Tubi teardown

The problem

Tubi has one of the largest free libraries in streaming, over 200,000 titles. A big library only helps if people can find something worth watching in it. The challenge is to turn volume into confident choices, so that viewers who arrive for free content stay because the next pick is easy and good.

• Algorithms: Make recommendations feel earned and responsive, including for new viewers with no history and for people who want something different.
• Findability: Let people search and browse by what they actually want, such as a decade, a mood or a country, without scrolling endless rows.
• Content quality: Help people tell the gems from the filler in a catalog of mixed quality, so a bad first pick doesn't end the session.

Research: how Tubi is doing

Tubi is growing and profitable, and viewers rate the app highly. The friction is about ads, picking from a mixed catalog, and search that doesn't read intent.

The numbers

• Scale: 100 million or more monthly active users and about 2.2% of US TV viewing, per company and press figures. Library size is reported as 200,000 titles and up, with some sources saying 275,000 to 300,000.
• Business: More than $1.1 billion in revenue in fiscal 2025, and profitable since the first fiscal quarter of 2026.
• Audience: Over a third of viewers are 18 to 34, and about 63% are cord cutters or cord nevers.
• Algorithm: Tubi engineers say the homepage is assembled by 70 or more machine learning models, trained on roughly a billion hours of monthly viewing. That 70+ figure comes from 2023 posts, so treat it as a past number.
• Other teardowns: The only full teardown I found is Appfigures' 2021 app-store teardown, which graded Tubi A-. It covers store optimization, not recommendations or search.

Ratings and reviews

Store figures are from trackers and the listings, and may differ slightly from the live pages. Trustpilot reviews skew toward people with a problem.

How the recommender works

What users complain about

1. Ads: The top complaint, including ad load and repeats within a break.
2. Search that misses intent: In my test, "scary movie from the 80s" returned horror titles but ignored the decade.
3. No way to narrow results: The results page is a poster grid with no filters or sort.
4. Hard to judge quality: Tiles show year, runtime, rating and genre, but no review score.
5. Menu sprawl: Browse has many sub-menus and lists genres twice.

Complaints 2 to 5 come from my own logged-out session on November 3, 2026, with two searches. It is one session, so it shows what is possible, not how often it happens.

Who uses Tubi

Three illustrative personas, built from the audience data and reviews above. They are hypotheses to test with real interviews, not research findings.

The user journey

Getting started and watching are smooth. The weak points are the middle stages where a viewer has to decide, and the last one where feedback should pay off.

Step by step

1. Land and start. The homepage loads rows of posters with no sign-in needed to browse. Strength: Almost no friction to get to a title.
2. Browse rows. Rows are ranked per viewer, with tiles showing year, runtime, maturity rating and genre. Gap: No review score or quality cue on the tile.
3. Use the Browse menu. Many sub-menus by genre and collection. Friction: Genres appear twice and the structure is deep.
4. Search. "korean thriller" returned relevant results. "scary movie from the 80s" returned horror but ignored the decade. Gap: No filters or sort on the results page.
5. Watch. Playback with ad breaks. Friction: Ads are the top complaint in reviews.
6. Give feedback and come back. Likes, dislikes and My List feed the models. Gap: The effect on your rows isn't visible, so it doesn't feel like it works.

How it holds up

• Function: Free, fast and no sign-in covers the core job. The recommender is serious engineering.
• Design: A poster-grid layout that is familiar and scannable. It has little metadata for deciding between titles.
• Performance: Reviewers mention buffering and crashes sometimes. I have not measured load times myself.

Where it can improve

1. Let people narrow results: Filters and sort on search and browse pages.
2. Show quality at a glance: A compact score or badge on every tile.
3. Make feedback visible: After a like or dislike, show the row change and say why.

How competitors compare

Each rival wins on one axis Tubi can borrow from: Pluto TV on lean-back viewing, Roku on distribution, Netflix and Prime on recommendation polish.

Three lessons carry into the ideas below. Pluto shows people sometimes want no choice at all. Netflix and Prime show that explaining a recommendation builds trust. Tubi's ChatGPT app shows it can lead on conversational discovery, which rivals have not done.

Five ideas to fix it

Each idea targets the same loop of finding, judging and trusting what to watch, and each touches the algorithm, findability and content quality in a different way.

1. Faceted search and sort. Add filters for decade, genre, country, runtime and rating, plus sort by popularity, newest or score, on search and browse results.
   • Solves: Search that ignores intent, and results nobody can narrow.
   • Why it matters: Turns "scary movie from the 80s" from a miss into a two-tap answer. It is mostly UI over metadata Tubi already holds.
2. Quality badges. Show a compact score (for example critic and audience ratings from sources Tubi already uses) and a short label like "Cult favorite" on every tile.
   • Solves: No way to tell a gem from filler.
   • Why it matters: Protects the first pick, which is the choice that decides whether someone stays. It also gives the ranker a visible quality signal.
3. Feedback that sticks. After a like or dislike, update the row right away and say why ("More like Heat, since you liked it").
   • Solves: Signals that seem to go nowhere.
   • Why it matters: Builds trust in the algorithm and produces more labeled data from engaged viewers.
4. In-app conversational search. Bring the ChatGPT-style experience into the app, so a viewer can type "a tense Korean thriller under two hours" and get a short, ranked answer with reasons.
   • Solves: Intent that keyword search can't read, and the feed-native expectation.
   • Why it matters: Extends Tubi's lead on AI discovery. Earlier attempts reportedly stalled, so this needs careful scoping.
5. Cold-start taste picker. For new or logged-out viewers, offer a 20-second pick of a few titles or genres to seed the first homepage.
   • Solves: Generic first rows for people with no history.
   • Why it matters: Improves the first session, but popularity fallbacks already do an acceptable job, so the gain is modest.

Sketch: filters on search results

Sketch: feedback that changes the row

What to build first

Ship filters and visible feedback first, because both are low effort and fix problems I hit myself. Quality badges and conversational search are bigger bets with higher payoff. The cold-start picker can wait. These placements are my initial estimates to confirm with engineering and usage data.

Sources

Opened and read, as of November 3, 2026:

• Tubi, hands-on logged-out session (homepage, Browse menu and two searches)
• Tubi engineering blog posts on the recommender, including embeddings, recall and exploration
• Appfigures, 2021 Tubi app-store teardown
• Product Teardown Worksheet template (Notion)

Also referenced from search results or snippets only, so details may differ: the TubiFM paper on arXiv, press coverage of Tubi's revenue, profit and audience figures, store and Trustpilot ratings, reports on Rabbit AI and the April 2026 ChatGPT app, and coverage of Freevee's retirement. One UI review I tried to open would not load, so I did not rely on it.