Most keyword advice written for Google transfers badly to YouTube. Search volume matters less, competition means something different, and a huge share of views never come from search at all. But the research still matters — not because it tells you what to rank for, but because it tells you what to make.
The goal is not a spreadsheet of 400 phrases. It is a shortlist of five topics where demand exists and the current results are beatable.
First: is this a search video or a browse video?
Before researching anything, decide which surface the video is for. They have different rules and mixing them up wastes effort.
Search videos answer a question someone typed. Tutorials, reviews, comparisons, "how to fix X". They accumulate views slowly and keep earning them for years. Keyword research is the whole game here.
Browse videos are served to people who were not looking for anything. Vlogs, essays, commentary, most Shorts. They spike hard and fade. Here the title and thumbnail carry the video; keyword research does almost nothing.
Check your channel's split in YouTube Studio under Analytics → Content → Traffic source types. If Search is under 10% of your views, keyword research will not transform your channel — improving titles and thumbnails will. If Search is 30% or more, this workflow is the highest-return hour you can spend.
Step 1: Harvest raw phrases
Start wide and cheap. Four sources, fifteen minutes:
YouTube autocomplete. Type your seed topic into the YouTube search box and read the suggestions. These are real queries, ranked by how often people type them. Then repeat with a letter appended — "resolve editing a", "resolve editing b" — to fan out the list. Tedious, and still the single best free source of intent data on the platform.
"People also watched" and search-result sidebars. These show you how the platform groups your topic, which is a different thing from what you assume it groups with.
Comments on competitor videos. The questions people ask under an existing video are literally unmet demand for the next video. This is where the best topics come from and the least competition is.
Competitor metadata. Run the top-ranking videos for your seed through the tag extractor to see the vocabulary they used, then look at what their titles have in common. If four of five titles contain the same qualifier, that qualifier is part of the query.
Step 2: Read intent, not volume
For each phrase, work out what the person wants. This matters more than any number a tool gives you.
| Query shape | What they want | Video that wins |
|---|---|---|
| "how to …" | A procedure they can follow now | Screen recording, chapters, no preamble |
| "best … for …" | A decision made for them | Ranked list with criteria stated |
| "… vs …" | A tie broken | Direct comparison, verdict early |
| "… not working" / "… error" | A fix, urgently | Fix in the first 60 seconds |
| "… review" | Confirmation before spending | Real usage, downsides included |
| "what is …" | An explanation | Plain language, examples |
A video that answers a different intent than the query implies will get clicks and lose them all in thirty seconds. The classic failure is answering "how to" with an essay about why the topic is interesting.
Step 3: Judge whether you can win
This is where most research goes wrong. People find a phrase with big numbers, make the video, and land on page four forever.
Search the phrase on YouTube and read the first page like a competitor analysis:
- Who ranks? If the first five results are all channels with a million subscribers, the phrase is not available to you yet.
- How old are they? Results dominated by videos from three years ago on a topic that changes annually is an opening.
- Do they actually answer it? If the top results are tangential — the query is "fix audio drift" and the results are general editing tutorials — the query is under-served.
- How long are they? A 25-minute video ranking for a question answerable in three minutes is an opening.
- Are they good? Watch thirty seconds of the top two. If they are dated, badly recorded, or padded, you can beat them with effort alone.
The best target is a phrase with modest demand and clearly weak results, not a phrase with huge demand and strong results. Ten videos ranking first for small queries beat one video ranking fortieth for a big one — and the small queries compound, because each one teaches the platform what your channel is about.
Step 4: Cluster into a series
Group the surviving phrases into clusters of three to six closely related videos. Clusters work better than isolated videos for a specific reason: viewers who finish one want the next, and the platform notices that your channel satisfies a whole topic rather than one query.
A cluster for a video-editing channel might be:
- "davinci resolve multicam setup"
- "davinci resolve audio sync not working"
- "davinci resolve timecode explained"
- "multicam editing: resolve vs premiere"
Each is a separate search intent. Together they own a topic. Link them to each other in the description and with end screens.
Step 5: Track what happened
Research without measurement is just opinion. After publishing, you need to know whether the video actually reached position 3 for the phrase you built it around — and that is not something Studio tells you directly.
Check position for your target phrase with the rank tracker at one week, one month, and three months. Three patterns and what they mean:
- Ranks well, few views. The query has less demand than you thought. Not a failure — a cheap lesson about the niche.
- Ranks well, views, poor retention. Intent mismatch. The video answers a different question than the one asked.
- Does not rank at all after a month. Either the competition is beyond your channel's current authority, or the video is not recognisably about that phrase. Re-check the title and the description opening.
Keep a simple sheet: phrase, video, position at 1/4/12 weeks, notes. After twenty videos you will have something no keyword tool can give you — a record of which query types your specific channel can win.
The tooling, honestly
You do not need much:
- YouTube autocomplete — free, and the highest-signal source there is.
- Tag extractor — competitor vocabulary in seconds.
- Rank tracker — did the video actually land where you aimed.
- A spreadsheet — phrase, intent, competition read, decision.
Volume estimates from third-party tools are directionally useful and precisely wrong. Treat them as a way to sort candidates, never as a reason to make a video you would not otherwise make.
The short version
- Decide first whether the video is for search or browse; only search videos need this.
- Harvest phrases from autocomplete, comments, and competitor metadata.
- Classify by intent, and build the video to match that intent exactly.
- Judge competition by reading the first page, not by a difficulty score.
- Cluster three to six related videos instead of making one-offs.
- Track position at 1, 4, and 12 weeks and write down what you learn.
