How to do YouTube keyword research, step by step

Most new creators do keyword research backwards. They film the video, open a tool five minutes before upload, and hunt for a phrase to paste into the title. By then the video is already made for nobody in particular, and the keyword cannot fix that.

This guide is the other way round. It is the full process I would follow before I write a single line of a script, from the first idea to the check you do a month after publishing. If you only want the tool that scores phrases for you, the free YouTube keyword research tool does that, and its page explains how each score is worked out. Here we focus on the method: what to do first, what to look at, how to decide, and what to do when the numbers disagree. I have been on YouTube since 2018 and blogging since 2010, and keyword research is the habit that carried over most directly from my blogging years.

What YouTube keyword research means

YouTube keyword research is the work of finding the words your viewers type into YouTube search, and choosing the ones your channel can rank for with a video that fully answers them.

Two parts of that sentence matter. The first is "your viewers". A keyword is useful only when the people searching it are the people you want on your channel. The second is "can rank for". A phrase that a million people search is no use to you if the first page is full of channels with ten million subscribers and videos from last week.

It helps to know what YouTube itself says about search. On its help page about how search works, YouTube lists three main elements: relevance, engagement and quality. For relevance it checks "how well the title, tags, description, and video content match your search query" (YouTube Help: How YouTube search works, checked 3 November 2026). So the keyword helps YouTube understand what your video is about. After that, how viewers react to the video decides how long it stays near the top.

In plain words: keyword research gets your video into the right search results. The video itself has to keep it there. That is why every step below looks at the viewer, and only some of them look at numbers.

Step 1: Check how much of your views come from search

Start by finding out how much search matters for your channel today, because it decides how much time keyword research deserves.

Open YouTube Studio, go to Analytics, and look at the traffic sources for your channel (in the Content or Reach part of Analytics, depending on your Studio layout). You will see sources such as Browse features, Suggested videos, YouTube search, Shorts feed and External. Look at the YouTube search line for the last 90 days and note its share of your views.

How to read your number:

• Search is one of your top sources. Your viewers come looking for answers. Keyword research should shape almost every video you plan.
• Search is small, Browse and Suggested are big. Your viewers find you on the home page and next to other videos. Keyword research still helps with titles and with a few evergreen videos, but topic and thumbnail ideas matter more. The outlier video guide fits this kind of channel better.
• You have no videos yet. Then you have no number, and that is fine. In my view, search is one of the few ways a viewer can find a brand new channel before YouTube learns who likes it. Treat keyword research as your main way to be found in the first months.

I am not giving you a magic percentage here, because I have not seen YouTube publish one and I would rather not invent it. Compare your search share with your other sources and let the biggest ones guide where you spend your time.

Step 2: Write 5 to 10 seed topics in your viewer's words

A seed is a short phrase of one to three words that names a topic your channel covers, written the way a viewer would type it.

Take a sheet of paper or a spreadsheet and write 5 to 10 seeds. Pull them from three places:

1. What you know. The topics you could talk about for an hour without notes.
2. What your viewers ask. Read the comments on your videos and on bigger videos in your niche. Questions in comments are often search phrases waiting to happen.
3. What your niche covers. If you have not picked a niche yet, do that first. My guide on how to find a YouTube niche walks through it.

Now the part most guides skip: write each seed in every language and script your viewers use. A viewer in Lahore who wants to earn online might type "freelancing", "online earning", "paise kaise kamaye" or the same phrase in Urdu script. Those are four different searches with four different sets of results. If your audience mixes English and Roman Urdu or Hindi, your seed list should mix them too.

Keep the seeds short. "Freelancing" is a good seed. "How I got my first client on Upwork after six months" is a video title, and it belongs later in the process.

Step 3: Turn each seed into a list of real search phrases

Expand every seed into 20 to 50 longer phrases, using sources that show what people type, and keep them all in one list.

Here are the free sources I would use, in this order:

1. YouTube autocomplete. Type your seed into the YouTube search bar and write down the suggestions. Then type the seed plus a space and each letter from a to z ("freelancing a", "freelancing b"), and then question words in front ("how freelancing", "what is freelancing", "is freelancing"). Do this in a private or signed-out window, because YouTube can personalise what it shows you based on your own history.
2. Studio's Trends tab. In YouTube Studio, open Analytics and then Trends, and search your seed. YouTube's help page says the Searched on YouTube section "shows top searches for the topic", and the Top searches card is based on your audience and your saved searches over the last 28 days (YouTube Help: Explore trends on YouTube, checked 3 November 2026). The same page notes some insights are limited to certain countries, languages and devices, and that an updated Studio layout started rolling out in July 2026, so yours may look different. My post on the YouTube Research tab explains which version you have.
3. Titles that already work. Look at what phrases recent videos on your topic use in their titles. The keyword research tool pulls these from video titles and groups them by angle. For phrases that are rising right now, check trending YouTube keywords.
4. Competitor tags. Tags tell you which spellings and variations other creators thought about. The tag extractor shows them for any video. Use them as ideas only. YouTube says tags "play a minimal role" in discovery (YouTube Help: Add tags to your video, checked 3 November 2026).

