How Subtext works
Subtext helps Australian parents make informed reading choices. It is not censorship, not a safety guarantee, and not a substitute for reading the book yourself or talking with your child.
What we analyze
- Public book metadata from Google Books and Open Library (title, author, publisher description, categories).
- Optional enrichment from allowlisted review sites (Common Sense Media, Kirkus, The StoryGraph) in Deep mode only.
- We do not read the full text of the book.
Result states explained
Subtext uses four honest states. We never show a "Comfort Read" or cozy all-clear badge.
- Content warnings found
warnings— We found themes that may matter for your family. Review each advisory below.- No warnings in our taxonomy
clear_confident— We completed analysis on adequate metadata and, with high confidence, found no warnings in our taxonomy. This is not a guarantee the book is risk-free.- Not enough to be sure
low_confidence— We ran analysis but available metadata was thin or evidence was weak. Do not treat this as an all-clear.- Could not analyze
could_not_analyze— Analysis did not complete successfully. We cannot provide guidance for this book right now. Nothing here means the book is clear.
Quick vs Deep
Quick uses publisher metadata only. It is faster but relies on thinner evidence — a missing or vague description may produce a "Not enough to be sure" result rather than an all-clear.
Deep adds allowlisted third-party review snippets. It is slower and richer, but still metadata-based — not a full-text analysis.
Limitations
- Metadata may be incomplete, outdated, or vague — especially for new or niche titles.
- Severity is AU-style guidance inspired by classification language, not legal ACB equivalence.
- Errors, rate limits, and timeouts produce "Could not analyze" — never a false all-clear.
- AI analysis is constrained to a fixed taxonomy and evidence from the provided text only.