When millions of people fled Ukraine following Russia’s invasion in 2022, governments and humanitarian organizations used mobile phone and online platform data to track movements and identify where support was needed.
Similar approaches have been used to monitor population displacement in Cameroon, and in Gaza during the Israel-Hamas war. This is part of a wider shift in which governments, organizations and researchers are using a new generation of data — including mobile phone records, social media activity, satellite imagery and online search patterns — to understand migration.
The value of these tools is clear. Migration is dynamic. These digital traces reveal population movements in real time, often months or years before official statistics become available.
But every form of measurement captures some realities while obscuring others. As researchers working on migration and digital technologies, we have found that while digital data are analytically valuable, they exclude key populations and cannot capture migration as a lived experience.
Migration’s digital traces
In 2024, Canada’s population grew by 744,324 people; 97.3 per cent of them were international migrants, according to Statistics Canada. Yet the number of migrants in Canada at any one time depends on how we measure migration in the first place — what counts and who is included?
For instance, the Canadian Census counts country of birth. Administrative records track visas and immigration status. Surveys can capture intentions, experiences and settlement trajectories. Data on temporary migrants are based on temporary work or study permits, but such data were not even part of the immigration policy debate until 2023.
None of these sources measures migration directly. Each is a proxy — a stand-in for something that is difficult to track — that captures just one aspect of a complex reality.
As Canada now advances its national AI strategy, we must ask: what happens when migration is increasingly understood through digital proxies that do not directly observe migrants themselves?
Read more:
Canada’s AI strategy must reckon with the environmental implications of data centres
Elderly, digitally excluded
Digital data are not inherently flawed. On the contrary, they often reveal patterns that traditional sources cannot.
Consider mobile phone data. Researchers often infer movement from changes in the location of a device. This can reveal important mobility patterns at a scale that surveys could never achieve.
A moving phone does not automatically indicate migration, however. It may reflect commuting, tourism, temporary displacement or even a device being shared among family members. The data show movement, but not necessarily migration.
The same challenge appears with social media data. Researchers can use online activity to study migrant networks, integration processes and migration-related discussions. Yet social media users are not representative of all migrants. Those who are older, digitally excluded, undocumented or economically vulnerable are often less active and hence less visible in these datasets.
The people most affected by migration policies may therefore be among the least represented in the data used to study migration.
Lived experience matters
Migration has always been measured through proxies such as country of birth, nationality and status. But none of these dimensions fully capture a person’s migration history, identity or sense of belonging. What is new about digital proxies is their scale, speed and growing role in decision-making.
This is where artificial intelligence enters the picture. AI systems do not generate migration knowledge from scratch. They learn from existing data, including mobile phone records, social media activity, administrative databases and other digital traces.
In other words, AI does not observe migrants directly either. It relies on the same measures. The difference is that AI can process these data at unprecedented scale and speed, turning proxy-based indicators into forecasts, classifications and decision-support tools. As a result, the assumptions built into migration data become increasingly consequential for policy development, enforcement or support for integration.
This distinction matters because migration is not simply movement. A satellite image can reveal population change. A phone signal can reveal mobility. An online search can indicate interest in moving abroad. Yet none of these signals can tell us why a person moved, whether that movement was voluntary or forced, how long it will last or what it means for the people involved.
These dimensions of lived experience matter more than recording movement in real time.
Data-driven governance
The challenge for policymakers is not whether to use digital data. Increasingly, these data are too useful to ignore. The challenge is understanding what each source can and cannot tell us.
No single dataset offers a complete picture. Administrative records provide legal and demographic precision but often arrive slowly. Surveys reveal motivations and lived experiences but are costly and difficult to conduct during crises. Digital traces offer speed and scale but may overlook important populations and contexts.
Each source captures different dimensions of migration. Together, they provide a richer understanding than any could alone.
As governments invest in artificial intelligence and data-driven governance, this lesson becomes increasingly important. The availability of real-time big data should not obscure other types of data that complement the picture.
If AI is truly to work for all, as the Canadian AI strategy suggests, we must look beyond algorithms themselves and pay closer attention to the data on which they depend.
The question is not whether we use proxies to understand migration. We always have. The real question is which proxies we use, what they reveal and what they leave unseen.
Anna Triandafyllidou receives funding from the Social Sciences and Humanities Research Council of Canada, Canada First Research Excellence Fund, Horizon Europe, the Mariam Assefa Foundation, MITACS in Canada.
Tuba Bircan receives funding from the European Commission Horizon Europe, MSCA, Belmont Forum, Flemish Scientific Agency (FWO) and Vrije Universiteit Brussel.
