When the American Alliance of Museums asked 2,045 US adults who should produce museum content, more than 70% rejected the use of artificial intelligence to write exhibition texts. Another 19% were comfortable with staff using it to develop ideas, but not to write the final text. Nine per cent supported its use in exhibitions.
The survey, conducted in January 2026, did not cover every possible use of AI. The alliance notes that it excluded some behind-the-scenes work, such as caring for collections or identifying objects. That distinction matters: the strongest reservations concerned what museums communicate directly to visitors.
This is one of the boundaries explored in Five Ways That AI Can Humanize Museums, a report by Remuseum, an independent research project hosted by the Crystal Bridges Museum of American Art. Drawing in part on a January 2026 meeting with representatives of seven US museums, it considers five areas of use, from staff training to the visitor experience.
At the Cleveland Museum of Art, AI helps produce visual descriptions of artworks. Before extending the project to its online collection, the museum developed a guide, tested different models and wrote a set of descriptions by hand to serve as a reference. Descriptions made with AI assistance are labelled, and visitors can give feedback. The technology helps the museum work across a large collection, but people still set the criteria and check the results.
The National Gallery of Art in Washington is using AI to extract and summarise information from research the museum has already produced, and to connect its art, library and archive collections. It is exploring how knowledge developed for specialists might be organised for people with different needs. The starting point remains material selected and preserved by the institution.
At Crystal Bridges, AskRosie is designed for people visiting the museum. By pointing a phone towards an artwork, a visitor can start a spoken conversation drawing on collection data, interpretive texts and materials produced by staff. The tool is intended to support looking at the work without requiring a code or drawing attention away from it.
The Remuseum report acknowledges questions about intellectual property, data, labour, ethics and environmental impact. It does not treat them as settled, or replace the policies each museum needs to establish. Staff training, approved tools and review processes are among the conditions it identifies for deciding when AI is appropriate and when it is not.
These examples do not erase the reservations expressed in the survey. They do show why the uses need to be considered separately. AI may help people encounter artworks and find their way into existing research; the museum remains responsible for the knowledge it draws on, the tool’s limits and the accuracy of what it shares.
