ProovenMethodology
Where the formats and figures published by Prooven come from, and how they are checked.
Which videos are analysed?
Public TikTok and Instagram (Reels) videos that reached a high number of views, most often over one million. They are found by keyword and by niche (business, health, content creation, lifestyle, education, and more). Only public data is used.
What is recorded for each video
- Public figures: views, likes, comments, shares and saves when the platform shows them, publishing date, length.
- The hook: the sentence spoken and the text shown in the first seconds, word for word.
- The structure: the video split into steps (hook, context, proof, payoff, call to action…), with what appears on screen at each step.
- The mechanics: the emotion targeted, what drives comments, shares or saves, and the filming format.
How the analysis is done
Transcription and breakdown are produced by AI models (Gemini and Claude), from the video itself and not just its caption. Results are stored in a database, then checked: inconsistencies are flagged and corrected.
Which formats become recommendations?
A format is suggested in the app only if it is reproducible: it has to work because of its structure, not because of the creator's fame, a filming budget or luck. Videos whose success mostly depends on the person in them are left out.
How figures are published
Every figure published on this site states its period and the set of videos it is based on. When a platform hides a statistic, it is marked as hidden and left out of calculations, never replaced by zero.
Limits
- The videos analysed are successes: they show what viral videos have in common, not a guaranteed result.
- Platforms often change their rules and displays. Figures are updated and dated.
- A video's real retention rate is only visible to its creator. Prooven does not make it up: it only infers it when public signals allow, and says so.
A mistake or a question about the method: antonin@prooven.fr. See also who is behind Prooven.