The appeal of asking
Most exhibits tell visitors what the museum has decided to say. A conversational exhibit reverses that: the visitor asks, and the exhibit answers. The format is attractive because curiosity does not follow a label’s structure. A child may want to know whether a ship’s crew was ever seasick, while an adult wonders about the cost of a tool. A question box can meet those interests directly.
Conversational formats range from a simple screen with a set of tappable questions to an open text or voice interface that accepts almost any wording. They look similar to the visitor, but they behave very differently, and the differences matter more than the surface.
Two broad designs
Scripted question and answer. The museum writes the questions and the answers. The visitor picks from a list, or types or speaks a question that is matched to the nearest prepared one. Everything the exhibit says has been reviewed in advance.
Generative or open-ended chat. The system composes answers on the fly from a body of source material. It can respond to almost any phrasing, including questions nobody anticipated. It can also produce wording nobody reviewed.
A third, hybrid approach pairs a free-text box with a fixed set of vetted answers, and shows the closest match or admits it has none. Many cautious projects choose it for exactly that reason.
What conversation can do well
- Lower the barrier. Typing or speaking a question feels more natural to some visitors than navigating a menu.
- Surface the unexpected. The questions visitors actually ask, recorded and reviewed (with attention to visitor privacy), can show what labels fail to explain.
- Support different levels. The same answer can be offered in a short form and a longer one.
- Give a human-like voice. A character or historical figure can be presented as a storyteller, though this needs care about what is invented and what is documented.
Staged well, a conversational station can feel like a friendly guide at the elbow. For a school group, a chance to ask a question may be more engaging than a panel to read.
What it cannot promise
This is where planning needs honesty.
It cannot promise to know everything. Any exhibit has a limited body of source content. The design should say plainly what the exhibit is about and what it covers, so that visitors do not assume it can answer any question.
It cannot promise accuracy without work. A scripted exhibit is only as correct as its script. An open-ended system may state something confidently that is wrong, or blend facts from different periods. Whatever the technology, the museum remains responsible for what is said in its galleries. Reviewing outputs, restricting the system to approved source material, and testing with awkward questions are all sensible precautions, and no precaution removes the risk entirely.
It cannot promise neutrality. Choices about what to include shape every answer. The issue of collection bias applies as much to a chatbot as to a label, and sensitive content needs particular care because a casual, friendly tone can sit badly with distressing subjects.
It cannot speak for communities. Where an exhibit presents a living culture’s views or history, words put in a simulated voice require consultation with source communities. Cultural safety applies when a voice is imitated or a person is represented.
It cannot promise confidentiality unless designed for it. Typed or spoken questions are personal data of a sort. Before anything is recorded or stored, the project should decide what is kept, for how long, and who can see it, and should tell visitors in plain words.
Handling the unanswerable
Every conversational exhibit will meet questions it cannot answer, off-topic remarks and the occasional rude input. The quality of the design shows in how it fails. Useful habits include:
- a clear “I can’t answer that, but here is what I can help with” response
- suggested questions on screen so visitors know what is in scope
- filters for abusive or inappropriate language, with a calm reply
- a visible route to a staff member or a printed source
- a limit on session length and an automatic reset between visitors
Because the people who use the station will often be groups, with several voices and some mischief, testing in realistic conditions matters. Prototype testing with real visitors, and watching how they treat the exhibit, usually shows problems that a quiet internal test misses.
Voice or text
Text suits quiet galleries and gives visitors time to think and correct themselves, but typing on a shared screen is slow and may exclude some people. Voice is quicker and feels natural, but gallery noise, accents and the discomfort of speaking aloud in public can all reduce reliability. Offering both, along with tappable suggestions, covers more visitors. Whatever the input, answers should be available as text, and spoken replies need captions or transcripts. Support for screen readers and for plain language matters here, because a conversational screen can look simple while remaining difficult to use in practice.
Is conversation the right format?
A conversational exhibit is worth considering when visitors have varied questions that a fixed layout cannot anticipate, and when the museum is prepared to maintain it. It is a poorer choice when answers must be exact, when the subject is sensitive, or when nobody will be able to review and update the content. A well-organized set of tappable questions, with answers written and checked by staff, achieves much of the benefit with far less risk, and for many galleries it is the more dependable starting point.