Artlas AI: How Museums Are Using Personalised Guides to Transform Visitor Experiences

What Artlas Actually Does

Artlas builds an AI layer on top of a museum’s existing knowledge infrastructure. It takes collection records, curatorial texts, and educational materials and structures them into a system that generates personalised visitor experiences in real time.

The result is an app that combines AI-generated audio guidance, artwork recognition, navigation tools, and conversational Q&A — all adapted to the individual standing in front of a painting.

Since December 2025, the platform has generated more than 25,000 personalised guides. It’s currently being piloted at Mori Art Museum in Tokyo, Dib Bangkok, and the Institute of Contemporary Art Miami. It supports more than 20 languages.

That’s not a proof of concept anymore. That’s early traction.

The Problem With Traditional Museum Interpretation

Here’s what Yao told The Art Newspaper directly:

“Most museum interpretation is still one direction. A wall label, an audio guide or a museum app usually speaks to everyone in the same way.”

Think about what that means in practice. A seven-year-old and a PhD art historian stand in front of the same Seurat painting and receive identical information. Neither gets what they actually need.

Traditional audio guides were a step forward in the 1990s. In 2026, they’re a bottleneck.

The deeper issue is structural. Human-guided tours are constrained by time, language, staffing, and cost. Not every visitor can join a tour at the right moment. Not every museum can afford multilingual educators across every gallery, every day.

AI doesn’t replace that human expertise. But it can scale it.

How Personalisation Works in Practice

Artlas uses a visitor’s interests, available time, language preference, and knowledge level to shape what they receive. The same artwork can generate completely different experiences depending on who’s asking.

Yao uses Georges Seurat’s A Sunday Afternoon on the Island of La Grande Jatte as a concrete example:

• A child gets guided through a visual game — spotting details, finding patterns, engaging with the surface of the painting.
• An adult receives social and historical context — what 19th-century Parisian leisure culture looked like, what Seurat was reacting against.
• An art specialist gets a technical deep-dive — colour theory, pointillist technique, how specific pigments have shifted over time.

Same painting. Three completely different doorways into the work.

“Personalisation is not just about making the language easier or shorter,” Yao says. “It is about choosing the right doorway into the artwork.”

That framing matters. It shifts the product from a translation tool into an interpretive layer — which is a fundamentally different and more valuable proposition for cultural institutions.

The Hallucination Problem — and How Artlas Addresses It

Any serious conversation about AI in museums has to confront accuracy. Hallucinations — where AI generates plausible-sounding but incorrect information — are a real risk, and in a museum context the stakes are higher than most.

“A wrong date is one thing,” Yao says. “A careless interpretation of identity, religion, colonial history or an artist’s intention can be much more serious.”

Artlas handles this through source restriction. The AI is instructed to stay within approved museum content and verified sources. When information isn’t available, the system is designed to say so rather than fabricate an answer.

Partner museums can review, edit, and approve content generated through the platform. Institutional control over interpretation is preserved even as AI scales that content across dozens of languages and audience types.

This is the right architecture for a trust-sensitive environment. Museums aren’t just repositories of facts — they’re stewards of cultural meaning. Any AI tool operating in that space needs to respect that responsibility explicitly.

Privacy and Data Protection as a Core Design Principle

Artlas is built to minimise the visitor data it collects. Museum content, interpretation standards, and approval processes stay under institutional control.

“For museums, trust is not only about factual accuracy,” Yao says. “It is also about privacy, data protection and clear boundaries around how AI systems use cultural and visitor information.”

This is a smart positioning move, but it’s also genuinely important. Museums operate in a public trust context. Visitors expect their behaviour inside a gallery not to be harvested for ad targeting or behavioural profiling.

By designing data minimisation into the platform from the start, Artlas removes a significant barrier to institutional adoption — and signals that it understands the cultural sector’s values, not just its technology needs.

The Competitive Reality: AI Is Already in the Museum

Here’s the argument Yao made at Berlin Gallery Weekend — and it’s a sharp one.

Visitors are already using ChatGPT, Gemini, and Claude to look up artworks while standing in galleries. General-purpose AI is already inside the museum. The question isn’t whether AI will enter the space. It already has, through visitors’ phones.

The real question is whether museums will have a responsible, institutionally guided AI layer — or whether they’ll cede that interpretive role to tools that have no understanding of the museum’s collection, curatorial framing, or content standards.

That’s a compelling case for institutional adoption. It reframes Artlas not as an optional upgrade but as a necessary response to something already happening.

What This Means for AI Adopters and Cultural Institutions

If you’re evaluating AI tools for visitor experience, content personalisation, or multilingual engagement, Artlas represents a specific and instructive model.

What makes it worth watching:

• Domain-specific AI built on verified institutional content, not general web data
• Multilingual at scale — 20+ languages without proportional staffing costs
• Institutional control preserved — museums approve content, retain interpretive authority
• Privacy-first architecture — minimises visitor data collection by design
• Conversational and adaptive — not a static audio track but a responsive experience

The platform is still in pilot phase. But the underlying approach — using AI to personalise access to expert knowledge while keeping human expertise and institutional trust at the centre — is a model that applies well beyond museums.

The Bigger Picture

The most interesting thing about Artlas isn’t the technology. It’s the philosophy behind it.

Yao isn’t trying to replace curators or educators. She’s trying to give every visitor the kind of experience that was previously only available to those who could afford a private tour, spoke the right language, or happened to arrive at the right time.

That’s a genuine democratisation of cultural access — and it’s built on a foundation of institutional trust rather than in spite of it.

Museums that adopt this model aren’t outsourcing interpretation. They’re extending it. And in a world where visitors are already turning to general-purpose AI for answers, that distinction matters more than ever.

The institutions that move first to establish a responsible, accurate AI layer will shape how their collections are understood for the next generation of visitors. The ones that wait will find that interpretation has already happened — just without them.