# Answer Bubbles: Information Exposure in AI-Mediated Search

arXiv: https://arxiv.org/abs/2603.16138 · HTML: https://mlsanity.com/paper/2603.16138/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 53 references that are in the map.
- First posted on arXiv 17 Mar 2026.

## Builds on (most cited first, 21 of 21)

- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [BERTopic: Neural topic modeling with a class-based TF-IDF procedure](https://mlsanity.com/paper/2203.05794/) (2022)
- [FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation](https://mlsanity.com/paper/2310.03214/) (2023)
- [Evaluating Verifiability in Generative Search Engines](https://mlsanity.com/paper/2304.09848/) (2023)
- [Dear Sir or Madam, May I introduce the GYAFC Dataset: Corpus, Benchmarks and Metrics for Formality Style Transfer](https://mlsanity.com/paper/1803.06535/) (2018)
- [ZeroSearch: Incentivize the Search Capability of LLMs without Searching](https://mlsanity.com/paper/2505.04588/) (2025)
- [Bias and Unfairness in Information Retrieval Systems: New Challenges in the LLM Era](https://mlsanity.com/paper/2404.11457/) (2024)
- [Neural Retrievers are Biased Towards LLM-Generated Content](https://mlsanity.com/paper/2310.20501/) (2023)
- [Don't lose the message while paraphrasing: A study on content preserving style transfer](https://mlsanity.com/paper/2308.09055/) (2023)
- [TyDiP: A Dataset for Politeness Classification in Nine Typologically Diverse Languages](https://mlsanity.com/paper/2211.16496/) (2022)
- [Collapse of Dense Retrievers: Short, Early, and Literal Biases Outranking Factual Evidence](https://mlsanity.com/paper/2503.05037/) (2025)
- [Dwell in the Beginning: How Language Models Embed Long Documents for Dense Retrieval](https://mlsanity.com/paper/2404.04163/) (2024)
- [Generative AI Search Engines as Arbiters of Public Knowledge: An Audit of Bias and Authority](https://mlsanity.com/paper/2405.14034/) (2024)
- [MoltNet: Understanding Social Behavior of AI Agents in the Agent-Native MoltBook](https://mlsanity.com/paper/2602.13458/) (2026)
- [Auditing Google's AI Overviews and Featured Snippets: A Case Study on Baby Care and Pregnancy](https://mlsanity.com/paper/2511.12920/) (2025)
- [Do We Know What They Know We Know? Calibrating Student Trust in AI and Human Responses Through Mutual Theory of Mind](https://mlsanity.com/paper/2601.16960/) (2026)
- [The Language of Approval: Identifying the Drivers of Positive Feedback Online](https://mlsanity.com/paper/2509.10370/) (2025)
- [The Hidden Toll of Social Media News: Causal Effects on Psychosocial Wellbeing](https://mlsanity.com/paper/2601.13487/) (2026)
- [Does Reasoning Make Search More Fair? Comparing Fairness in Reasoning and Non-Reasoning Rerankers](https://mlsanity.com/paper/2603.10332/) (2026)
- [Transparency, Privacy, and Fairness in Recommender Systems](https://mlsanity.com/paper/2406.11323/) (2024)
- [PromptDSI: Prompt-based Rehearsal-free Continual Learning for Document Retrieval](https://mlsanity.com/paper/2406.12593/) (2024)

## Built on by (most cited first, 3 of 3)

- [Inform, Coach, Relate, Listen: Auditing LLM Caregiving Support Roles](https://mlsanity.com/paper/2605.29473/) (2026)
- [How People Evaluate AI-, Expert-, and Peer-Style Financial Advice](https://mlsanity.com/paper/2608.09019/) (2026)
- [Masking or Mitigating? Deconstructing the Impact of Query Rewriting on Retriever Biases in RAG](https://mlsanity.com/paper/2604.06097/) (2026)

## Abstract (arXiv)

Generative search systems are increasingly replacing link-based retrieval with AI-generated summaries, yet little is known about how these systems differ in sources, language, and fidelity to cited material. We examine responses to 11,000 real search queries across five systems---vanilla GPT, Search GPT, Perplexity Search with Grok, Google AI Overviews, and traditional Google Search---at three levels: source diversity, linguistic characterization of the generated summary, and source-summary fidelity. We find that generative search systems exhibit significant textitsource-selection biases in their citations, favoring certain sources over others. Incorporating search also selectively attenuates epistemic markers, reducing hedging by up to 60% while preserving confidence language in the AI-generated summaries. At the same time, AI summaries further compound the citation biases: Wikipedia and longer sources are disproportionately overrepresented, whereas cited social media content and negatively framed sources are substantially underrepresented. Our findings highlight the potential for textitanswer bubbles, in which identical queries yield structurally different information realities across systems, with implications for user trust, source visibility, and the transparency of AI-mediated information access.