VeriAd: A CRAG-Orchestrated Multimodal Compliance Auditing Framework for Short-Form Advertisement Videos with Temporal Evidence Grounding and Cross-Modal Contradiction Detection

In this paper we introduced VeriAd, a multimodal framework for auditing short-form advertisement videos for compliance. Rather than simply sending unstructured text to a language model, VeriAd creates a timeline of evidence based on the timestamped information. Rules are then graded using a jurisdiction aware CRAG loop. Each disclaimer is checked to ensure they are sufficiently similar, lengthy, legible and consistent to qualify the claim. LangGraph ensures we can track the multi-stage process. LangSmith means we can trace all decisions made and we provide a three-way verdict to send any uncertainty to a human reviewer. The whole framework uses only managed services and no model training meaning it is feasible for small teams and agencies. Future work will include completing the annotated data set and providing the results of our ablation study shown in table 7. New languages will be added for regional languages within India. We will also expand our visual analysis to include before and after images and tampered media. An automated pipeline will be created to notify maintainers if the source regulation changes. Using feedback from a reviewer to automatically adjust the thresholds is also a promising area of work.

1.  Y. Chen, Z. Hu, M. Xue, T. Cao, L. Liu, P. Shu, H. Yu, J. Jiang, "R³: Advertisement Compliance Rectification via Group-Relative Experience Extractor and Curriculum Reinforcement", Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL), Industry Track, 2026. https://arxiv.org/abs/2607.07318 2.  Y. Yang, Z. Liu, Y. Yuan, Y. Song, X. Ma, Y. Song, X. Zeng, L. Sun, Y. Wang, H. Zhou, S. Cui, Z. Gong, J. Zhang, "BLM-Guard: Explainable Multimodal Ad Moderation with Chain-of-Thought and Policy-Aligned Rewards", arXiv preprint arXiv:2602.18193, 2026. https://arxiv.org/abs/2602.18193 3.  Microsoft, "Azure AI Video Indexer Insights Overview", Microsoft Learn Documentation, 2026. https://learn.microsoft.com/en-us/azure/azure-video-indexer/insights-overview 4.  Microsoft, "Get Optical Character Recognition (OCR) Insights", Microsoft Learn Documentation, 2026. https://learn.microsoft.com/en-us/azure/azure-video-indexer/ocr-insight 5.  Microsoft, "Transparency Note for Azure AI Video Indexer", Microsoft Learn Documentation, 2026. https://learn.microsoft.com/en-us/legal/azure-video-indexer/transparency-note 6.  LangChain, "LangGraph Documentation", LangChain Inc., 2026. https://langchain-ai.github.io/langgraph/ 7.  LangChain, "LangSmith Documentation", LangChain Inc., 2026. https://docs.smith.langchain.com/ 8.  Microsoft, "Vector Search in Azure AI Search", Microsoft Learn Documentation, 2026. https://learn.microsoft.com/en-us/azure/search/vector-search-overview 9.  GitHub user raghavbankar, "Video Compliance System: Automated Video Compliance QA Using Agentic RAG and LLMs", GitHub Repository, accessed October 2026. https://github.com/raghavbankar/Video_Compliance_Analyzer 10.  D. Ji, Y. Yang, L. Liu, P. Shu, H. Wu, S. Tang, X. Chen, S. Ma, T. Chen, L. Zhu, "RAVEN++: Pinpointing Fine-Grained Violations in Advertisement Videos with Active Reinforcement Reasoning", Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP), Industry Track, 2025. https://arxiv.org/abs/2511.19168 11.  D. Ji, Y. Yang, H. Wu, S. Ma, T. Chen, L. Zhu, "RAVEN: Robust Advertisement Video Violation Temporal Grounding via Reinforcement Reasoning", Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL), Industry Track, 2025. https://arxiv.org/abs/2510.16455