Unintended Effects of Geographic Conditioning in Large Language Models

• URL: http://arxiv.org/abs/2606.18124v1
• Date: Tue, 16 Jun 2026 16:23:36 GMT
• Title: Unintended Effects of Geographic Conditioning in Large Language Models
• Abstract summary: We evaluate location leakage: the phenomenon where a model generates geographic references despite receiving a geographically neutral user prompt.<n>Across both creative writing and open-ended Q&A prompts, even state-of-the-art LLMs systematically favor region-specific outputs when exposed to location metadata.<n>Our analysis further shows a novel structural conditioning effect: replacing the injected location with the placeholder "Unknown" still elevates leakage by up to 72 times above baseline.
• Score: 11.644241437138339
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Modern conversational AI systems frequently rely on user metadata to localize responses, yet the unintended regional biases introduced by this hidden context remain poorly understood. In this work, we evaluate location leakage: the phenomenon where a model generates geographic references despite receiving a geographically neutral user prompt. Across both creative writing and open-ended Q&A prompts, even state-of-the-art LLMs systematically favor region-specific outputs when exposed to location metadata, with leakage spiking by up to 793 times above baseline (e.g., from 0.04% to 31.7% for Llama 3.1-8B, and 21.3% and 8.8% for Qwen3-8B and Claude Sonnet 4.6, respectively). Our analysis further shows a novel structural conditioning effect: replacing the injected location with the placeholder "Unknown" still elevates leakage by up to 72 times above baseline, demonstrating that the user profile frame itself, independent of any geographic content, acts as a generative conditioning signal.

Related papers

• Evaluating the Hidden Costs of Personalization in Large Language Models [55.07935163900691]
  We identify three emerging risks: (1) irrelevant personalization, where models reference personal information in unnecessary contexts; (2) preference narrowing, where models reinforce informational echo chambers; and (3) sycophantic bias, where models agree excessively with user opinions.<n>We propose PRISK, a dynamic evaluation framework with automated data generation and tailored metrics that uncovers systematic limitations in current AI assistants.
  arXiv  Detail & Related papers  (2026-08-28T20:05:12Z)
• Terrain signatures in Welsh settlement names [0.0]
  We analysed 3,757 Welsh settlements using a frozen, source-audited 24-element lexical framework.<n>High-terrain names occupied locations 24.4 m higher relative to their 2-km surroundings.<n>Adding terrain-name polarity to a non-lexical spatial and settlement baseline reduced geographically held-out mean squared error by 4.63%.
  arXiv  Detail & Related papers  (2026-08-27T11:24:11Z)
• Region Matters: Efficient and Reliable Region-Aware Visual Place Recognition [46.97190804063947]
  We introduce FoL++, a method combining robust discriminative region modeling with adaptive re-ranking.<n>FoL++ achieves state-of-the-art performance with a lightweight memory footprint, improving inference speed by 40% over FoL.
  arXiv  Detail & Related papers  (2026-04-24T09:28:35Z)
• From Black Box to Glass Box: Cross-Model ASR Disagreement to Prioto Review in Ambient AI Scribe Documentation [43.148402136307716]
  Cross-model disagreement among heterogeneous ASR systems can act as a reference-free uncertainty signal.<n>We transcribed 50 publicly available medical audio clips with eight ASR systems spanning commercial APIs and open-source engines.<n>Low-agreement regions were enriched for content disagreements, with the content fraction increasing from 53.9% to 73.9% across quintiles of high-risk mass.
  arXiv  Detail & Related papers  (2026-03-02T13:02:13Z)
• LocalBench: Benchmarking LLMs on County-Level Local Knowledge and Reasoning [9.319308493696893]
  Large language models (LLMs) have been widely evaluated on macro-scale geographic tasks, but their ability to handle hyper-local knowledge remains poorly understood.<n>We present LocalBench, the first benchmark designed to evaluate LLMs on county-level local knowledge across the United States.<n>Using LocalBench, we evaluate 13 state-of-the-art LLMs under both closed-book and web-augmented settings.
  arXiv  Detail & Related papers  (2025-11-13T16:26:13Z)
• Where on Earth? A Vision-Language Benchmark for Probing Model Geolocation Skills Across Scales [61.03549470159347]
  Vision-language models (VLMs) have advanced rapidly, yet their capacity for image-grounded geolocation in open-world conditions has not been comprehensively evaluated.<n>We present EarthWhere, a comprehensive benchmark for VLM image geolocation that evaluates visual recognition, step-by-step reasoning, and evidence use.
  arXiv  Detail & Related papers  (2025-10-13T01:12:21Z)
• Around the World in 24 Hours: Probing LLM Knowledge of Time and Place [18.17538075862074]
  We present the first evaluation of the ability of language models to jointly reason over time and space.<n>We evaluate eight open chat models of three different model families for different combinations of temporal and geographic knowledge.<n>We do not find clear correlations of performance with specific geographic regions.
  arXiv  Detail & Related papers  (2025-06-04T14:14:28Z)
• Distortions in Judged Spatial Relations in Large Language Models [45.875801135769585]
  GPT-4 exhibited superior performance with 55 percent accuracy, followed by GPT-3.5 at 47 percent, and Llama-2 at 45 percent. The models identified the nearest cardinal direction in most cases, reflecting their associative learning mechanism.
  arXiv  Detail & Related papers  (2024-01-08T20:08:04Z)
• Jalisco's multiclass land cover analysis and classification using a novel lightweight convnet with real-world multispectral and relief data [51.715517570634994]
  We present our novel lightweight (only 89k parameters) Convolution Neural Network (ConvNet) to make LC classification and analysis. In this work, we combine three real-world open data sources to obtain 13 channels. Our embedded analysis anticipates the limited performance in some classes and gives us the opportunity to group the most similar.
  arXiv  Detail & Related papers  (2022-01-26T14:58:51Z)
• Enhancing Prototypical Few-Shot Learning by Leveraging the Local-Level Strategy [75.63022284445945]
  We find that the existing works often build their few-shot model based on the image-level feature by mixing all local-level features. We present (a) a local-agnostic training strategy to avoid the discriminative location bias between the base and novel categories, and (b) a novel local-level similarity measure to capture the accurate comparison between local-level features.
  arXiv  Detail & Related papers  (2021-11-08T08:45:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.