Targeted AI R&D and governance, not overall AI vibrancy, raise tourism’s share of GDP; human-capital gains materialize with a one-year lag while COVID cut tourism by about 37%.

Main Finding

Aggregate national AI vibrancy (Stanford HAI composite score) does not have a measurable within-country effect on tourism’s direct share of GDP over 2017–2023. However, disaggregating the AI ecosystem shows two robust positive drivers of tourism’s GDP share: the AI-related R&D pillar (β = 1.811, p = 0.005) and the Policy & Governance pillar (β = 0.353, p = 0.037). The Talent pillar delivers a positive effect with a one-year lag (β = 0.183, p = 0.025). The COVID-19 pandemic sharply reduced tourism’s GDP share (~37% decline; β = –0.455, p < 0.001), and AI development did not significantly moderate that decline.

Key Points

• Data/sample: Unbalanced panel of 33 countries, 2017–2023; 169 country-year observations (166 after dropping missing controls).
• Dependent variable: Tourism’s direct contribution to GDP (% of GDP), log-transformed (ln(y)); alternative outcomes used in robustness checks: tourism employment share and tourism GVA share.
• Main regressors: Stanford HAI AI Vibrancy composite score and seven pillars — R&D, Economy, Talent, Policy & Governance, Infrastructure, Responsible AI, Public Opinion.
• Aggregate vs. pillar results:
  • Composite AI Vibrancy: no significant within-country effect on ln(tourism GDP share) (β ≈ 0.061, p = 0.622).
  • Pillar decomposition:
    • R&D: large, positive, and robust (β = 1.811, p = 0.005).
    • Policy & Governance: positive and robust (β = 0.353, p = 0.037).
    • Talent: significant positive with 1-year lag (β = 0.183, p = 0.025); suggests human-capital channel needs time to affect sectoral outcomes.
    • Other pillars (Economy, Infrastructure, Responsible AI, Public Opinion) not reported as robust positive drivers in main specifications.
• COVID-19 effect: substantial negative impact on tourism’s GDP share (≈ 37% drop).
• No evidence that higher AI vibrancy mitigated the pandemic’s effect on tourism GDP share.
• Robustness: results robust to clustered country-level SEs (CRV1), HC1 SEs, two-way fixed effects, lagged/dynamic specifications, alternative variable transformations and dependent variables.

Data & Methods

• Sources:
  • Tourism measures: UN Tourism database (Tourism Satellite Accounts); OECD indicators for alternative dependent variables.
  • AI measures: Stanford HAI Global AI Vibrancy Index (composite score + seven pillars).
  • Controls: World Bank WDI — FDI/GDP, internet users, trade openness, GDP per capita (ln), tertiary enrollment; COVID dummy for 2020–2021.
• Transformations:
  • Dependent variable: Box–Cox suggested λ ≈ 0.117 → used ln(y).
  • Composite AI Vibrancy: log(1 + x) to reduce right skew.
  • Pillars: Yeo–Johnson transformations to handle zeros/negatives and reduce skew.
• Econometrics:
  • Main estimator: one-way entity fixed-effects panel model (Hausman test favored FE over RE).
  • Standard errors: clustered at country level (CRV1); HC1 SE as alternative.
  • Extensions: pillar decomposition (nested models adding pillars as data availability permits), one-year lagged pillars (to reduce reverse causality), dynamic model with lagged dependent variable (Nickell bias cautioned), interactions (AI × digital infrastructure, AI × COVID) and mean-centering for interactions.
  • Diagnostics: tests for heteroskedasticity, serial correlation, cross-sectional dependence, VIFs (multicollinearity checked).
• Sample limitations: unbalanced panel, short T (≤7), progressively reduced sample when later pillars included (Responsible AI, Public Opinion available only from later years).

Implications for AI Economics

• Aggregate AI metrics can mask crucial heterogeneity. Policy and research should analyze AI ecosystem components (R&D, governance, talent, infrastructure) separately when assessing sectoral impacts.
• R&D and policy/governance matter more than overall AI vibrancy for tourism’s macro economic contribution:
  • Investing in AI-related R&D and creating coherent governance frameworks appear to strengthen tourism’s capacity to capture the economic benefits of AI (innovation, product/service quality, destination management).
  • Policy design should prioritize institutional and innovation-capacity levers, not only diffusion or headline AI adoption metrics.
• Human-capital investments show lagged returns for sectoral outcomes:
  • Talent development (skills, tertiary training) supports tourism’s gains from AI but effects take time to emerge — implying medium-term workforce policies rather than expecting immediate impacts.
• AI does not automatically increase resilience to extreme shocks (COVID example):
  • No evidence here that higher AI vibrancy buffered the pandemic-induced drop in tourism GDP share; resilience likely requires combined institutional, financial, and sectoral measures alongside digital/AI capabilities.
• For AI economics research:
  • Sectoral analyses should use disaggregated AI ecosystem indicators and allow for dynamic/lags and heterogeneity across countries.
  • More granular microdata (firm-level adoption, SME access to AI, consumer uptake) and causal identification strategies are needed to unpack mechanisms (productivity augmentation vs. labor substitution).
  • Consider distributional effects and digital-divide constraints: sectoral gains may concentrate among better-resourced firms/locations unless accompanied by inclusive policies.
• Policy takeaway for tourism stakeholders:
  • Targeted support for AI R&D, clearer governance/regulatory frameworks, and investments in relevant skills can raise tourism’s economic contribution more effectively than broad “AI promotion” alone.

Reference: Rahmanov et al. (2026), “Which dimensions of AI development shape tourism’s direct contribution to GDP? Evidence from a multi‑country panel,” Knowledge and Performance Management, 10(2):78–102 (DOI: 10.21511/kpm.10(2).2026.06).