Problem
The assumption was that bedtime reading is boring because it is generic, and that a child hearing their own name in a story would be more engaged than one hearing about a stranger. Personalization looked like the obvious missing feature.
That assumption turned out to be wrong, which is the useful part of this entry.
Why I built it
Because it was the cheapest way to test a belief I actually held. Generation was a solved problem by then, so the build was short and the answer would come from users rather than from me arguing with myself. The point was never the product; it was finding out whether personalization was the thing parents wanted.
Solution
The product took a child profile and a short parent prompt and generated a custom story from them.
Outcome
- Activation was fine — first-time visitors got through the flow and generated a story.
- Return usage dropped sharply after the first successful generation.
- Users described the output as interesting, but not worth coming back for.
- Given the choice, parents repeatedly preferred high-quality pre-written stories over a personalized generated one.
What they were raising was a quality question, not a feature question. Nobody asked for more controls. They asked whether the output could be trusted — and for something you read to a child at night, that is the whole decision.
Learnings
Personalization does not beat narrative quality in a product people use repeatedly. Novelty gets the first session; coherence gets the second one. If the generated version loses to a curated version on a straight comparison, adding settings on top of the generated version does not change the result.
The longer write-up lives in the AI Story Maker PMF analysis.