Netflix AI Discover Alternative for Your Own Streaming Service
Netflix does not sell this. Below is every element it has described, what TasteRay provides for it, and what your team would still have to build.
Netflix does not sell AI Discover: as of November 2026 it is a test inside its own TV app. A service that wants something similar can use TasteRay's API for mood-based ranking and written reasons, then build the tab, themes, clips and rejection memory itself. Large operators can add Enterprise terms.
If your board has read the Netflix announcement and asked what your service is doing about it, the honest answer has two parts. The ranking and the reasons can be bought today. The experience around them is your product work, and this page lists it line by line.
What Netflix has said, and when
These facts come from Netflix's post of 3 November 2026 and its Help Center article, both read on 5 November 2026.
- A new tab in the TV app, for a small group of US members who use Netflix in English on supported TVs, opening to more people "over the coming weeks".
- Themes previewed by a clip from a show or movie, such as "Something Light", "Scare Me", "Family Movie Night" and "Quick Watches".
- Recommendations based on watch history, preferences and time of day, according to the Help Center.
- Netflix calls it a "test kitchen" and says ideas may reach the main experience. It is optional and absent from Kids profiles.
- It is powered by "a new intelligence layer". Netflix names no model or vendor and gives no metrics, so nobody outside can claim to work the way it does.
Each element, what we provide, what you build
Read the third column before you read anything else. TasteRay is the engine and the reasons, and a usable tab needs more than those.
| AI Discover element | TasteRay provides | You build |
|---|---|---|
| A Discover tab in the TV app | Nothing on screen: a REST API, with no TV SDK | The tab, its icon and the screens, in your own apps |
| Themes such as "Scare Me" | No theme endpoint. A theme is a string you send in context.preferences | The theme list and its copy. AI-written themes need your own model call |
| A clip previewing each theme | Nothing. The API returns text, not video | Clip selection and playback from your assets |
| Matching recommendations | POST /v1/recommend with items ranks your list, up to ten per request, with a match score | Preselecting candidates for each theme from your catalog |
| Narrower themes and "Something Different" | A new request with narrower preferences | Navigation, state, and the next batch of candidates |
| "Tune Your Recommendations", remembered | Takes up to 50 rated items in context.history per request. The API itself remembers nothing | Storing rejections, and filtering them out before you send |
| Results explained | why_match and key_factors on every result | Where the reason sits on your card |
| Based on history, preferences, time of day | Reasons from whatever you send: history, preferences, a free-text profile | Deciding what to send, and sending it each time |
| An intelligence layer that "keeps learning" | Stateless self-serve API. Enterprise can train on your content | Your feedback loop and measurement |
| Not described in Netflix's post: typing a request in your own words | context.profile takes free text | The input box or voice capture |
A theme is one request
To test the idea we sent "Theme: Scare Me" with two preferences and a list of ten titles, including Paddington, The Holiday and Moneyball, none of which came back. The theme name means nothing to the API on its own, so the sentences you write for it do the work.
{
"vertical": "movies",
"context": {
"preferences": ["Theme: Scare Me", "genuinely frightening", "atmospheric dread over jump scares"]
},
"items": [
{"id": "h1", "name": "The Babadook", "description": "A widowed mother and her son are haunted by a children's book monster.", "metadata": {"year": 2014, "runtime_minutes": 94}},
{"id": "h2", "name": "Hereditary", "description": "A family unravels after a death, as something older stirs in the house.", "metadata": {"year": 2018, "runtime_minutes": 127}},
// ...and 8 more titles: Paddington, The Holiday, Moneyball and others
],
"options": {"count": 3}
}meta.processing_time_ms 1,847. The model's sentences are verbatim.Hereditary
98 out of 100, taste_match_score
You don't want to be startled—you want to be unsettled. This film is the definitive masterclass in atmospheric dread, ensuring that the fear doesn't leave when the credits roll.
- It respects your intelligence. The narrative is dense, layered, and requires your full attention. You'll feel challenged by the mystery, not just scared by the visuals.
- The acting, particularly Toni Collette's, is terrifyingly real. You won't feel you're watching a performance; you'll feel you're witnessing a genuine psychological breakdown in real-time.
The Babadook
95 out of 100, taste_match_score
This is a masterful case study in how grief manifests as horror. It delivers the 'atmospheric dread' you crave by making the home itself feel like a hostile, claustrophobic entity.
- It’s shorter (94 minutes) than Hereditary, making it a quicker, more intense dose of dread without the longer narrative buildup, perfect for a focused watch.
