Someone searches for how to freeze the top row of a spreadsheet. The video's thumbnail shows exactly that, but the author spends the opening minutes talking about their experience, and the command only appears near the end. Search brought the viewer to the video, and then the video made them hunt for the answer.
Take an online analytics school. Yerasyl, one of its teachers, records a short explainer using a lead spreadsheet: the viewer should learn to keep headings visible while scrolling and get to know Yerasyl as a teacher along the way. All numbers in this lesson are illustrative.
Start with the viewer's action
Before recording, Yerasyl defines the result: the viewer freezes the top row and sees for themselves that the headings stay put while scrolling. If the steps depend on the editor, device or version, say which. The lesson names no commands on purpose: take them from the interface you are actually recording.
It is easiest to build the script around one spreadsheet. First scroll it and show the headings disappearing. Then perform the action and scroll the sheet again. The viewer sees the problem, the fix and the check on the same data.
The video promise and its demonstration
Title promise
Establish the editor, device and version before recording.
- Demonstrate on the same data
- Lead-tracking spreadsheet
Demonstration script
- Task
Scroll the sheet to show the headings disappearing.
- Action
Freeze the row. Take the command name from the interface being recorded.
- Check
Scroll the sheet again: the headings stay visible.
Cursor, narration and picture address the same step. Captions do not obscure the action.
The cursor, voice and picture all show the same step. Say the command's name at the moment it is visible on screen. Zoom in on the important area and place subtitles where they do not cover the action. Fix mistakes in feature names before publishing: viewers repeat what they hear.
If the task needs any preparation, say so at the start. Cover an alternative method or a common mistake after the main result. Let the task set the length: a drawn-out demo and a rushed one both make the action hard to repeat, just in different ways.
Match the query, promise and content
In search, the person has already asked their question. YouTube names relevance, engagement and quality as the main elements of ranking. How well the title, description and content match the query matters, and the user's history can also affect results. So the position the author sees on their own screen need not match what a future viewer sees.
A title such as "How to freeze the top row in [editor]" names the action and the environment. The square brackets hold the place for the editor's real name and do not stay in the published title. The thumbnail shows a result the video actually delivers. Promising to fix all your reporting would go beyond what is shown.
The description helps clarify conditions, find the right segment and move on to the next material. Put the link to the next lesson after the current task is solved. If someone only needs a frozen row, the free explainer should get them there without requiring a course purchase.
YouTube treats tags as secondary, except for words that are often misspelled. As metadata, the title, thumbnail and description matter more. A long tag list will not rescue a video that promises one thing and shows another.
After publishing, look at the queries people actually arrive with. If they are looking for how to freeze a column while the video only shows a row, Yerasyl decides whether to record a separate video or narrow the promise. Driving more clicks without fixing the mismatch just sends more people to an answer that does not fit.
Separate viewing sources
Search, a click from another video and an external link are three different situations. Someone from search is already looking for the action, while a viewer of a neighbouring video may not even know why they need it. Crediting every view to search optimisation just because the title contains a key phrase is a mistake.
YouTube impressions CTR relates views to counted thumbnail impressions. Not every view comes from such an impression. So first work out what is in the numerator and the denominator, and only then compare percentages.
The first table includes only views from counted impressions for each source; everything else is left out.
| Source | First-period impressions | Views from those impressions | CTR |
|---|---|---|---|
| Search | 1,000 | 80 | 8% |
| Browse features | 5,000 | 150 | 3% |
| Total | 6,000 | 230 | About 3.83% |
Suppose another 70 views came through external links, with no counted thumbnail impressions. The overall counter would then show 300 views. Dividing 300 / 6,000 gives 5%, but the numerator is now a different one, and that 5% has nothing to do with the CTR of counted impressions.
Lower overall CTR can hide unchanged source results
In the second period, search produced the same 1,000 impressions and 80 views. Browse features produced 15,000 impressions and 450 views. Within each source the rates are still 8% and 3%. Yet the overall figure is now 530 / 16,000 ≈ 3.31%.
CTR fell from about 3.83% to 3.31% even though nothing changed within the sources. The total simply contains more impressions from the source with the lower CTR. Meanwhile views from counted impressions rose to 530 from 230.
The calculation shows that the impression mix changed. It cannot tell you why the platform started showing the video more widely, or whether the thumbnail works equally well for different audiences. Keep the source, period, volume and every change to the publication for the analysis.
Changing titles one after another does not give equal conditions either: demand and audience may shift in the meantime. In a report on such a comparison, describe its limits and do not declare every increase a proven effect of the new word. Equal conditions come from A/B testing in YouTube Studio: up to three titles or thumbnails run in parallel, and the winner is picked by share of watch time, not by CTR.
Watch the disputed segment alongside the graph
The retention graph points to the segment worth rewatching. One line cannot establish the cause: the person may have got the answer, skipped a familiar step or failed to spot the command. Yerasyl compares that stretch with the queries, comments and the recording itself.
| Observation | Possible explanation | What to inspect in the video |
|---|---|---|
| Leaving before the demonstration | The answer takes too long to arrive | Introduction and first action |
| Repeating one step | The step is useful or unclear | Interface scale, narration and speed |
| Leaving after the result | The task is complete | Whether the viewer has a next need |
In a row-freezing video, leaving right after a successful scroll may be normal: the task is done. Stretching the intro for watch time makes no sense, because it only pushes back the help you promised. Instead, ask someone who does not know the method to repeat the action from the recording. The point where they stumble gives the video editor a clearer task than a retention percentage.
Calculate production cost and the linked result
The school sells a course for 25,000 ₸, with 15,000 ₸ contribution after variable teaching costs and before acquisition. It spent 60,000 ₸ preparing the video, by its own budget. That is a condition of the exercise and says nothing about market production prices.
Over an agreed period that has already closed, the team linked 6 non-refunded payments to the video. Revenue from them was 150,000 ₸ and contribution before production 90,000 ₸. After 60,000 ₸ for the video, 30,000 ₸ is left before other expenses. Keep repeat touchpoints, but count each payment once in the overall report.
For the video to pay back, it needs 60,000 / 15,000 = 4 additional payments. Six linked payments do not mean the video added all six: some of those people may already have been choosing the school. So show the observed result and the estimate of added sales separately, and the estimate needs a suitable comparison.
The video may keep getting views after the chosen period, but future purchases are not facts yet. Keep the publication date, costs and payment history. Then you can discuss the next videos from a known result and clearly stated expectations.
Exercise: should the thumbnail change?
The school's owner sees CTR drop from 3.83% to 3.31% and suggests bringing back the old thumbnail. Write them a short explanation based on the two periods. Then pick one segment of your own video that is worth testing for clarity.
In the example, the rates within sources did not change: search still gives 8% and Browse features 3%. The second source gained weight. A drop in the overall figure alone does not show that the thumbnail got worse. If you still want to compare thumbnails, put them into an A/B test in YouTube Studio, which picks the winner by watch time, not by CTR. While you are at it, check that the video keeps the thumbnail's promise.
For the video itself, do this: open the row-freezing moment and repeat the action without the author's prompts. If the command is not visible or scrolling never shows the result, you have found what to fix. That fix is for the viewer, while the video's payback stays in its own calculation of four additional payments.
Video: a creator's view on promotion
Listen to how the author explains promotion and compare his advice with your own recording: which query it answers, where it keeps its promise and which observations can actually change the next release.