You will end up with a long, messy list. Do not judge it yet. Just remove duplicates and anything off topic.

Step 4: Sort the phrases by what the viewer wants

Every search phrase carries an intent, meaning the kind of video the viewer expects to get, and your video has to match it.

Go through your list and give each phrase one of five labels. This table shows the labels, a typical phrase for each, and the kind of video that fits.

Here is how to use the table. Say the phrase is "fiverr vs upwork". The viewer has already narrowed the choice to two platforms and wants you to help them pick. A video called "My freelancing story" with Fiverr and Upwork mentioned at minute 12 will lose them, even if the phrase sits in the title. A video that puts both platforms side by side in the first minute and gives a clear verdict will keep them watching, and that watching is the engagement signal YouTube talks about.

Some phrases carry two intents. "Freelancing in Pakistan" could mean "how do I start" or "how do I get paid". When you are not sure, the results page in the next step tells you which intent YouTube thinks is right.

Step 5: Study the results page before you commit

Search each shortlisted phrase on YouTube and read the first page of results, because they show you exactly what you would be competing with.

Use a private window again, and set the country and language your viewers use if you can. Then answer five questions for the first 10 results. The table below lists each question, what a good sign looks like, and what a warning sign looks like.

Read the table as a score out of five. Say you search "upwork profile tips" and find that four of the top ten videos are three years old, two are from channels under 20,000 subscribers, and three spend most of their time on Fiverr. That is three or four good signs: the phrase is worth a video. If instead every result is a fresh video from a huge channel and each one answers the question well, look for a narrower version of the phrase.

Two cautions. First, your results are personalised. YouTube says it may use your search and watch history to shape results, so two people can see different pages for the same phrase. Second, a single viral video at the top does not make a phrase impossible. What matters is whether small channels appear anywhere on the first page.

What our data shows: one seed, three phrases

To show how much the numbers change as a phrase gets narrower, I pulled three phrases from TubeMind's keyword data on 3 November 2026. Our index held 1,785,460 videos from 21,452 channels that day, collected through the official YouTube Data API. Each row counts indexed videos from the last 90 days with the phrase in the title.

Let me explain what this tells you, row by row.

"Freelancing" is the broad seed. It had 112 recent videos from 41 channels, and only 38% of those videos came from channels under 50,000 subscribers. The median video got 1,011 views, which means half did better and half did worse. Uploads on the word rose 87% compared with the 90 days before. That is a busy, growing topic, and the Difficulty of 51 says a small channel will have to work for a place on it. The Difficulty score is TubeMind's own measure; the keyword tool page explains how it is calculated and how to read it, so I will not repeat that here.

"Upwork" is one level narrower. It had 26 videos, but 62% came from small channels, the median was higher at 1,546 views, and Difficulty dropped to 28. Uploads on it more than tripled against the previous 90 days. Inside the "freelancing" titles, the Upwork group of 9 videos had a median of 8,914 views, the strongest of any angle. For a small Pakistani freelancing channel, that is the kind of sub-topic I would look at first.

"Freelancing for beginners" is the narrowest. Only 5 indexed videos used it, 80% from small channels, and Difficulty was 9. But 5 videos is a thin sample. A low score on very few videos can mean "open", or it can mean our index has not seen the bigger videos on that phrase. This is exactly when Step 5 matters: search the phrase on YouTube yourself before you trust the number.

Two more things stand out. First, format changes with the phrase. More than half of the recent "freelancing" and "upwork" videos were Shorts, but none of the "freelancing for beginners" videos were. People who type "for beginners" seem to expect a full walkthrough of around 9 minutes. Second, language is part of the keyword. Among the "freelancing" titles, 38% also used "earning" and about 29% used the Roman Urdu word "kamao". Those viewers are mixing English and Urdu in one search, and the titles that reach them do the same.

One last lesson from this data: about 16% of the "freelancing" titles also contained the name of one city, Sheikhupura, and 28% contained "seminar". A pattern like that usually means one or a few channels upload many similar videos. One busy uploader can fill a phrase. When a related word looks odd, open the videos and check who is behind it. Method and definitions are on how our scores work.

Step 6: Pick one main keyword per video and build a keyword map

Give every planned video one main keyword, and write the plan down as a keyword map so two of your videos never chase the same phrase.

A keyword map is a simple table: one row per video, with the main keyword, two or three close variations, the intent, and the format. Here is a hypothetical map for a new freelancing channel, built from the phrases above.

Look at how the map works. Each row has its own main keyword, so no two videos compete for the same search. The variations are phrases the same video can also answer, so you can use them once in the description or in a spoken line. The intent and format columns come straight from Steps 4 and 5. And the five videos together cover one topic from several sides, which is what a viewer who finds one of them wants to watch next.

How do you choose which row to film first? I would start with the phrase that scored best in your results page check, where small channels already appear. Save the broad phrases, like "freelancing" on its own, for later. Once you have a few videos ranking on narrow phrases, YouTube has more reason to trust your channel on the topic. My guide to micro niches follows the same idea at the channel level.

