If you are using AI to write a video script, put the effort in the first line
We tested which parts of a script move reach. The opening line does, corrected for account size. Length and pace do not. Queried 28 September 2026.
What the numbers say
We measured three things a script generator can control and asked which of them actually moves reach, correcting every figure for account size first. The opening line survives that correction. Reels whose first line asks a question reached a median of 0.48 views per follower across 410 Reels, against 0.31 for openings that begin with the word how across 60, and 0.17 for openings that name their audience, the classic if you are a first time buyer shape, across 106. That is roughly three to one between the best and worst shapes. The body of the script does not survive it. Grouped by transcript length, the bands run 0.52, 0.47, 0.26, 0.47 and 0.42 across 2,946 Reels, which is not a pattern, it is noise. Grouped by speaking pace the bands run 0.46, 0.50, 0.43, 0.41 and 0.41 across 2,803, a gentle slope with nothing in it worth acting on.
This is a useful thing to know before you pay for words. A generator that writes you 200 tidy words and a flat opening has optimised the part of the script our data cannot distinguish from noise. Ask it for twenty openings and pick one, rather than asking it for one script and accepting the whole thing. And treat the audience-naming opening with suspicion: it is the shape most tools produce by default and the worst performer we measured.
Sample of 2,946, measured on 2026-09-28. Scrptly production database, queried 2026-09-28. Hook shapes measured across 3,219 Reels with a recorded opening line; script length across 2,946 with a transcript and a duration between 5 and 180 seconds; pace across 2,803 on the same basis. All normalised by the posting account's follower count.
The measurement
Opening lines grouped by their grammatical shape. Raw median views is shown beside views per follower because the two orderings differ, and the corrected column is the one to read. Naming the audience looks mid-table on raw views and is the worst shape once account size is removed.
| Opening shape | Reels measured | Median views per follower | Median views raw |
|---|---|---|---|
| Asks a question | 410 | 0.48 | 208,283 |
| Anything else | 2,107 | 0.47 | 177,887 |
| Leads with a number | 227 | 0.39 | 138,238 |
| Opens in first person | 309 | 0.32 | 84,568 |
| Starts with the word how | 60 | 0.31 | 53,950 |
| Names its audience | 106 | 0.17 | 106,424 |
One Reel, beat by beat
One Reel broken into its five beats: @rakshith_suresh_, 17 seconds, 2,703,659 views. Read the time column against the beat: the hook gets three seconds of the seventeen, and the payoff gets five.
| At | What happens |
|---|---|
| 0:00 to 0:03 | Hook: bold text claim set against a visual contradiction, a forced painful smile in the dark |
| 0:03 to 0:04 | Hard cut to black, interrupting to reset attention |
| 0:04 to 0:10 | Build: outdoor scene carrying stress and frustration, reinforcing the text |
| 0:10 to 0:12 | Setup reveal: the MEANWHILE text drops and names the other group |
| 0:12 to 0:17 | Payoff: a genuine happy expression reveals the viewer is that group, completing the irony |
What to do next
- Ask the generator for twenty opening lines and nothing else. That is the part of the script our measurements can actually distinguish.
- Delete any opening that begins with the words how to. It was the second worst shape we measured, at 0.31 views per follower.
- Be careful with openings that name the audience. If you are a founder, if you are a nurse, and so on scored 0.17, the worst of every shape, and it is the shape most tools reach for first.
- Do not let a tool talk you into a target word count or a target pace. Neither showed a usable relationship with reach once account size was removed.