Academic attention to AI in tax compliance has surged since 2019 but remains fragmented: studies concentrate on enforcement effectiveness while legitimacy and sustainability are underdeveloped, and most published research originates from emerging-economy universities rather than OECD tax administrations.

Main Finding

AI in tax compliance is an emergent but rapidly growing interdisciplinary field (2015–2025) characterized by three overlapping thematic waves—enforcement (anomaly detection/audit targeting), governance/legitimacy (algorithmic transparency, trust, contestability), and sustainability (digitalization tied to broader fiscal and environmental goals). The literature is accelerating but not yet mature: AI improves detection/effectiveness, yet algorithmic opacity and weak governance risk undermining taxpayer legitimacy. The paper introduces the AI Tax Compliance Governance Matrix (ATCGM) — an integrative framing that links effectiveness, legitimacy, and sustainability — and documents a surprising geographic and disciplinary skew (research intensity concentrated in several non‑OECD universities and finance journals).

Key Points

• Corpus and growth
  • Analytical corpus: 1,803 Scopus-indexed, peer‑reviewed (Q1–Q2) articles (2015–2025).
  • Annual growth rate: ~36.0% per year; average document age 2.35 years.
  • International co-authorship: 36.77%.
• Thematic trajectory
  • Three compressed waves in a single decade: enforcement → governance/legitimacy → sustainability.
  • Central tension: efficacy of AI (better anomaly detection and targeting) vs. legitimacy risks (opacity, contestability, data privacy).
• Geographic and institutional patterns
  • Geographic inversion: high productivity from Sumy State University (Ukraine) and Universiti Teknologi MARA (Malaysia), and notable activity from Jordan, Romania, Azerbaijan—rather than traditional OECD administrative centers.
  • Interpretation: research intensity driven more by reform urgency and opportunity‑seizing than by administrative maturity.
• Disciplinary and network structure
  • Top publishing outlets skew toward finance and applied risk journals (e.g., Finance Research Letters; International Review of Economics and Finance).
  • Public administration journals are under‑represented.
  • Co-authorship networks form relatively closed epistemic clusters with little cross-cluster brokerage; Anglo‑American research centers are notably absent from the core.
• Empirical signals and case evidence
  • Operational deployments show measurable compliance effects (example: Peru’s e‑invoicing rollout reportedly raised declared VAT liabilities by >5% in year one), but governance conditions behind such gains are under‑researched.
• Contribution
  • Presents ATCGM (effectiveness, legitimacy, sustainability) as an organizing analytical matrix.
  • Identifies five empirically grounded research gaps (paper highlights governance and legitimacy research gaps tied to observed effectiveness gains; full list in original article).

Data & Methods

• Data source and search
  • Single-source query on Scopus (July 2025) using a comprehensive Boolean string combining AI terms (e.g., machine learning, NLP, anomaly detection, generative AI) and tax compliance terms (e.g., tax evasion, VAT fraud, tax audit, digital taxation).
  • Initial returns: 7,059 documents.
• PRISMA-style filtering (five phases)
• Temporal boundary: 2015–2025.
• Subject-area filter: retained Economics, Business, and Social Sciences (reduced to 3,679).
• Document type: peer‑reviewed articles only (3,378).
• Language: English (3,299).
• Quality filter: Scimago Q1–Q2 journals → final corpus 1,803 articles.
• Tools and analyses
  • Biblioshiny (v4.1.3) for descriptive bibliometrics and trend metrics.
  • VOSviewer (v1.6.20) for network visualizations: co-authorship (threshold ≥3 publications per author) and keyword co-occurrence (threshold ≥12 occurrences; 91 keywords).
  • Thematic evolution inferred from temporal overlay visualizations and cluster interpretation.
• Limitations noted by authors
  • Scopus-only and English-language restriction (may omit regionally important or non‑English scholarship).
  • Q1–Q2 selection may exclude operationally relevant practitioner/regional journals.
  • Citation-lag bias (2024–25 items under-cited); rapid AI evolution may outpace peer-reviewed literature.

Implications for AI Economics

• Modeling and measurement
  • AI changes how tax gaps and compliance are measured (real‑time reporting, risk scoring). Economists should incorporate algorithmic accuracy, selection effects, and measurement bias when estimating behavioral responses and revenue impacts.
• Policy design and legitimacy
  • Technical effectiveness is necessary but not sufficient; governance structures (transparency, explainability, contestability, privacy safeguards) are central to sustaining voluntary compliance. Economic models of enforcement should include legitimacy and trust as endogenous factors influencing behavioral responses.
• Distributional and institutional considerations
  • The geographic inversion suggests heterogeneous adoption paths: emerging economies may produce more scholarly work tied to reform urgency; OECD administrative practices may be operational but under‑documented academically. Comparative studies should account for institutional capacity, legal frameworks, and political economy drivers of AI adoption.
• Research agenda priorities (derived from paper)
  • Empirically test the ATCGM: quantify trade‑offs and complementarities between effectiveness, legitimacy, and sustainability in deployed systems.
  • Evaluate governance conditions behind observed revenue/compliance gains (e.g., e‑invoicing and VAT changes) to separate technical from institutional drivers.
  • Study algorithmic transparency and appeal mechanisms in tax contexts (legal explainability, contestability), and how they shape taxpayer behavior and administrative costs.
  • Integrate environmental/sustainability outcomes where taxation digitalization intersects with green fiscal policy.
  • Broaden data sources and interdisciplinary engagement: include public administration scholarship, practitioner reports, non‑English outputs, and administrative data to reduce selection bias.
• Practical implication for policymakers and economists
  • When assessing AI interventions in taxation, evaluate both direct revenue/effectiveness metrics and governance/legitimacy inputs; investments in explainability, oversight, and public communication can materially affect behavioral responses and long‑term revenue durability.

If you want, I can: - Extract the five specific research gaps enumerated in the paper (full text permitting), or - Produce a short list of empirical study designs that would operationalize the ATCGM (e.g., difference‑in‑differences evaluations, randomized audits, survey experiments on legitimacy).