The pitch for AI audio guides is compelling on paper. Multilingual. Conversational. Personalized. Available to every visitor. But museum professionals have heard compelling pitches before. What they want to know is: what actually happens when a real, named museum deploys one?
This article frames the answer honestly. The AI audio guide market is young enough that large-scale, peer-reviewed studies don't exist yet. What we can do is point at one named deployment we know intimately, one named deployment from a peer institution that has also published results, and the wider research on audio guide adoption that puts those numbers in context.
The named deployment: Museo Miraflores, Guatemala City
Museo Miraflores is an archaeological museum at the Kaminaljuyú site in Guatemala City. In 2021 they launched a traditional handheld audio guide built with a major partner. The project was expensive, the production cycle was long, and once labels started changing, the guide couldn't keep up. By 2024 the devices had been moved to the archive room and were only handed out on request.
In 2025, director Maria Gadsden decided to try again. Miraflores became the first museum to deploy Musa as a paid, permanent AI audio guide. The full story is in the Living Guides at Museo Miraflores write-up, including how the spatial layer works and how the team handled the connectivity dropout in the tomb gallery. The numbers below are the headline results.
Headline results from Museo Miraflores (Musa internal data, Museo Miraflores deployment, 2026):
- 120,000 annual visitors
- Roughly 5x higher uptake than the previous handheld devices
- Average tour sessions over 40 minutes, with some near two hours
- Launched as a paid, permanent guide in roughly eight weeks, including a four-week pilot
Maria Gadsden, Director of Museo Miraflores, on the experience:
This is the future of guided experiences. All our visitors prefer Musa over traditional audio guides and we've learned so much about their interests.
A visitor at Miraflores, Karin Cardoza:
It made navigating the museum very easy. I loved taking up the escalators while it showed me interesting details along the way.
These numbers are specific to one venue, with one collection, in one city, with one team. They're a strong signal, not a benchmark. The honest framing is that Miraflores is an early-mover deployment that worked, and a useful reference point for any museum running a similar pilot.
A second named data point: Smithsonian American Art Museum
Musa is not the only AI guide with a named, public deployment. The Smithsonian American Art Museum (SAAM) launched an AI-powered audio guide alongside its September 2023 reopening, and MuseumNext has published the results. Per MuseumNext's writeup, within three months SAAM saw 35,000 unique users, 25% of whom returned for further engagement, with about 70% of users under the age of 40. SAAM also flagged something we recognise from our own work: previous traditional audio guide attempts at the museum had faltered partly because of staffing and distribution friction, not because visitors didn't want interpretation.
Two named museums on different continents, working with different partners, are now reporting that AI-powered guides are pulling adoption out of the floor that traditional handheld and app-based guides had been stuck on.
What that floor actually looks like
For context: the published research on traditional audio guide adoption is sobering. Nubart's analysis of 175 museum audio guide apps found an average adoption rate of about 2.47%, meaning on a typical day, fewer than three out of every hundred visitors actually downloaded the museum's audio guide app. Industry observers like Locatify describe how dedicated handheld rental systems have largely disappeared outside specialised contexts, replaced by web-based and BYOD (bring your own device) experiences. Even those have struggled to climb out of single digits when the experience is a static, pre-recorded tour locked behind an app store.
Colleen Dilenschneider's IMPACTS data suggests that visitors are also shifting how they discover and engage with cultural organisations, with AI tools moving into the discovery layer itself. That doesn't directly measure audio guide adoption, but it points the same direction: visitor expectations of conversational, personalised digital experiences have already moved.
This is the backdrop against which Miraflores's roughly 5x lift, and SAAM's 35,000 users in three months, become legible. They're not unbelievable numbers. They're what happens when a guide gets two things right at once: the friction is low enough that visitors actually start it, and the experience is good enough that they actually finish.
What we hear from operators we work with
Beyond the venues we've named, we hear consistent themes from operators in our network. We're not going to attribute specific numbers to anyone who hasn't published their own data, but the pattern is worth flagging.
Tera Greenwood of SSA Group, on what today's guest-journey awards recognise:
removing friction and creating those seamless, personalized experiences that today's guests expect.
The friction point comes up in every conversation we have with directors who've lived through a traditional audio guide deployment: the desk, the line, the device, the deposit, the language menu, the dead battery. Each of those steps is a place where a visitor can simply give up. AI guides delivered via QR code on a visitor's own phone collapse most of that.
What to measure when you run your own pilot
The most useful case study for your museum is the one you generate yourself. A pilot with your collection, your visitors, and your data produces evidence that no external case study can match. With that in mind, here's what a useful 30-day pilot tracks:
Adoption rate. What percentage of visitors actually start the guide? Compare against your previous baseline if you had one. If you didn't, just record where you land.
Completion rate. How far do visitors get through the tour? High starts with low completion is a content quality signal, not an adoption signal.
Engagement depth. With a conversational guide you can measure how many visitors ask questions, what they ask about, and how long they stay engaged per stop. Pre-recorded guides simply can't produce this data. It often surfaces objects that need better interpretation.
Language distribution. Which languages are your visitors actually using? This frequently surprises museums and reshapes decisions far beyond the audio guide.
Downstream effects. Review scores, return visits, time in museum, gift shop spend. Hard to attribute cleanly, but these are the metrics directors and boards care about. Track them before and after, look for trends, don't over-claim.
For a structured walkthrough, see How to Pilot an AI Museum Guide and How to Launch a Museum Audio Guide in 30 Days.
Where the evidence still has gaps
Honesty requires noting what these case studies don't yet prove:
Peer-reviewed, multi-venue studies. The published academic literature on AI audio guides is thin. Most of the strongest evidence right now is single-venue, vendor-adjacent, or both. Treat individual numbers as venue-specific until that catches up.
Long-tail effects. Both the Miraflores and SAAM data points are still relatively recent. Effects on repeat visitation, membership conversion, and reviews accumulate over years, not months.
Generalisability. Miraflores is an archaeological museum in Guatemala City with rich narrative content. SAAM is a major art museum with international visitor flow. A small regional history museum or an outdoor heritage site will see different patterns. Pilot, measure, decide.
What is reasonably solid: the friction floor for traditional and app-based audio guides has been documented at painfully low single-digit adoption rates. Two named AI guide deployments at very different institutions are now publicly reporting much higher engagement. That's not a peer-reviewed conclusion, but it is the strongest evidence the field currently has, and it points one way.
If you're considering a pilot and want to structure it for maximum learning, with measurement that survives scrutiny when you bring the numbers to your board, let's talk about what that looks like for your institution.