Academic research on AI for tax compliance has surged since 2020, dominated by machine-learning fraud detection but rapidly pivoting toward explainability and governance; the authors synthesize 527 Scopus records into a practical five-step implementation framework spanning risk design, data integration, model choice, governance, and continuous monitoring.

Main Finding

A bibliometric and science-mapping review of 527 records (2004–2026) shows that research on AI and tax compliance has accelerated exponentially since 2020, driven by advances in machine learning, big data analytics, and digital tax administration. Core technical themes are AI/ML-based fraud detection and data mining; emergent governance topics (explainable AI, transparency, trust, ethics) are becoming central. The literature supports a practical five-step AI-for-tax-compliance framework: (1) risk concept formulation, (2) data integration, (3) model selection, (4) governance and explainability, and (5) continuous monitoring.

Key Points

• Scope and scale: 527 Scopus records published between 2004 and 2026 were analyzed using PRISMA-guided selection.
• Tools and mapping: Biblioshiny and VOSviewer were used for bibliometric performance mapping and science mapping (co-citation, bibliographic coupling, keyword co-occurrence, thematic evolution).
• Growth pattern: Exponential publication growth since 2020, reflecting technical advances and increased digitalization of tax systems.
• Geographic leadership: China, Germany, and the United States are the most prolific contributors.
• Dominant research themes: taxation, artificial intelligence, machine learning, fraud detection, and data mining.
• Emerging/priority themes: explainable AI (XAI), governance frameworks, transparency, public trust, ethical AI adoption, and data governance.
• Practical output: A proposed five-step implementation framework for AI in tax compliance (risk formulation → data integration → model selection → governance & explainability → continuous monitoring).
• Novelty of the study: Combined performance mapping with science mapping to identify emerging themes and chart future research directions.

Data & Methods

• Literature retrieval: Structured search strategy applied to Scopus, with PRISMA methodology for screening and selection.
• Dataset: 527 published records spanning 2004–2026.
• Analysis software: Biblioshiny (R Bibliometrix) for bibliometric performance indicators and thematic evolution; VOSviewer for network visualization (co-authorship, co-citation, keyword co-occurrence, bibliographic coupling).
• Analyses performed:
  • Publication trends and growth rates over time.
  • Performance mapping: prolific authors, journals, countries, and citation counts.
  • Science mapping: co-citation networks, bibliographic coupling clusters, keyword co-occurrence maps, and thematic evolution over periods.
  • Identification of emergent themes and synthesis into an implementation framework.

Implications for AI Economics

• Fiscal capacity and efficiency: AI-driven detection and risk-assessment tools can materially reduce the tax gap and increase revenue collection efficiency, changing fiscal capacity models and short-term revenue forecasts.
• Cost–benefit and investment decisions: Economic evaluation should account for development/deployment costs, expected revenue gains, administrative savings, and ongoing monitoring/compliance costs; heterogeneous returns across country contexts imply prioritization decisions.
• Incentives and behavioral responses: AI enforcement changes auditing probabilities and detection technologies—economic models must incorporate strategic taxpayer responses (e.g., tax avoidance innovation, privacy-driven behavior), and potential shifts in compliance equilibria.
• Labor market effects: Automation of audit, risk-scoring, and processing tasks will reconfigure public-sector labor demand, shifting toward data-science and governance roles; transition costs and retraining needs are economic policy considerations.
• Distributional and equity impacts: Differential effects on taxpayer types (firms vs. individuals, large vs. small businesses, informal sector) can alter effective tax incidence and require equity-focused policy design.
• Explainability and trust as economic inputs: Explainable AI and transparent governance increase voluntary compliance by affecting perceived fairness and legitimacy; trust is an economic multiplier for compliance that must be factored into deployment strategies.
• Regulation and enforcement externalities: Data-sharing, privacy constraints, and cross-border information flows shape the feasible set of AI tools and create coordination problems across jurisdictions; regulatory design affects adoption speed and efficiency.
• Research and evaluation priorities for economists:
  • Causal impact studies (RCTs, difference-in-differences) on AI tools’ effect on compliance, revenue, and taxpayer behavior.
  • Cost-effectiveness and welfare analyses comparing AI approaches with conventional enforcement.
  • Game-theoretic and dynamic models of strategic taxpayer response to AI-enabled enforcement.
  • Distributional assessments and policy designs to mitigate regressive enforcement outcomes.
  • Institutional analyses of governance, accountability, and procurement that affect implementation quality.
• Policy guidance: Economic policy should couple technical deployment with governance safeguards (XAI, transparency, auditability), phased roll-outs with evaluation, and investments in organizational capacity to realize efficiency gains without undermining trust or equity.