YouTube Data Extraction: Video Metrics for Influencer & Content Research
YouTube data scraping turns a public video listing into a table you can sort, filter and chart. For influencer teams and content strategists, that table answers practical questions. How long should a review run? Do short titles perform better? Which creators are gaining views right now?
To show what this looks like in practice, we collected 100 public tech review videos and published them as a free file: the YouTube tech review videos sample dataset. This post walks through what the data contains, what it tells us, where it falls short, and how to use a YouTube video dataset in real research.
Why YouTube video data matters
YouTube is where many buyers research a product before they purchase it. A review that reaches a few million viewers shapes opinion in a way a product page cannot. That makes video metrics useful for three groups:
- Influencer and partnership teams who need to compare creators on reach, not on gut feeling.
- Content teams who want to know which formats, lengths and title styles tend to get views in their niche.
- Brands and analysts who track how often their products are reviewed and how much attention those reviews get.
Doing this by hand means opening dozens of tabs and copying numbers into a spreadsheet. A structured youtube video dataset replaces that work with one file. It also makes the analysis repeatable, so you can run the same comparison next month on fresh data.
What our YouTube sample shows
We collected 100 unique videos from public YouTube search results for smartphone, laptop, gadget and tech review queries. Each row has a video ID, title, relative publish age, duration in seconds, view count, video URL and thumbnail URL. Here are the first five rows:
| video_id | title | published_relative | duration_sec | view_count |
|---|---|---|---|---|
| c347oYQO57A | So This is Peak Smartphone | 6 months ago | 797 | 6000000 |
| bYZhnS-qWic | Apple is DONE? China’s New Phones Are Insane | 1 month ago | 729 | 4099999 |
| Hi2tjMLVpdQ | Which Smartphone Do They ACTUALLY Use? --- MKBHD, Austin Evans, Linus + More | 8 years ago | 479 | 8700000 |
| x-0gefsoDr0 | Why I STOPPED Chasing the Latest Smartphones | 3 months ago | 315 | 72000 |
| zPAY2VxfFBk | The Problem with these Smartphone Batteries | 8 months ago | 512 | 2300000 |
Here is what we computed from the full 100 rows:
- Typical video length is about 13 minutes. The median duration is 795.5 seconds, and 43% of the videos (43 of 100 records) run 15 minutes or longer.
- Views are heavily skewed. The median video has 1,250,000 views, while the mean is 3,547,243. The highest is 28,000,000 and the lowest is 842. A handful of hits pull the average far above the typical video.
- Length alone does not predict views. The correlation between duration and the logarithm of views is 0.03, which is effectively none.
- The 10 to 15 minute band had the highest median views. Its 34 videos had a median of about 4.05 million views. The 15+ minute group (43 videos) had a median of 917,000, and the 5 to 10 minute group (19 videos) had 338,000. Only 4 videos were under 5 minutes, with a median of 9,450.
- Shorter titles sit with higher view counts. The median title is 44 characters. The 33% of videos with titles under 40 characters had a median of 4.6 million views. The 51 videos with 40 to 59 characters had 1.2 million, and the 16 videos with 60 or more had 64,500. The correlation between title length and log views is -0.50.
- The sample leans recent. By YouTube's relative labels, 58% of the videos were published "months" ago, 17 "days" ago, 14 "years" ago, 7 "weeks" ago and 4 "hours" ago.
How to read these numbers
Be careful with the title-length result. It is a pattern in this sample, not proof that shorter titles cause more views. Large channels may simply favor punchy titles, and a big audience lifts views whatever the title says. The same applies to the duration buckets: the 10 to 15 minute band contains several very popular uploads that raise its median.
The sample also mixes several creators and comes from search results, so it is not the full upload history of one channel. It is a snapshot designed to show the shape of the data. For a rigorous test you would pull a whole channel or a larger set of videos and compare like with like, for example only videos from channels of similar size.
Video length vs views: how to test it properly
The angle most content teams ask about first is duration. To test it on your own data:
- Convert duration to seconds, as in the
duration_seccolumn, so it can be sorted and bucketed. - Group videos into bands such as under 5 minutes, 5 to 10, 10 to 15, and 15 or more.
- Compare the median views per band, not the mean. A single viral video can distort a mean badly, as the gap between 1.25 million and 3.5 million above shows.
