What we research
CreatorsCompanion helps creators find posts that are travelling in their niche and write their own versions in their own voice. The same research features collect public metrics for the posts they find. We publish what those metrics show in aggregate, so creators can replace rules of thumb with measured numbers, and we state the sample and its limits next to every figure. Each study links back here.
Where the data comes from
Every study uses public post metrics only: views, likes, comments, shares, saves where the data has them, the account’s follower count, video length, caption length, hashtag count and publish time, plus a label for how each post’s first sentence opens, set by a fixed rule from text we never publish. They were collected through third-party data providers as part of the product’s research features. The figures in the current studies were computed with read-only queries; nothing in the source data was changed.
1. The search library
Posts found by keyword searches on TikTok, Instagram and YouTube: searches our users run, the niche monitors they schedule, and the builds of our niche library (20 categories). Each search keeps the posts with the most views after an automatic relevance check drops off-topic ones, and every post is stored once even when several searches find it.
2,692 posts: 2,002 TikToks (1,696 viral), 584 Instagram Reels (501 viral) and 106 YouTube videos (25 viral). The TikToks were published between September 10 and November 11, 2026, mostly in October; the Reels mostly between August and October 2026.
2. The older TikTok sample
TikToks found by our first viral finder, which ran keyword searches sorted by most liked with a United States region setting. Collected from November 2025 to March 2026: 3,229 unique videos (2,817 viral), of which 3,175 were published within 35 days of collection. We use it as a second, independent check.
3. Tracked creators
The full post timelines of creators our users follow with the product’s competitor tracking: 4,319 posts from 59 accounts, mostly on Instagram. We use them for anything that needs a creator’s own normal, such as what a typical Reel gets, which Reels broke out and how many likes and comments a Reel gets per view.
4. Repeated view checks
The view count of a tracked post, recorded at each scan: 7,286 checks of 3,079 posts. We use them to see how early views compare with a later count.
Sampling and its biases
- Search favours big posts. Platform search shows popular posts first, so our samples describe what search surfaces, not what a typical post gets. The sort matters too: the older sample, sorted by most liked, has a median viral TikTok of 2.12 million views against 803,000 in the search library.
- The niches follow our users. The searches are the ones our users and our niche library run, so business, marketing, fitness, real estate and money topics are over-represented, and most captions are in English. Results for one niche can differ a lot from the overall figures.
- A short window. Most posts in the search library are from a few months of 2026, mostly October for TikTok. We have not tested seasons or year-on-year change.
- Time zones. Publish times are exact instants, analysed in UTC. The search library does not record a creator’s time zone, so its times stay in UTC. The older sample records the account’s country, so the posting-time study converts those posts to local time with one zone per country (US accounts in Eastern time, checked again in Central time); for countries with several zones that local time is approximate. Tracked creators’ times are in UTC.
- Metrics have different ages. A post’s views are read when a search finds it. Viral TikToks were read within their first 30 days (116 under 2 days old, 777 at 14 to 30 days); most viral Reels were read more than 30 days after posting (379).
- The definition favours small accounts. Passing your follower count is easier when the count is small.
- Tracked creators were chosen by our users. They are mostly large accounts, and their follower counts are today’s, not the count on the day each post went up.
- Platform counting. Instagram views are play counts as publicly reported, and Instagram has changed how it counts views over time. A looping short Reel counts each replay as a view.
Rules we apply before publishing a number
- Every reported group has at least 30 posts; in the posting-time study, so does every weekday and every 3-hour block on its own. Percentiles (medians, middle halves) need at least 100 posts, and a figure computed per creator needs at least 30 creators. A comparison within the same creators needs at least 30 posts and at least 10 breakouts per group. One result falls below the 100-post line for percentiles: how much of a tracked Reel’s later view count is already there in its first days, which has fewer than 100 Reels in most rows and is marked directional on its page.
- Shares and rates come with Wilson 95% intervals. Within-creator comparisons are observed breakouts over expected ones, where the expected count adds up each post’s own creator’s usual breakout rate, with a Byar 95% interval; we say when an interval includes 1, meaning no clear difference.
- When we test whether two groups differ, we only call a difference real at p < 0.01. A difference in mix (weekdays, hook forms, caption bands) uses a chi-square test, run only when every expected cell holds at least five posts; a difference in a rate or a length uses a Mann-Whitney rank test, with the chance that a random post from one group beats one from the other.
- Anything under these thresholds is left out, even when it would make a better headline. Some results are marked directional on the page when the sample is small or measured unevenly.
What we never publish
The studies are aggregates. We never publish a handle, display name, profile picture, post link or id, caption, transcript, hook text, on-screen text, thumbnail or bio, never a row about a single account or post, and never the list of creators our users track. We do not publish anything about who searched for what. Creators are people, so even public numbers about them are personal data: an exact view count, follower count and date together can point to one post, which is why every figure is rounded and no study shows a “top post” example.
What we left out, and why
- YouTube: only 25 YouTube videos in the search library passed the viral line, under our minimum.
- Per-niche numbers: each niche category comes from a small number of searches taken once, so its figures move too much to publish. To see your own niche, use the free niche check.
- A best time to post down to the day and hour, or by niche: split that finely, no group reaches 30 posts. When viral posts went up mostly shows when creators post, so the posting-time study publishes weekdays and 3-hour blocks only and compares viral posts with the posts that did not go viral and with each creator’s own breakout rate.
- Posting frequency and follower growth: too few cases to say anything reliable.
- Hook forms with fewer than 30 posts in a group, such as how-to and negative openings among viral TikToks: they print as under 30. Engagement for Reels found by search: Instagram search results carry no like, comment or share counts, so Reels engagement comes from tracked creators only.
Rounding
Views and follower counts are rounded to three significant figures (so the median viral TikTok prints as 803,000 views). Percentages are whole numbers from 10%, with one decimal between 1% and 10% and two significant figures under 1%. Multiples and ratios have two significant figures, seconds, characters, words and days are whole numbers, and interval bounds have two decimals as computed. Engagement rates follow the percentage rule; a Reel’s rate over its creator’s usual rate keeps two decimals, like a lift. Breakout rates in the posting-time study keep one decimal, because the gaps there are fractions of a point. A p-value has two decimals, one significant figure under 0.01, and anything smaller than 0.001 prints as p < 0.001. Sample sizes are exact. Each study’s CSV download lists every printed figure with the result file, row and field it comes from.
Updates and corrections
Each study shows when its figures were computed and when the page was last changed. When we refresh a study with new data, we update both dates and say what changed. If you find an error, email support@creatorscompanion.app; we will check it, correct the page and date the change.
The current studies: how many views is viral, viral video length, the best time to post and TikTok engagement rate, hooks and captions.
See what is viral in your niche
These numbers describe many niches at once. To see the posts beating their creator’s following in yours, type your niche: the free niche check ranks recent TikToks and Reels by views against followers, the same test this research uses. It needs no account.
How to cite this
CreatorsCompanion, “Research methodology: how we study viral posts,” published , https://www.creatorscompanion.app/research/methodology
You are welcome to quote these numbers with a link to this page. Spotted an error? Email support@creatorscompanion.app and we will correct the page and date the change.
More research
- How many views is viral? What 1,696 viral TikToks and 501 viral Reels show
- Viral video length: how long viral TikToks are, and which Reels break out
- Best time to post on TikTok and Instagram: does timing decide what goes viral?
- TikTok engagement rate: what viral TikToks get per view, and how their hooks and captions compare