YouTube SEO: How Videos Actually Get Ranked and Recommended

YouTube runs two largely separate discovery systems. Search answers a query someone typed. Recommendation predicts what someone will watch next without being asked. They reward different things, and a video optimised purely for one often fails at the other.

How search ranking works

YouTube search behaves much like a search engine: it matches a query to relevant videos, then orders them by expected satisfaction. The inputs that matter:

• Relevance of title, description, and spoken content to the query. YouTube transcribes your audio and uses it.
• Engagement signals on the video relative to others surfaced for the same query — click-through rate and watch time in particular.
• Overall video and channel performance history for related queries.
• Freshness, weighted heavily for queries about current events and barely at all for evergreen ones.

The important nuance: ranking is query-specific. Your video is not "ranked" in general — it is ranked for each search term, and it can sit first for one phrasing and nowhere for a near-identical one.

How recommendation works

The recommendation system optimises for predicted watch time and satisfaction for a specific viewer, based on their history. It is personalised in a way search is not. What drives it:

• Click-through rate on the impression, which is almost entirely a thumbnail-and-title problem.
• Watch time and retention once clicked.
• Whether the viewer continues watching on YouTube afterward, rather than leaving.
• Explicit signals — likes, subscriptions, and "not interested" responses.
• Topical similarity to what that viewer has watched recently.

Notice what is absent: keywords. Recommendation does not care about your tags. It cares whether people who see the video watch it and stay.

What actually matters, in order

1. Thumbnail and title as a pair. They determine whether an impression becomes a view. Nothing else in this list matters if this fails.
2. The first thirty seconds. The largest single drop in retention happens here on nearly every video.
3. Overall retention. Both systems weight it heavily, and it is the clearest quality signal available to YouTube.
4. Topic-to-audience fit. A video that is excellent but unrelated to what your audience expects will underperform a mediocre on-topic one.
5. Title relevance to real search phrasing — for search-driven videos specifically.
6. Description, primarily for context and links rather than ranking weight.
7. Tags, which carry very little weight today.

Practical search optimisation

For videos you intend to be found through search rather than recommendation:

• Target the phrasing people actually type. Use YouTube's own autocomplete as your primary research — it reflects real queries.
• Put the target phrase near the start of the title, in natural language. "How to clean a espresso machine" beats "Espresso Machine Cleaning Guide Tutorial 2026".
• Say the topic out loud in the first fifteen seconds. YouTube transcribes it, and it confirms to the viewer they are in the right place.
• Write a description that explains the video for a human. Two useful paragraphs beat a keyword list.
• Add timestamps. They improve navigation, and they surface as key moments in search results.
• Answer the question early. Search viewers arrive with intent and leave immediately if you make them wait.

Tactics that no longer work

• Tag stuffing. Tags are a minor signal, mostly useful for handling common misspellings of your topic.
• Keyword-stuffed descriptions. They read badly to humans and add little.
• Copying a successful channel's tags. Their performance came from their content, not their metadata.
• Chasing high-volume search terms your channel has no authority for. A smaller, precisely matched query you can win is worth more than a large one you cannot.
• Misleading thumbnails. They inflate click-through and destroy retention, which the systems read as dissatisfaction.

The two-track content strategy

Search videos and recommendation videos are different products. Search videos answer specific durable questions, accumulate views slowly, and keep working for years. Recommendation videos ride interest and topicality, spike hard, and decay.

A channel built entirely on recommendation lives or dies on each upload. A channel built entirely on search grows slowly but compounds. Most durable channels run both deliberately, and know which one each video is for before they make it.

“Decide whether a video is meant to be found or meant to be served, before you write the title. The two demand different titles.”