In T379908, we learned popular LLMs (e.g. Gemini, Claude, and ChatGPT) in include unique HTML elements along with text people paste from these services.
Through this investigation, we also concluded (T379908#10419299) that the HTML these service generate is likely to evolve over time and thus, not reliable/stable enough to be used a signal to configure Paste Check with.
This ticket involves the work of revisiting the HTML these popular LLMs include in text people paste from them to:
- **Learn** if and how the HTML has changed since we first investigated this in November 2024
- **Decide** whether we think the HTML is stable enough to be used as a signal Paste Check can be configured with/off of
=== Related
- [Wish: Automatic updated list of newly created articles possibly generated by artificial intelligence](https://meta.wikimedia.org/wiki/Community_Wishlist/Wishes/Automatic_updated_list_of_newly_created_articles_possibly_generated_by_artificial_intelligence)
- [WETBench: A Benchmark for Detecting Task-Specific Machine-Generated Text on Wikipedia](https://arxiv.org/html/2507.03373v1)