Firms using AI export slightly more, a review of 22 studies finds

The effect varies widely. Heterogeneity was extremely high (I2 = 99.09%), so the authors say the evidence does not support a universal claim that AI increases export performance. It points instead to a small average association whose size depends on context and measurement.

Meta-regression showed that a country's development level, its digitalization capacity, the research design and the way export performance is measured systematically shape the AI–export link. The authors also note that some studies found a short-term negative effect that later turned positive, and that some results were not significant.

The study synthesises earlier empirical research on whether firms' AI adoption improves export outcomes such as export intensity, product quality and market diversification. Pooling 22 studies with 404,664 observations, it finds a statistically significant positive association of r = 0.0633.

The authors stress that the average hides large differences between studies. An I2 of 99.09% means nearly all variation reflects real differences between contexts and methods rather than chance, so the average should not be applied to any single firm or country.

Meta-regression identified four moderators: development level, digitalization capacity, research design and the way export performance is measured. The authors say the results can inform policymakers and managers considering AI for sustainable economic outcomes, while some primary studies showed a short-term negative effect that later turned positive.