October 10, 2026
I Tested 24 AI Bedtime Stories for a Hard Night. With Only a Lesson, None Mentioned It.
In July I ran a small test on the stories Z Tales writes, because of one result I could not stop thinking about. In a trial run where the family dog had just died, the story the app wrote was called "Ada's Kind Adventure in the Enchanted Forest". It was about pulling a thorn out of a rabbit's paw. The dog was not in it.
The reason was simple. Back then the story maker asked one question about the night: "What should tonight's story teach?" It offered a list of virtues. A parent whose dog has died does not want to teach anything, so they pick the nearest one, kindness, and the AI takes kindness literally.
The test
Four hard nights, the kind parents actually come to Z Tales with:
- The first day of preschool is tomorrow
- A new baby sister is arriving
- The family dog has died
- Two siblings fought all day and share a room tonight
Everything else was held the same: one child (Ada, age 4 to 5), one story world (an enchanted forest), one AI model (gpt-4o-mini, the one the app used at the time), and the app's real story instructions, copied word for word.
Version A was what the app did then: the night squeezed into a lesson, so courage for preschool, kindness for the new baby and the dog, sharing for the siblings.
Version B gave the model the situation itself, in the words a parent would use, plus one rule: say what is happening early, keep the child the hero, do not fix what cannot be fixed, end with the child at home and going to sleep, and no moral at the end.
Each version wrote two stories for each night. Every story was read in full, not just scored by a script, because in the first round the automatic checks missed the main problem completely: they marked the lesson-only stories as clean while every one of them ignored the parent's night.
What happened
With only a lesson, 8 stories:
- 0 of 8 mentioned what the child was facing. No preschool, no baby, no dog.
- 5 of 8 ended on a spoken moral, like "Kindness makes the world brighter!"
- 2 of 8 ended with the child at home and going to sleep. The rest ended in the forest.
With the situation and the rule, 8 stories:
- 8 of 8 named what was happening, 6 of them on the very first page.
- 0 of 8 ended on a moral.
- 8 of 8 ended at home and asleep. In both dog stories, Ada falls asleep thinking of her dog, by name.
What it did not fix
The specific worries a child carries into a hard night (which hook is mine, who picks me up, where is the dog now) showed up in only about a third of the stories, even with the rule. The model also still loves a talking squirrel, and most stories wandered off into the forest before coming home.
It was a small test: two stories per version per night, one model, one child. Read it as a clear direction, not a law. One more thing worth saying plainly: these are bedtime stories, not therapy, and nothing here says a story changes how a child copes.
What changed in Z Tales
The story maker now asks about tonight's situation as well as the lesson. There are 47 of them, from the dentist and a new babysitter to two homes and a hospital stay, and choosing one replaces the lesson rather than sitting next to it. Every free story page about a situation opens the maker with that situation already picked.
The free social stories on the site are written by hand to the same rule, and most now start with the steps in order and come with a free printable.
If you use any AI to write a bedtime story
You do not need Z Tales for the main finding. Whatever tool you use:
- Tell it what tomorrow actually holds, in your own words, not the virtue you hope it teaches.
- Name the one detail your child keeps asking about.
- Ask for the story to end with your child in their own bed, falling asleep.
- Ask for no moral at the end.
Damon
Damon Scaife builds Z Tales on his own — the story rules, the illustration prompts, the code and the safety guardrails. He answers hello@ztales.com himself. More about Z Tales.