ai.txt is a Spawning-proposed text file at the site root that declares AI training and citation consent in robots.txt syntax.
It's distinct from llms.txt: llms.txt tells engines what's worth indexing, ai.txt tells them whether they're allowed to use it for training in the first place.
Syntax
Same User-Agent / Allow / Disallow shape as robots.txt:
User-Agent: *
Allow: /
Disallow: /private/
Allow: / means "AI training is permitted." Disallow: / blocks training entirely. Per-path rules work the same way — common pattern is to allow public content and disallow authenticated dashboard surfaces.
When to opt in
If your business model depends on AI citations (most B2B SaaS, most editorial sites), opting in is the obvious move. Refusing AI training is mainly relevant for creative-portfolio sites — Spawning grew out of the visual-artist compensation conversation.
Does anyone honor it
OpenAI, Anthropic, and Perplexity say they respect both ai.txt and robots.txt's explicit AI-crawler User-Agent rules. The honor system isn't bulletproof, but compliance is the norm — and shipping the file is the cheapest way to make your stance machine-readable. BCited's scanner checks for it.