- Repeat the comparison within one channel, then across channels of similar size.
If the pattern holds inside a single channel, you have a stronger signal that length matters for that audience.
Title length and publish timing
Title length is easy to measure and easy to act on. Count characters, bucket them, and compare medians as we did above. Pair that with a look at title style: question marks, numbers, all-caps words and brand names in the first few words.
Publish timing is the other classic question: which weekday or hour works best? Our sample records YouTube's relative age label ("2 weeks ago") rather than an exact date, so we cannot answer a day-of-week question from it. Exact publish dates are available on individual video pages, and a full dataset can include them. With exact timestamps you can group by weekday and compare median views for videos of similar age, which matters because older videos have had longer to collect views.
Which fields matter and why
- video_id is the stable key. Titles change, IDs do not. Use it to deduplicate and to join datasets across time.
- title supports length analysis, keyword tracking and topic grouping.
- published_relative gives a rough age. Always compare views against age, since a month-old video and a six-year-old video are not equal.
- duration_sec supports length bucketing.
- view_count is the main attention metric. In listings YouTube shows abbreviated figures, so large numbers are rounded.
- video_url and thumbnail_url let your team review the creative behind a result.
- scraped_at_utc records when the number was read. Views move, so a snapshot without a timestamp is hard to trust later.
Likes and comments are the fields teams usually want next, because engagement rate (reactions per view) separates an audience that watches from one that responds. Search listings do not expose those figures, so this sample leaves them out. A full channel-level dataset can include them.
Use cases for a YouTube video dataset
- Influencer shortlisting. Rank creators in a niche by median views per video rather than by subscriber count, then check how consistent they are.
- Content planning. Find the length and title patterns that cluster among the top videos in your topic.
- Competitor monitoring. Track how often rival products appear in reviews and how much attention each video gets.
- Campaign measurement. Snapshot a sponsored video's views on a schedule to see how quickly it gathers attention.
- Trend spotting. Watch which topics, such as a new phone launch, produce a cluster of high-view videos in a short time.
If you work on other platforms too, our guides to extracting data from Instagram and extracting data from Twitter/X cover the same influencer data workflow for those networks.
Common challenges in YouTube data scraping
Dynamic pages. YouTube loads much of its content in the browser, so a simple HTTP request often returns an incomplete page. Reliable collection needs rendering or a structured source.
Anti-bot measures. Repeated requests from one address get throttled or blocked. Production collection needs rotating infrastructure and sensible request pacing.
Pagination and result limits. Search listings return a limited number of results per request, and different queries overlap. We merged several queries and deduplicated by video ID to reach 100 unique videos.
Rounded and relative values. Listings show "1.2M views" and "3 weeks ago". These are fine for ranking and bucketing, but not for exact calculations. Know which of your fields are approximate.
Layout changes. YouTube changes its markup and internal data structures regularly. A parser that works today can break next quarter, so monitoring and maintenance are part of the job.
Data quality. Reposts, compilations and re-uploads can sit alongside original reviews. Filter by channel and check titles before drawing conclusions.
How often should you refresh?
It depends on the question. For influencer shortlisting, a monthly refresh is usually enough. For launch tracking, such as a new phone release, daily or even more frequent snapshots show how views build over the first week. For campaign measurement, capture the same video at fixed intervals (day 1, day 3, day 7, day 30) so the curves are comparable. Always keep the scrape timestamp.
FAQ
Is it legal to collect YouTube video data? We only collect publicly visible information, and we do not collect personal data. Legal questions depend on your use case and jurisdiction, so review our guide on web scraping legality and take advice for your situation.
Can I get likes, comments and exact publish dates? Yes, a custom dataset can include them where they are publicly shown. This free sample contains only the fields that search listings expose.
Why are the view counts rounded? YouTube shows abbreviated figures such as "28M views" in listings, so the sample reflects that precision. Exact counts are available on individual video pages.
Can you collect a specific channel instead of search results? Yes. A channel's video tab can be collected in full, which gives you a cleaner, like-for-like dataset than a mixed search sample.
Get YouTube data for your own research
Download the free YouTube sample dataset to see the structure and test your analysis. When you are ready to go beyond the sample, request a free sample of your own target and we will show you what your channels or keywords look like as clean data. Datasets start from $49, see pricing.
Author: DataHarbor Team