Turning the page…

Last updated: 2026-06-22

Every Fables story is generated by AI, which means a person isn't reading each page before your child does. That worries a lot of parents. It should. So we built the review your child's story passes through, and we'd rather show you the machinery than ask you to take "it's safe" on faith.

This page is honest about one more thing: what is automated and what isn't. The checks below run on every story, every time. Human review by early-childhood educators is on our roadmap, not in place today, and we won't pretend otherwise (see "What a person does, and doesn't, review", below).

Our content policy

Fables makes warm, age-appropriate stories for young children. We do not allow:

• Content that targets, identifies, or could harm a real child other than your own.
• Hateful, sexually explicit, or graphic/extremely violent content.
• Personal information (emails, phone numbers, addresses, or ID-like numbers) anywhere on a child's page.

You own the stories you make, you can regenerate any page you don't love, and you can delete everything any time. The full rules live in our Terms and Privacy policy.

The safety screen: what runs before a story reaches your child

Before a finished story is shown to anyone, every page passes through a safety screen with two layers. It is fail-closed: if a check can't run, errors, or isn't sure, the page is held back, never shown by default.

1. A personal-information backstop (deterministic, runs first). Before any AI judgment, a plain pattern-matcher scans each page for emails, phone numbers, and long digit runs that look like addresses or IDs. If it finds any, that page is blocked immediately, independent of, and ahead of, the AI judge. This means even if the smarter judge had an outage, personal information still can't slip onto a child's page.

2. An age-appropriateness judge (AI, fail-closed). Each page is then judged for age-appropriateness against your child's reader profile. The judge returns safe, warn, or block, and anything it can't return cleanly is treated as block, never safe. If any page is blocked, the whole story is held back rather than shipped with a bad page in it.

What the quality gates enforce

Stories aren't only words. A page that reads as broken art breaks trust just as fast. A set of deterministic gates runs over the artwork in our test suite, on every published experience, so these guarantees can't quietly regress:

• Legibility (figure–ground contrast). Every character must stand apart from its background by value, not blend into it. A figure that collapses into its backdrop (a green creature on a green meadow) fails the check.
• What the words name, the picture shows. When the narration names a concrete creature ("an owl hooted hello"), that creature has to actually appear on the page. Words and picture have to agree.
• Pose matches the moment. If the words say a character is fast asleep, the art can't show them standing, walking, or waving. The clear contradictions are caught.
• The scene is finished. Every page must be lit, every character grounded with a contact shadow and shaded so it reads as solid, the backdrop must have depth, and no placeholder art is allowed to ship.

Words your early reader can actually read

For early-reader and controlled-vocabulary modes, a checker flags any word outside the decodable set (common sight words plus your story's own character and prop names) so the page stays readable for a child just starting out. This is something a fixed library of pre-written stories can't promise across a constantly growing supply of new ones; because Fables generates each story, it can check each one.

Unusual names and unusual premises

Children have names from every language, and the stories families dream up are wonderfully strange. We've designed the checks so they don't punish either:

• Names are not treated as suspicious. The personal-information backstop looks for emails, phone numbers, and ID-like digit runs, never ordinary names. An unusual first name is only a name; it sails through.
• In early-reader mode, your story's own names are always allowed. The decodable-word check adds every character and prop name in your story to the allowed set, so "Zephyrine the dragon" never trips it.
• Unusual premises still get judged on substance. A story about a lonely volcano or a worried little ghost is fine; the age-appropriateness judge reacts to what actually happens on the page, not to a premise sounding odd.

What a person does, and doesn't, review

We want to be straight with you, because this is exactly where products tend to overpromise:

• Automated, on every story: the safety screen (personal-information backstop + age-appropriateness judge) and the art-quality and decodable-word gates described above. These run without anyone asking.
• You, in control: you can preview a story, regenerate any page you don't like, set a story private, and report a page that missed the mark. Reporting helps us tighten the checks.
• Not yet in place, on our roadmap: review of our content standards by early-childhood educators. We have not done this yet, so we don't claim it. When it's real, it will be described here in the present tense, with who did it.

Generation is statistical, and no automated check is perfect. Sometimes a page still misses. Read the first one with your kid, keep the stories you love, and skip the ones that don't fit. Questions about safety? Email hello@fablesfactory.com.