mlsanityData to 7 Oct 2026

How Do LLMs Cite? A Mechanistic Interpretation of Attribution in Retrieval-Augmented Generation

arXiv 2606.28358 ↗First posted 9 Jun 2026cs.IR cs.AI cs.CLcited by 0

On the map: Retrieval-augmented generation → RAG robustness and adaptive retrieval

Abstract (arXiv)

Retrieval-Augmented Generation (RAG) aims to enhance the trustworthiness of Large Language Models (LLMs) by grounding their outputs in external documents, often using inline citations for verifiability. However, the faithfulness of these citations -- whether the model genuinely uses a source to generate an answer -- remains a critical, unverified assumption. This paper offers the first mechanistic account of how a large language model decides whether to attach an inline citation while answering a factoid question. Using the Llama-3.1-8B-Instruct model in a controlled experimental environment based on the PopQA dataset, we employ an activation patching approach. We map the underlying mechanism responsible for citation, discovering that it is not a single, localized component but a distributed, multi-stage "attributional ensemble" of attention heads and MLP layers. We show that amplifying or attenuating only those critical heads and MLPs repairs over 90% of missed citations and eliminates 69% of spurious ones on PopQA without harming answer accuracy. Although gains on the multi-document HotpotQA benchmark are modest, the same component set still moves citation rates in the intended direction, indicating that the underlying mechanism is not dataset-specific. The results reveal a potential disconnect between the model's apparent reasoning and its internal computational pathway, suggesting that inline citations can create a false sense of security.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in RAG robustness and adaptive retrieval, one of 15 directions in Retrieval-augmented generation. Builds on 19 of its 31 references that are in the map. First posted on arXiv 9 Jun 2026.

Builds on

19 of 31 references.

Closest ancestors · 1

Same direction, newest first

  1. 2024Model Internals-based Answer Attribution for Trustworthy Retrieval-Augmented Generation19 Jun

Other references · 18

  1. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  2. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,686
  3. 2023Retrieval-Augmented Generation for Large Language Models: A Surveycited by 2,185
  4. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,130
  5. 2022When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memoriescited by 1,103
  6. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,081
  7. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 590
  8. 2023Enabling Large Language Models to Generate Text with Citationscited by 584
  9. 2025Reasoning Models Don't Always Say What They Thinkcited by 388
  10. 2023Evaluating Verifiability in Generative Search Enginescited by 297
  11. 2024A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Modelscited by 266
  12. 2024Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Toolscited by 170
6 more
  1. 2022Attributed Question Answering: Evaluation and Modeling for Attributed Large Language Modelscited by 108
  2. 2024Are self-explanations from Large Language Models faithful?cited by 101
  3. 2024ContextCite: Attributing Model Generation to Contextcited by 93
  4. 2017Explainable AI: Beware of Inmates Running the Asylum Or: How I Learnt to Stop Worrying and Love the Social and Behavioural Sciencescited by 84
  5. 2023Creating Trustworthy LLMs: Dealing with Hallucinations in Healthcare AIcited by 33
  6. 2024Peering into the Mind of Language Models: An Approach for Attribution in Contextual Question Answeringcited by 25

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Concise and Sufficient Sub-Sentence Citations for Retrieval-Augmented Generation
  2. 2025VeriCite: Towards Reliable Citations in Retrieval-Augmented Generation via Rigorous Verification
  3. 2025Transparentize the Internal and External Knowledge Utilization in LLMs with Trustworthy Citation
  4. 2024Ground Every Sentence: Improving Retrieval-Augmented LLMs with Interleaved Reference-Claim Generation
  5. 2024On the Capacity of Citation Generation by Large Language Models
  6. 2024Evaluation of Attribution Bias in Generator-Aware Retrieval-Augmented Large Language Models
  7. 2024Attribute or Abstain: Large Language Models as Long Document Assistants
  8. 2024ClashEval: Quantifying the tug-of-war between an LLM's internal prior and external evidence