How it works
How we know
Every line on a KidSight card comes from watching the video and measuring it against a written set of criteria. This is exactly what that means.
The reading
What KidSight AI looks at
The analysis behind every card — we call it KidSight AI — starts with the video itself. Not the title, not the description, not a summary somebody else wrote. The whole thing goes to the model, frames and audio together, and what comes back is an account of what actually happens on screen.
It is told, in as many words, that it is an observer and not a judge. It reports; it does not rate. Around sixty separate readings come back from one watch: what a video teaches and how it goes about teaching it, how it moves and how it sounds, who is in it, what it asks a child to do, and a set of safety readings.
KidSight AI is not trained to have opinions about children’s video. It applies the same written criteria to every video, in the same order, every time.
The scoring is separate from the watching
The part that decides whether a video is any good for a child of a given age is ordinary code, not the model. It reads a versioned rulebook — fixed criteria, fixed weights — and reaches the same answer from the same account every time. Run it again and the numbers land where they landed before.
That separation is deliberate. If the model both described a video and rated it, a change in our standards would be impossible to tell apart from a drift in the model. Keeping them apart means we can change our minds about what matters, run the whole library again, and see exactly what moved.
The rulebook is checked before it can go live. A criterion cannot reference a framework that hasn’t been declared. An attribution cannot attach to a criterion that doesn’t exist. A weight that fails to resolve stops the release rather than quietly scoring as zero.
Age
How KidSight AI reads pace
Pace is a judgment, and we treat it as one. KidSight AI places a video on a four-point scale running from calm to frenetic, and reads how much is happening on screen and in the audio alongside it.
What matters is what happens next. Pace is never scored as good or bad on its own. It is priced differently depending on how old the child watching is.
Pace, weighted by age
Calm peaks at two, then falls away. Fast does the reverse.
at two
ages 0–2
at four
ages 3–5
at seven
ages 6–8
at ten
ages 9–10
Calm
slow cuts, long takes
1.0
0.9
0.7
0.6
Moderate
steady, varied rhythm
0.6
0.8
1.0
1.0
Fast
quick cuts, high energy
0.2
0.4
0.6
0.7
These are the live weights: how much a video’s pace counts toward its fit at that age, on a scale of 0 to 1. They are not a score. Current as of August 2026.
A calm video counts fully toward a good fit for a two-year-old, and a good deal less by the time a child is nine. A fast one runs the other way. Same reading, same video, opposite verdict — and the same thing happens with whether a video is soothing or upbeat, and with how much of it is carried by music.
One video, more than one answer
Every video in the library is scored separately for the youngest children, for preschoolers, for school age, and for the oldest children we cover. Separate results, stored separately. When you open a card, you are reading the answer for your child’s age — not a general rating with an age note bolted on.
The criteria themselves change as a child gets older, and they tighten at both ends.
01
For the youngest, the bar is narrow
A video has to build something up, piece by piece, for it to count as teaching at all. Explaining a thing outright earns nothing until around six.
02
Most of the social side is switched off
A video cannot earn credit for something like naming feelings before a child is three. Not marked down — unavailable, because that isn’t what the research says is happening yet.
03
At the older end it tightens the other way
A video that assumes a nine-year-old knows nothing is demoted for it, by two full rungs, with its own recorded reason.
04
Which is how a good video stops counting
A counting song that genuinely teaches a four-year-old is, by nine, just something a child watches. The card says so rather than pretending otherwise.
Provenance
What the criteria are built on
Every criterion is written against published child-development research, and each one carries its attribution inside the system itself — not in a marketing document, but in the configuration the scoring engine reads on every analysis.
How to weigh them is our judgment: what ought to count for more at three than at eight, where a bar sits, what earns credit and what doesn’t. We hold those as editorial calls and we’d rather name them as such than dress them up as calibration.
No one can pay to be in the library.
No creator has a relationship with us, and none has ever had a say in what a card says. No research organisation has reviewed KidSight, approved it, or is affiliated with us. We read what they publish, the same as anyone can.
In practice
From a measurement to a lineup
A video reaches a shelf only by earning it — a real criterion has to qualify, at your child’s age. Where a child’s age can’t support a shelf, the shelf doesn’t appear at all. We’d rather show you fewer shelves than pad one out.
No show takes more than three places on a shelf. There are over a hundred and fifty Bluey episodes in the library, and without that rule one show would swallow a shelf whole.
The daily pick clears a higher bar than anything else on the page: watched in full, at the top of the range, teaching something real, and clean on every safety check — including one we run and never display.
Nothing here learns from you. There is no behavioural tracking anywhere in this, because we link out to YouTube and never see what got played. Two children the same age see the same library.