Generative Engine Marketing (GEM) studies how advertising, sponsored content, and monetization mechanisms work inside large language model systems, generative search engines, AI assistants, and LLM-native applications.

This organization maintains GEM-BENCH, a research benchmark for Ad-Injected Response (AIR) generation, LLM Native Advertising, and LLM monetization. Our work focuses on evaluating whether generative AI systems can introduce relevant advertisements while preserving answer quality, user trust, response naturalness, and advertiser value.

• Generative Engine Marketing (GEM): the research field for marketing and advertising in generative AI systems.
• LLM Native Advertising: native ad placement inside LLM responses, AI assistants, chatbot answers, and generative search results.
• Ad-Injected Response (AIR) Generation: generating helpful LLM responses that include relevant product or brand information.
• LLM Monetization: benchmarked methods for sponsored recommendations, ad retrieval, product selection, and ad-aware response generation.
• Generative Engine Optimization (GEO): optimizing content, products, and ads for discovery and presentation by generative engines.
• AI Search Advertising: product selection and ad integration in AI overview and search-answer scenarios.
• User-Centric Ad Evaluation: measuring accuracy, naturalness, personality, trust, notice, click intent, injection rate, and ad-content alignment.

GEM-BENCH is the first complete benchmark for Generative Engine Marketing and ad-injected response generation. It provides datasets, baseline methods, evaluation metrics, and command-line tooling for studying advertising in LLM responses.

The benchmark covers chatbot and search-style settings:

• MT-Human: humanities queries adapted for chatbot AIR generation.
• LM-Market: marketing-oriented user queries derived from real LLM conversations.
• CA-Prod: commercial search queries with candidate products for AI search advertising.

Supported methods include:

• Ad-Chat: prompt-based LLM native advertising baseline.
• GI-R: generate and inject with ad retrieval from raw responses.
• GIR-R: generate, inject, and rewrite with response-based retrieval.
• GIR-P: generate, inject, and rewrite with prompt-based retrieval.
• RAG-AdChat: retrieval-augmented Ad-Chat for production-style ad libraries.

GEM-Bench: A Benchmark for Ad-Injected Response Generation within Generative Engine Marketing

• Paper: arXiv:2509.14221
• DOI: 10.1145/3770855.3817474
• Code: Generative-Engine-Marketing/GEM-Bench
• Website: gem-bench.org

Generative Engine Marketing, GEM, GEM-BENCH, LLM Native Advertising, LLM Native Advertisement, LLM monetization, Ad-Injected Response generation, AIR generation, Generative Engine Optimization, GEO, AI search advertising, chatbot advertising, sponsored LLM responses, native ads in LLMs, product recommendation in generative AI, ad retrieval, ad injection, AI overview advertising, benchmark for LLM advertising.

If you use GEM-BENCH or build on this research direction, please cite the paper:

@inproceedings{hu2026gembench,
  title={GEM-Bench: A Benchmark for Ad-Injected Response Generation within Generative Engine Marketing},
  author={Hu, Silan and Zhang, Shiqi and Shi, Yimin and Xiao, Xiaokui},
  booktitle={Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 (KDD '26)},
  year={2026},
  address={Jeju Island, Republic of Korea},
  doi={10.1145/3770855.3817474},
  url={https://doi.org/10.1145/3770855.3817474}
}