It Follows
88 out of 100, taste_match_score
This film creates a unique, paralyzing anxiety through the simple concept of an inescapable threat. It’s a test of nerve, forcing you to watch a character try to outrun a fate that feels inevitable.
- It’s a different flavor of dread: kinetic anxiety. Instead of the slow, creeping horror of a haunted house, this is the stress of a ticking clock and a relentless pursuer.
Fields shown: item.name, metadata.taste_match_score, explanation.why_match, explanation.key_factors.
The same call, for a different evening
Change the preferences and the list, and the reasons change with them. Here the theme carries an age range, and the ten titles included Se7en and Blue Is the Warmest Colour, which did not come back. Treat that as a result to test on your catalog, and keep your own content rating filter in front of the API.
{
"vertical": "movies",
"context": {
"preferences": ["Theme: Family Movie Night", "suitable for ages 6 to 12", "all ages can enjoy"]
},
"items": [
{"id": "f1", "name": "Paddington 2", "description": "A gentle, funny story of a bear framed for theft.", "metadata": {"rating": "PG", "runtime_minutes": 104}},
{"id": "f2", "name": "Spider-Man: Into the Spider-Verse", "description": "Animated adventure of Miles Morales.", "metadata": {"rating": "PG", "runtime_minutes": 117}},
// ...and 8 more titles: Se7en, Blue Is the Warmest Colour and others
],
"options": {"count": 3}
}meta.processing_time_ms 1,963. The model's sentences are verbatim.Paddington 2
98 out of 100, taste_match_score
You're looking for a movie night that brings the whole family together in a warm, safe bubble. Paddington 2 is the emotional anchor of family cinema—it makes children giggle with delight and adults feel a gentle nostalgia, ensuring everyone at the table smiles by the end.
- At 104 minutes, it respects the attention spans of your younger viewers while providing a complete, satisfying narrative arc for the older ones. No one gets bored; everyone stays engaged.
Coco
95 out of 100, taste_match_score
Coco bridges the gap between childhood wonder and deeper emotional meaning. It allows the 6-12 age group to enjoy a vibrant musical adventure, while the underlying themes of family and memory offer a touch of maturity that makes the older kids and parents feel seen.
- It’s short enough (105 mins) to keep engagement high, yet deep enough to be the topic of dinner conversation afterwards.
Kiki's Delivery Service
92 out of 100, taste_match_score
If you want a quiet, calming evening, this is it. Kiki’s Delivery Service is a gentle slice-of-life that feels like a warm blanket. It’s perfect for winding down, offering a story about independence and kindness that appeals to the whole spectrum of ages.
- The pacing is slow and soothing. This is ideal if your family has been through a chaotic week and needs to decompress. It doesn't demand high energy; it invites relaxation.
Fields shown: item.name, metadata.taste_match_score, explanation.why_match, explanation.key_factors.
For a small service, for a large one
- A small service
- Start on a free key with a handful of themes written by your editors. Precompute the rows, ship a separate tab to some of your viewers, and pay $49 or $199 a month if it earns its place. The guide has the code.
- A large operator
- The same pattern, with custom AI trained on your content, white-label delivery, an SLA and custom limits, starting from a pilot. See Enterprise and security.
Netflix is a separate company; see its own site. TasteRay is not affiliated with it.
Sources and further reading
- Introducing AI Discover (Netflix, 3 November 2026)
- Explore in new ways with AI Discover (Netflix Help Center)
- TasteRay API documentation
Frequently Asked Questions
Can I license Netflix AI Discover for my own service?
Nothing Netflix has published offers that. As of its 3 November 2026 post, AI Discover is a test tab inside the Netflix TV app for a small group of US members, with no API or licence mentioned. A service wanting something similar has to build it, with a vendor or on its own.
Does TasteRay generate themes the way AI Discover does?
No, the TasteRay API has no theme endpoint. You write the themes or generate them with your own model, then send each as preferences in a normal request. Netflix says AI helps create its themes, so a service that wants the same has to add that step itself, on top of TasteRay.
Does TasteRay work the same way as Netflix AI Discover?
That cannot be known. Netflix describes a new intelligence layer without naming a model, a vendor or how it works, as of November 2026. TasteRay is a separate company, not affiliated with Netflix, and its documented API can be compared only with what Netflix has said its feature does.
What is the fastest way to test this idea on our own catalog?
Open the TasteRay playground, which needs no key, and paste ten titles from your catalog with a theme written as preferences. Read the reasons as your viewers would. If they hold up, get a free key, precompute a few rows, and show them to a small group in a separate tab.