Recommendation is the interesting problem, so it gets the attention. Search is treated as solved — a box, a query, a list of results — and inherits whatever the platform provides by default.
Then you look at the logs and find that search is where people go when they have a specific intent, which is the moment they are least tolerant of failure.
The queries that matter are the hard ones
Search traffic in a media product is not evenly distributed. A large share is:
- Exact titles, often misspelled or half-remembered.
- People. An actor, a presenter, a pundit — frequently the strongest signal of intent there is, and frequently unsupported because contributors were stored as free text.
- Descriptive queries. “That documentary about the trawlers.” No title, no contributor, just a memory of what it was about.
- Franchise and series, where someone wants a brand and gets a single episode from 2019 ranked first.
A search that only matches title strings handles the first of those, badly, and none of the rest.
Failure is silent
The worst property of search is that its failures leave no trace in the product. Recommendation failing is visible — nothing appealing on the home screen. Search failing produces an empty result page, and the person leaves.
Zero-result queries are the most useful log a media business has and almost nobody reads them. They are a direct statement of what an audience expected you to have, in their own words. Some of it is content you do not own, which is buying intelligence. Much of it is content you do own, described in a way that did not match how anyone thinks about it.
What good search needs
Mostly the same foundation as recommendation:
- Typo tolerance and partial matching, because titles are misremembered.
- Structured entities for people and series, so a contributor search works at all.
- Synonyms and colloquial names — what the audience calls something rather than what the rights holder calls it.
- Availability applied to results, so nothing surfaces that cannot be played.
- Semantic matching over transcripts and descriptions, which is where models genuinely help — the trawler documentary is findable if the content itself has been indexed rather than just its title.
The order to do it in
Search first, recommendation second. It is cheaper, its failures are measurable, it serves the highest-intent users, and it exercises the same metadata foundation that any recommender will need.
A business that fixes search usually discovers the discovery problem was smaller than it looked.
Looking at discovery? Get in touch, or read about our entertainment and media work.