# AI Word Explorer

> AI Word Explorer (aiwordexplorer.com, formerly LexA-Index) is an interactive tool for exploring which words large language models systematically overuse compared to human writing, across 34 languages, multiple registers, and several AI models. It accompanies the research paper *AI-Associated Lexical Shifts Across 34 Languages* by Thomas Stephan Juzek (Florida State University; https://tjuzek.com).

## Key findings

- Chat-aligned language models show consistent lexical preferences across **34 languages** (WMT News Crawl corpus), measured by comparing GPT-4.1 mini continuations against human text using log prevalence ratios.
- AI-overused vocabulary rose in real news writing after ChatGPT's release: prevalence increased in **26 of 34 languages** (mean change **+15.1%**), while matched baseline words did not (**−4.5%**).
- Substantial cross-lingual semantic convergence: "emphasize"-type verbs are overused in **24 of 34 languages**.
- Diachronic comparison (2020–2021 vs. 2023–2024) and longitudinal analysis of 10 languages show post-2022 increases exceeding earlier periods, though smaller than effects previously found in Scientific English.
- Colloquially, these AI-overused words are known as "delve" words, AI slop, ghost words, AI buzzwords, Modelese, or Algospeak.

## Paper

- [AI-Associated Lexical Shifts Across 34 Languages (arXiv:2605.25358)](https://arxiv.org/abs/2605.25358): the main study behind the Explorer. Author: Thomas Stephan Juzek, FSU. Subjects: cs.CL, cs.AI, cs.CY.

## Explore

- [AI Word Overuse Explorer](https://www.aiwordexplorer.com/): interactive site to inspect AI-overused words by language, register, and model.
- [Words AI overuses, by language](https://www.aiwordexplorer.com/words.html): a static, readable list of the top 20 AI-overused words per language (ranked by Log Prevalence Ratio), across 34 languages.
- [About & Methodology](https://www.aiwordexplorer.com/about.html): windowed prevalence method, Lexical Alignment Score (LAS), related papers, and citation details.

## Code & data

- [GitHub repository](https://github.com/fsu-nlp/lexa-index): website source, CSV outputs of the 34-language word lists, and the data-build script.

## Citation

```bibtex
@misc{juzek2026lexical,
  title         = {AI-Associated Lexical Shifts Across 34 Languages: Cross-Lingual Convergence and Diachronic Uptake in News Writing},
  author        = {Juzek, Thomas Stephan},
  year          = {2026},
  eprint        = {2605.25358},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2605.25358}
}
```