Parental controls
How the AI content filter works
In short
Every video added to a library is screened against 21 content patterns using the video’s own transcript. Per profile, you decide which patterns block a video outright and which only flag it for your review. Every decision is visible and can be overridden.
What it actually reads
The filter works from the video’s transcript — the actual spoken words — rather than from its title, its thumbnail, its tags, or its YouTube age rating. Those metadata signals are exactly what misleading content is optimised to game, and they are the reason a video called "Fun Learning Colours" can turn out to be nothing of the sort.
Working from the transcript means the filter is judging what is said in the video. A creator who swears in minute eleven of an otherwise gentle video is caught; a video with an alarming title and entirely benign content is not falsely blocked.
It is a second line rather than a first. The library is the primary boundary — the filter exists to catch drift inside a set you already approved, which is the failure mode curation on its own does not handle. Channels change, creators have off days, and a channel you added in March is not necessarily the channel it was.
The 21 patterns
The patterns cover the range of things parents actually raise, not just the obvious categories. Language and realistic violence are there, but so are hidden advertising and sponsored content, gambling mechanics in games, engagement-baiting, parasocial and emotional manipulation, materialism and wealth-flexing, beauty and body pressure, dangerous internet challenges, and toxic behaviour modelling.
That breadth is deliberate, because the content most parents object to is usually not age-restricted. A twenty-minute unboxing that is functionally an advert breaks no platform rule. Neither does a creator whose entire register is escalating outrage, or a stunt that is technically harmless and obviously imitable by a seven-year-old.
Some patterns are positive or neutral markers rather than concerns — educational framing, for instance. Those are never counted as concerns, so a video is not penalised for being explicitly educational.
Each pattern is scored rather than treated as a simple yes or no, which is what lets the same pattern block for a six-year-old and merely flag for a twelve-year-old without maintaining two separate systems.
Setting what happens per pattern
For each profile, every pattern is set to one of these. Defaults are seeded from the profile’s birth year and can all be changed.
- Block
- A video matching this pattern does not enter the library and never appears on the child screen. Use for the patterns you consider non-negotiable at that age.
- Flag for review
- The video is held and shown to you with the reason it was flagged. You decide. Best for patterns that depend on context — the same one can be fine in one video and not in another.
- Allow
- The pattern is not treated as a concern for this profile. Sensible for an older child where a pattern that matters for a six-year-old is simply not relevant.
Overrides and why they are per video
Any individual decision can be overridden. If the filter held something back and you disagree after looking at it, allow it for that profile and it joins the library.
Overrides are stored per video and per family rather than as a change to your pattern settings. That is deliberate: one video being an exception should not quietly widen the rule for everything that follows it. If you find yourself overriding the same pattern repeatedly, that is the signal to change the pattern setting itself rather than keep making exceptions.
Overrides do not expire and are not revisited, so a video you allowed stays allowed. Removing it from the library removes it regardless of the override.
Nothing is filtered silently. Everything the filter held back is visible in the dashboard with the pattern that triggered it, so you always know what was withheld and why — you are never in the position of wondering whether something was blocked or simply did not exist.
Common questions
Does the filter replace me choosing the library?
No, and it is not meant to. The library is the boundary; the filter is a safety net over content you already approved. Its job is catching drift — an approved creator changing register, a channel that expanded into something new — not deciding what your child may watch in the first place.
What if the filter blocks something that is fine?
Override it. The video and the reason it was flagged are both visible in the dashboard, so you can look at what triggered the decision and allow it for that profile. Repeated overrides on the same pattern usually mean that pattern is set too strictly for that child.
Can the filter miss something?
Yes. It reads transcripts, so it does not catch purely visual content with no spoken component, and it cannot screen live streams before they air. It is a meaningful reduction in risk over a set you already chose, not a guarantee — and we would rather say that plainly than imply otherwise.
Are different children screened differently?
Yes. Pattern settings are per profile, seeded from each child’s birth year. The same video can be blocked for a six-year-old and available to a twelve-year-old in the same family without you maintaining two libraries by hand.
Does screening happen again if a video changes?
New uploads from channels in a library are screened as they appear. An already-screened video is not continuously re-analysed — the meaningful risk is a channel’s new content rather than an old video being edited after publication.
Related articles
Try it with your own kids
The free trial needs no card, and nothing installs — build a library, set a limit, and hand the device over.