Keyword research for video goes wrong when it produces a list of terms instead of a list of videos worth making. Useful youtube keyword research ends with topics you can film well, phrased the way viewers actually ask.
Start from questions, not volume
Search suggestions are the cheapest source of real phrasing available: begin typing a subject and read what completes it. Those completions come from actual searches, and they show the wording people use rather than the wording your industry prefers.
Your own analytics are the second source. The search terms already bringing people to your videos indicate where you have some standing, and expanding around those usually outperforms attacking unrelated high-volume subjects from nothing.
The method
- Collect completions for your core subjects, recorded verbatim.
- Add the search terms already reaching your channel.
- Add recurring questions from comments and messages.
- Group them by underlying need rather than by wording.
- For each group, judge whether you can genuinely make the best version.
- Check what already ranks; if it is thorough and current, pick a sharper angle.
- Convert survivors into titled video ideas with a stated promise.
That honesty check matters. Targeting a subject you cannot cover better than the existing results wastes a production slot, however attractive the search volume looks.
Search is only one route
Much video viewing comes from suggestion and browsing rather than search, and those respond to whether people watch rather than to phrasing. Keyword work therefore has a real but bounded role: it helps the right people find a video that then has to hold them.
Avoid contorting a title to fit a phrase. A title that reads awkwardly loses more in clicks than it gains in matching, and the resulting video often disappoints the people it attracts. Metadata application is covered in the video SEO checklist, and audience language in the audience research guide.
Common questions
Are third-party volume estimates reliable?
Treat them as rough ordering, not measurement. Their absolute figures are estimates built on limited samples.
Should every video target a search term?
No. Some of the best-performing video answers no query at all and travels through suggestion instead.
How often should research be redone?
Once or twice a year, plus whenever your analytics show unfamiliar terms arriving.
What about very competitive subjects?
Narrow the angle rather than abandoning the subject. A specific question is winnable where a broad one is not. See contact.
Does the same research apply to shorts?
Less so. Short-form reaches people through browsing rather than search, so phrasing matters far less than the opening seconds.
Should research drive the whole schedule?
No. A channel made only of searched-for topics tends to feel mechanical, and the videos people remember are often the ones nobody was looking for.