When AI Becomes the Front Door – Reflections from the CityDNA Helsinki Trend Room

The trend room session on artificial intelligence at CityDNA Helsinki was short by conference standards, but it carried more weight than its format suggested. Organised as the first step in a wider CityDNA conversation that will continue with an online workshop in June and a white paper in the autumn, the panel brought together Joshua Ryan Saha from the University of Edinburgh's Futures Institute, Stefan Kapel from Austria Tourism, and Jonathan, co-founder of Give Me and the network's AI partner.

The framing was deliberate. This was not a session about whether AI matters for destination management. That debate is closing. The more precise question raised in the room was what happens when the architecture of discovery itself starts to change, and when the destination website, long the central asset of most DMOs, ceases to be where the first conversation with the traveller takes place.

The website is no longer the front door

The data presented in the room was not dramatic in isolation, but the cumulative picture was difficult to ignore. Roughly 40 percent of US travellers now use AI for trip planning, a figure that has doubled in nine months. Organic website traffic for destination organisations in North America has dropped between 20 and 40 percent year on year. Surveys in the UK suggest AI adoption in travel planning has already reached around 50 percent, and European markets appear to be tracking a similar path.

What this points to is not a gradual shift in channel performance. It is a reorganisation of how travellers arrive at a destination idea at all. The website remains operationally important, but it is no longer the surface where the first impression is formed. Increasingly, that impression is shaped inside an AI model, based on what the model has been trained on, what it considers a reliable source, and how it synthesises what it finds.

A live poll made the gap visible. Only 58 percent of attendees were somewhat confident that their destination appears accurately in AI search results. For an industry whose core task is perception, this is a meaningful signal.

One of the more counterintuitive data points of the session concerned value rather than volume. Visitors who discover a destination through AI appear, on average, to be around four and a half times more valuable than those arriving via organic search. Lower traffic, in other words, is not the whole story. The intent behind AI-driven discovery, and the quality of match between traveller and place, seems to be significantly higher. Stefan made a related point. The longer a traveller interacts with an AI tool, the better it understands them, and the more precise its suggestions become. That produces customers who are more willing to pay for a closer fit to their own preferences, rather than simply comparing on price.

Visibility is becoming a statistical problem, not a deterministic one

One of the more useful conceptual points raised during the panel was the non-deterministic nature of these systems. If every person in the room asked the same question about Helsinki, each would receive a materially different answer. It can take thousands of queries before the same response surfaces consistently. This changes what visibility actually means in practice. It is no longer a matter of ranking on a page. It is a matter of how frequently, and with what emphasis, a destination is represented across a distribution of answers.

Stefan suggested a practical discipline that follows from this. Ask the same question of an AI model a hundred times, save the responses, feed them into another model, and ask it to identify the patterns, the gaps, and the weak spots. Statistics rather than single queries. It is a small operational shift, but it reflects a larger conceptual one. Destinations now need to think about their AI presence the way one thinks about a probability distribution, not a fixed position.

An adjacent point deserves attention. AI models still distinguish between reliable and less reliable sources. DMOs, national tourism organisations, and other officially mandated bodies have a structural opportunity here, provided the quality and depth of their data justify that trust. Being a reliable source in the AI era is not a matter of claim. It is a matter of what the data looks like when a model actually examines it.

Data sovereignty is a governance question, not a storage question

The discussion moved from visibility to data, and here the conversation became more strategic. Booking.com holds close to 69 percent of the European hotel OTA market, which means that most of the behavioural data generated by travellers interacting with European accommodation sits with a single platform. A recent digital maturity benchmark placed DMOs at 11.8 out of 20 on data and analytics capability. The asymmetry is considerable.

The panel was careful not to frame this purely as a deficit. Platforms contribute real value by structuring accommodation-level data at a scale that would be unrealistic for destination organisations to maintain alone. The more productive question is whether DMOs can position themselves meaningfully above the transactional layer, at the level of context, values, and regional narrative. This is the layer that AI models draw on when a traveller is still at the inspiration stage, well before they reach a booking platform.

On European AI sovereignty, the conversation was honest about the current state. Mistral is the most notable European base model, but most LLMs used by travellers today are developed in the United States or China. If destination organisations build their AI capabilities entirely on those supply chains, they inherit geopolitical and commercial dependencies that are difficult to unwind later. The issue is not a sudden shift, but a need for deliberate choices about where the core organisational infrastructure sits.

The more immediately tractable sovereignty question concerns destination data itself. Examples from New Zealand, where work is under way to curate national business data that AI models can draw on, and from Scotland, which is pursuing a similar direction, were offered as emerging models. As Stefan put it, the point is not where the data is stored. The point is the governance structure, the API standards, and the agreements that keep data accessible, reliable, and attributable to trusted local sources.

The role of the DMO is becoming more, not less, defined

There is a tendency, when AI is discussed in organisational contexts, to assume that technology will compress the space for human work. This view has a practical consequence worth naming. Around 42 percent of DMOs anticipate some level of funding risk within three years, and entry-level marketing roles across the industry have already declined by roughly 30 percent. The risk, as Joshua put it, is that AI is read as a reason to reduce destination organisation budgets precisely at the moment when trust and authenticity are becoming more, not less, important to how travellers make decisions.

The panel took the opposite view to the compression narrative, and the argument was persuasive. If AI can handle structured data, summarisation, and basic synthesis, the scarce capability becomes context. The ability to bring local knowledge, emotional understanding, and trust into the loop is precisely what models cannot generate on their own.

In a five-year view, Joshua referenced Jeanette Roush at Brand USA and the concept of the destination organisation as data steward. Curating, enriching, and representing the data that shapes how a destination appears in AI systems. He also raised the possibility that destination organisations could re-emerge as coordinating platforms for their SMEs, enabling more direct booking relationships in a landscape where agent-based systems may otherwise route value away from local operators.

Stefan framed the human role somewhat differently, as the work of bridging technical capability and traveller emotion. Inboxes, slides, structural drafting, and repetitive synthesis can increasingly be handled by AI. What remains, and what becomes more valuable, is the work of storytelling, context-setting, and sustaining genuine connection to place. On this framing, the destination organisation of the next decade is less an executor of marketing tasks and more a custodian of how a place is understood.

What Helsinki made visible

The wider implication of the session was not that destination organisations need to move faster on AI adoption. Many already are. It is that the terms on which destinations compete for attention are changing, and that the most consequential decisions now sit in areas that have historically been less visible. Data governance. Reliable-source positioning. A statistical understanding of AI visibility. And the cultivation of human context that cannot be automated.

The CityDNA initiative rightly treats this as a longer conversation. The June workshop and the autumn white paper will matter because the current landscape is not stable enough for quick answers. What Helsinki made clear, at least to me, is that the destination organisations that will remain relevant in the AI era are those that stop treating AI as a marketing channel and start treating it as a structural layer of how their destinations are known at all.