Courts use it. Social workers use it. Algorithms are now being trained to optimize for it. I've spent 20 years in the room with children in conflict — divorce, abandonment, foster care, school. I know what that phrase looks like when it's working. And when it isn't.
Every video traces back to the same problem: a legal standard nobody can define, applied by professionals who are drowning, now partly delegated to systems that have never sat in a room with a child.
Pattern-level, anonymized stories from 20+ years working with kids, teens, and young adults — camps, coaching, foster care, school, and yes, real crisis too. No names, no identifying detail, ever.
The actual UN Convention on the Rights of the Child, UNICEF's AI guidance, and the other source texts that are supposed to govern all of this — read and reacted to the way a practitioner reads them, not the way a lawyer does. More documents are added as the series continues.
What happens to "best interest of the child" when part of the deciding is done by a machine — risk-assessment tools, companion chatbots, government fraud algorithms — and what the evidence so far actually shows.
Written versions of the channel's core arguments — for readers, researchers, journalists, and anyone who wants the sourced, citable case for each claim.
Ratified by 196 countries, undefined in the text itself. What that openness does when it meets an overworked system — and what happens when it gets written into an algorithm's objective function.
The Dutch childcare benefits scandal: a fraud-detection algorithm, thousands of wrongly flagged families, and over a thousand children taken into care — for a decision no child was ever the direct target of.
A child's time is not an adult's time. The best interests standard asks what is best for a child, and almost never when.
Twenty-plus years working directly with kids, teens, and young adults — running youth programs, fifteen summers on the Leadership staff at American summer camps, coaching families through custody and divorce, and for the last several years, as a transitional foster carer. Currently completing a postgraduate psychology degree, deliberately in parallel with this project rather than before it.
It's the legal standard set out in Article 3 of the UN Convention on the Rights of the Child, requiring that a child's best interests be "a primary consideration" in all actions concerning them. The Convention does not define the phrase. UN guidance says it has to be assessed case by case — which is exactly why the same phrase can justify opposite decisions in near-identical situations.
Yes, in the 196 countries that have ratified the Convention, and in most national family and child protection law. How much it actually constrains a given decision depends heavily on the judge, panel, or agency applying it, precisely because the term itself is left undefined.
UNICEF's own guidance is explicit that when an AI system informs a decision affecting a child, the final decision must still be made by a human who can be held accountable for it — not the score or the flag itself. That requirement exists precisely because risk-assessment tools trained on historical case data can carry forward the same blind spots that already existed in how harm was recorded.
Article 12 gives a child capable of forming their own views the right to express them freely in all matters affecting them, including custody hearings and care reviews, with those views weighed according to the child's age and maturity. In practice, a child's view is often gathered in advance and summarized in a written report rather than heard directly in the room where the decision gets made.
Article 9 says a child shouldn't be separated from their parents against their will unless authorities, subject to judicial review, determine it's necessary for the child's best interest — and that a child separated from a parent keeps the right to maintain contact with both, unless that itself would be contrary to their best interest. In practice, the ongoing part — the maintained contact, the periodic review — gets far less resourcing than the original decision.
No. The US signed it in 1995 but has never ratified it, and remains the only UN member state that hasn't. Signing signaled an intention to eventually ratify — it doesn't make the Convention binding domestic law the way it is in the 196 countries that have ratified it.
A Dutch government algorithm flagged thousands of families for childcare-benefit fraud based on learned patterns rather than individual evidence, disproportionately targeting families with dual nationality. Families were ordered to repay benefits in full with little ability to challenge it, many were financially ruined, and over a thousand children were taken into foster care as a result. The scandal brought down the Dutch government in 2021.
The evidence so far says no. UNICEF's guidance names an "empathy gap" — the difference between a chatbot sounding understood and a child actually being understood — and research has documented engagement-driven emotional-manipulation patterns in companion apps. UNICEF's guidance requires chatbots to disclose they aren't human and avoid being designed to create emotional dependency in children.
New long-form and short videos are published weekly on YouTube, @childbestinterest. Written companion essays are published here as the series continues.