Step 7: Put the keyword to work, then check what happened

Use the main keyword where YouTube and viewers both read it, publish, and come back after 28 days to see which searches brought viewers.

The placement itself is short. Put the main keyword near the start of the title. Use it naturally in the first two lines of the description. Say it out loud in the first minute, so it is in the captions. Add a few tags for spellings and variations. That is enough. Repeating it five times reads like spam to viewers, and YouTube's help page warns that adding excessive tags to a description breaks its spam policies (YouTube Help: Add tags to your video). Before you publish, run the draft through the video SEO checker to catch anything missing.

Then the step almost every guide forgets: measure. After about 28 days, open the video in Studio, go to Analytics, then the Reach tab, and find the Traffic source: YouTube Search card. YouTube says this report "shows you what terms viewers searched when they found your content" (YouTube Help: Understand your video reach, checked 3 November 2026).

Compare that list with your keyword map. Three things can happen:

• Your main keyword is on the list. Good. Check your position for it with the YouTube rank checker and note it in your map.
• Other phrases brought the views. YouTube matched your video to searches you did not plan. Add those phrases to your map. They are often the easiest next video, because YouTube already links your channel to them.
• No search views at all. Check the title and thumbnail first, then the first minute. If the video does not deliver what the phrase promises, viewers leave and YouTube stops showing it.

My post on how to check where your video ranks covers the ranking side in more detail. The point here is the loop: your own search report feeds your next round of keyword research. After 10 or 15 videos, this report will often give you better keywords than any outside source, because it comes from your own viewers.

Keyword research for Shorts

Shorts need lighter keyword work, but the same logic applies: know what people want on a topic, then make the clip that answers it fast.

Most Shorts are found in the Shorts feed, where the viewer swipes rather than searches. So the keyword matters less for the first push. It still helps in two ways. Plenty of creators aim Shorts at search phrases: 57% of the recent "freelancing" videos in our data were Shorts. And YouTube's Trends tab has a card called Content gaps for Shorts, which it describes as Shorts search terms tied to topics your audience watched in the last 28 days (Explore trends on YouTube).

A simple routine works well: when you plan a long video from your keyword map, look for one "fix a problem" or "understand" phrase in the same topic and make it into a Short. Say your long video is "upwork profile tips". A Short answering "upwork profile rejected, what to do" covers a narrower search and points viewers to the full video.

Mistakes that waste your keyword research

These are the mistakes I see most often in the process itself. Each one makes good research useless.

1. Researching after filming. The keyword should shape the video, from the first line of the script to the thumbnail text. Picking it at upload time turns research into decoration.
2. Researching from your own signed-in account. Your suggestions and results are coloured by what you watch. Use a private window so you see something closer to what a new viewer sees.
3. Treating Google search volume as YouTube volume. People search Google and YouTube for different things. A phrase with large Google volume can be a phrase people want to read about, with little video demand.
4. Copying a competitor's keyword without a better video. Using the same phrase is fine. Making the same video with less experience is the problem. Find what their video leaves out and build yours around that gap.
5. Giving up after one week. Search views can build slowly as YouTube learns who your video satisfies. Wait for the 28 day check before you judge the keyword.
6. Ignoring the click. A keyword gets your video shown in search. The title and thumbnail decide whether anyone clicks. Check them against the top results before you publish.

When keyword research will not help much

Keyword research is a strong habit, but it is weak for some kinds of channels and some kinds of videos. Know where it stops working so you do not blame yourself.

• Entertainment and personality channels. Comedy, reactions, vlogs and challenge videos are mostly found on the home page and in Suggested. Few people type "funny vlog" into search. Topic, title and thumbnail matter far more here.
• Brand new topics. When something has just happened, people search before autocomplete or any tool catches up. A trend tool and the news will tell you more than keyword research in the first hours.
• Very small languages or regions. Autocomplete and tools have less data for small languages, so lists will be short. Comments and your own search report become your main sources.
• Tool numbers are estimates. No outside tool sees YouTube's private search counts. TubeMind's data, for example, counts videos and views in our own index, which covers part of YouTube. Use any score to compare phrases with each other, and confirm the winner with your own search on YouTube.

Sources and method

TubeMind keyword data for "freelancing", "upwork" and "freelancing for beginners" was pulled on 3 November 2026 for the last 90 days. Videos are indexed uploads whose title contains the phrase; small channels are those under 50,000 subscribers; trend compares uploads with the previous 90 days. The index held 1,785,460 videos from 21,452 channels that day, built through the official YouTube Data API. Definitions are on how our scores work. Official facts are from YouTube Help: How YouTube search works, Explore trends on YouTube, Understand your video reach and Add tags to your video, all opened on 3 November 2026. The top ranking guides for "how to do YouTube keyword research" were read the same day to find what they leave out.

Ready to score your own phrases? Run your seed through the free YouTube keyword research tool.

Research right on YouTube: the free TubeMind Chrome extension shows the outlier multiplier, a revenue estimate and similar channels on YouTube pages.