östh together with an international research team from the Netherlands and Turkey. The researchers combine large-scale digital data on a substantial number of Airbnb listings with AI-generated information about the surrounding environments. The analysis is based on a so-called hedonic pricing model, a method used to better understand how the prices of goods and services (such as Airbnb accommodation) vary depending on their different attributes. In housing research, this may involve estimating how factors such as floor area, location, standard or proximity to amenities contribute to price. In the present study, the method is extended to include AI-generated neighbourhood descriptions as variables that may explain housing prices.
The study analyses a large number of Airbnb listings in several major US cities. The results show that AI-generated variables – for example descriptions of neighbourhood vibrancy, historical character or creativity – account for up to twelve per cent of the price.
“This means that it is not only the actual attributes of the dwelling that matter, but also how the place is described and perceived. Since AI is trained on sources ranging from social media to books and public texts, it reflects these images of the city. This implies that truths, falsehoods and everything in between that shape our view of the world are incorporated and given weight. We are now continuing our research to examine whether AI-generated pricing models can also be used to analyse the broader European housing market,” says John Östh, Visiting Professor of Human Geography at IBF.
Article
The article in Computers, Environment and Urban Systems is available through Open Access.
Hedonic price models, social media data and AI – An application to the AIRBNB sector in us cities