Most editorial prose on Discover is drafted with the assistance of generative AI, grounded in source material. We disclose this because AI-assisted editorial should be labelled, not concealed.

What is AI-drafted

• Artist biographies, grounded in source material specific to each artist and reviewed against our content guardrails before publication.
• Movement editorials, grounded in scholarly art-history sources.
• FAQ answers, with a portion matched back to specific scholarly sources cited in the page footer.
• Where-to-see bio hooks: short summary paragraphs only. The structured visit data (museum lists, addresses, opening hours) is sourced from museum and reference data, not generated.
• Long-tail museum summaries. Top museums are hand-curated.

What is not AI

• Citation placement. The ^[N] superscripts beside specific claims are placed deterministically.
• Source records themselves. The pool of sources we cite is fetched from external reference systems at ingestion time. AI does not invent sources.
• Structured facts such as birth and death dates, nationality, movement tags, and museum holdings.
• Visit information such as opening hours, addresses, and ticket prices.
• Images. Every image is Creative Commons or public domain. AI does not generate or alter images on Discover.

Content guardrails

Before any AI-drafted content reaches a published page, it passes through a guardrail layer. The layer rejects content involving figures and symbols associated with hateful ideologies, and routes other historically sensitive material to an editorial review queue instead of auto-publishing.

These guardrails catch the worst failure modes; they are not a substitute for editorial review by domain experts.