How We Keep Stories Safe

Our Safety Promise

No story reaches your child unless it has passed four separate checks. If it fails any of them, it is regenerated — not published, not "probably fine." This page explains exactly what those checks are and why we built them this way.

We think parents deserve to understand the system, not just trust a badge.

Why Safety Is Hard for AI-Generated Content

A children's book written by a human author goes through editorial review, publisher guidelines, and age-rating checks. Those gatekeeping steps exist because human language is subtle — something that reads as neutral on the surface can carry an inappropriate implication for a young reader.

AI-generated content faces the same challenge without the same institutional guardrails. We've spent considerable effort building a safety pipeline specifically tuned for children aged 3–10, because the standard "content moderation" tools used in adult applications are not calibrated for this age group.

The Four-Layer Safety System

Layer 1 — Pre-Generation Filter

Before any story is written, we check your child's profile and interests against a list of topics and themes that should never appear in children's stories. The AI is given explicit constraints: it knows it is writing for a child of a specific age, with specific interests, and it is instructed on what is off-limits.

This is not a blacklist of words — it is a structural constraint on the type of story that can be requested. A child's interest in "fighting games" becomes "adventure and problem-solving" in the story. A child who lists "sharks" as an interest gets ocean adventures, not horror.

Layer 2 — Automated Content Moderation

After each chapter is written, it passes through an automated content moderation system trained on a wide range of text categories: violence, frightening content, mature themes, inappropriate language, and more. Any chapter that exceeds a conservative threshold is flagged and sent back for regeneration.

We set the threshold conservatively — meaning some perfectly acceptable chapters are occasionally flagged and regenerated, because we would rather err on the side of caution than publish something marginal.

Layer 3 — Children's Content Classifier

The general-purpose moderation in Layer 2 was trained on adult content. Children's content has different standards — something that passes adult moderation can still be too scary, too sad, too mature, or too complex for a 5-year-old.

We use a custom classifier trained specifically on children's text to catch these age-specific issues. It flags content that is "too frightening for young children," "inappropriately mature themes," "stereotyping or bias," and "content that depicts adults as threatening or unsafe" — categories that general moderation tools miss.

Layer 4 — Reading Level Verification

The fourth check is not about content type but about reading level. Stories are calibrated to your child's Lexile score. After writing, we measure the actual Lexile level of the generated chapter. If it falls more than 100 Lexile points outside the target range, the chapter is regenerated.

This matters for safety in a developmental sense: a story written at the wrong reading level is either inaccessible (discouraging) or not challenging enough (wasting the opportunity to grow). Both outcomes work against your child.

What Happens When a Story Fails

If a chapter fails any of the four checks:

1. It is automatically regenerated — up to three attempts.
2. If all three attempts fail, the story falls back to a safe, generic story from our pre-approved library.
3. The failure is logged for our team to review.

Your child never sees the failed attempt. From their perspective, they receive a story. From our side, we know something in the generation chain needs investigation.

Human Review (5% Sample)

In addition to automated checks, a random 5% of all stories are reviewed by a human before publication to the child's library. Every story that required more than one generation attempt — because the first attempt failed a check — is also flagged for human review.

We do not claim that every single story is reviewed by a human. That claim would not be true, and we think it is important to be honest about what we can and cannot guarantee at scale.

Parental Controls

You remain in control of what your child reads. From the parent dashboard you can:

• Flag any story — if something concerns you, tap "Report this story" and our team will review it within 24 hours.
• Delete any story from your child's library permanently.
• Update interests at any time to redirect the types of stories generated.
• Review all stories — every story generated for your child is visible in your dashboard, not just the ones they've read.

COPPA Compliance and Data Privacy

StoryKind is designed from the ground up for child safety — not just content safety, but data safety.

We comply with the Children's Online Privacy Protection Act (COPPA) in the United States and are designed to meet GDPR requirements in Europe. We collect only what is necessary to provide the service:

• Your child's first name (never last name)
• Age and reading level
• Interests you select from our list
• Reading history and quiz results

We never collect photographs, school names, precise location data, or contact information about children. We do not share or sell any data about children to third parties for marketing purposes.

A full description of our data practices is in our Privacy Policy.

Questions?

If you have a concern about a story your child received, or a question about how our safety system works, please contact us at [email protected]. We take every content concern seriously and respond within 24 hours.