TOML
Minimal, human-friendly config.
TOML is an explicitly-typed, line-oriented config format whose grammar is small enough to be parsed unambiguously by every major language, which is why it has become the default for tooling configs (Cargo, pyproject.toml, Hugo). For anything that has to read or rewrite human-edited metadata without a schema in hand, TOML's typed values remove the YAML-style guesswork: no date silently becoming a string, no "no" silently becoming false.
Geordy generates this format for every page on your site that warrants it - auto-detected by page type, served from your subdomain, and refreshed on the cadence below.
# page.toml for https://geordy.ai/features title = "Features" description = "The structured files AI systems actually read." url = "https://geordy.ai/features" updated = 2026-10-20 [owner] name = "Geordy" url = "https://geordy.ai"
How to make this format work for AI.
- Quote all string values; use bare keys for identifiers (`name = "foo"`) and reserve quoted keys for those containing dots or special characters.
- Prefer `[table]` headers and `[[array.of.tables]]` over deeply nested inline tables for readability and clean diffs.
- Use offset date-time literals (RFC 3339, e.g. `1979-05-27T07:32:00Z`) rather than strings so consumers get real timestamps.
- Pin to the spec version your parser supports (TOML 1.0.0 is current stable) and run files through a validator like `taplo` in CI.
- Re-defining the same key or table twice in one file - TOML 1.0 makes this an explicit error, not a silent override.
- Mixing types inside an array in TOML 1.0 parsers that haven't adopted heterogeneous arrays.
- Putting a sub-table after later keys in its parent table - once `[a.b]` is opened you can no longer add bare keys to `[a]` below it.
Use cases, scored for AI visibility.
Real sites already shipping TOML.
Tools & resources for TOML.
- Human readability with explicit types
- Fast parsing
- Adopted by major tools (Cargo.toml, pyproject.toml)
- Reduced flexibility for complex nested structures versus YAML
No AI engine asks for page.toml. It is here for a different job: typed, unambiguous page metadata your own pipelines can read, where a date is a date and nothing has to be guessed. Cargo, pyproject.toml, Hugo and Poetry all settled on TOML for that reason, and every mainstream language has a parser, so nothing downstream has to hand-roll one.
Written for these readers. Which of them actually request it is measured, not assumed: see how we measure.
- TOML: the project siteA minimal configuration file format that is easy to read because its semantics are obvious.
Not measured yet. Geordy will publish which crawlers request this file once enough independent sites are covered. See how we measure.