The opening playbook: what 1,666 AI creator posts put in line 1 and on slide 1

The first-line template, 9 cover formulas ranked by results, the last slide, and the sorting prompt for your own niche. From 1,666 public Instagram posts by 24 accounts in the AI niche, collected on 10 Oct 2026. As of October 2026.

The data in one paragraph

I collected the last posts of 24 accounts that teach AI on Instagram: 1,666 posts, 340 of them carousels. It cost $4.38 of a free plan's monthly credit. Then Claude read every caption and the top 60 carousels slide by slide. "Score" below means likes plus 3 times comments, divided by that account's usual post. So 1.0 is a normal post for that account and 2.4 is two and a half times better.

Line 1: the first-line template

714 of the 1,666 captions start with the same word: Comment. One account starts all of its last 80 posts that way.

What comes right after the word:

After the word	Posts	Real example (first line)
...and I'll send you [thing]	304	Comment "CRM" and I'll send you the repo plus the full setup guide.
...to get [thing]	200	Comment "GOOGLE" to get this new AI Mouse Pointer from Google Deepmind.
...for [thing]	188	Comment "Free" for stack
something else	22	Comment AGENT if you want access

Copy one of these and fill the blanks:

Comment "[WORD]" and I'll send you [the exact thing, with a number].
Comment "[WORD]" to get [the exact thing].
Follow and comment "[WORD]" and I'll send you [the exact thing].

The rules the data shows:

One short word. The median keyword has 5 letters. The most used: AI (91 posts), AGENT (49), CLAUDE (29).
In quotes. 571 of the 714 put the word in quotes, so people see exactly what to type.
Name the thing. Of the 714 first lines, 201 mention a link, 162 a guide, 87 a skill, 83 a setup and 64 a prompt.
Short. The median first line is 9 words.
Send it for real. The word only works if the thing exists and arrives.

Does the word pay? The honest numbers

All 1,666 posts: the ones that open with "Comment" have a median of 519 comments. The others: 79.
Inside the same account: 14 of 16 accounts got more comments on posts that open with the word.
Likes: more in only 9 of 15 accounts (one account hides its likes).

So the word brings comments, and many of those comments are just the word. It doesn't make people stop scrolling. That happens earlier, on slide 1.

Slide 1: 9 cover formulas ranked by results

From all 340 carousel covers. The median score of covers with each formula (one cover can have more than one):

Clone a famous name or "build your Jarvis": 5.06 (11 covers). "How to clone Hormozi inside of Claude. (One prompt. He runs your whole business.)"
Free, open source or "before it's gone": 2.42 (49 covers). "HOW TO USE FABLE 5 BEFORE IT'S GONE"
Connect a tool you know to Claude: 2.26 (30 covers). "I CONNECTED CLAUDE TO TRADINGVIEW. Now it analyzes every chart."
How to build it in N steps: 1.88 (47 covers). "HOW TO BUILD AN AI MEDIA BUYER · 1 MCP. 4 SKILLS."
Money for a period: 1.79 (34 covers). "Claude can make you $453 a day on YouTube." I don't use this one without my own real numbers.
I built, I connected, I replaced: 1.25 (39 covers). "I replaced my Meta Ads Agency with 5 Claude Code skills."
N things, a list: 1.09 (58 covers). "5 CLAUDE MCP that are actually useful."
News, "just dropped": 0.63 (37 covers). "GOOGLE JUST DROPPED 3 NEW GEMINI MODELS"
Picking a fight ("sucks", "fake", "stop"): 0.55 (40 covers). "CLAUDE SUCKS AT WRITING CONTENT"

Length. Covers of 6 words or fewer: median 0.60. 7 to 10 words: 1.01. 11 to 16 words: 1.01. 17 or more: 1.27. A short label ("CLAUDE CODE PLUGINS GUIDE") loses. A sentence wins: who or what, a verb, a result. In the top 60 carousels the median cover is 11 words.

The first five words. In a bigger set (3,558 posts, 43 accounts), posts that open with filler ("So", "Today", "In this video") did worse: median 0.88. Put a known name, a number or some tension in the first five words.

Fill-in for slide 1:

[Known tool] + [what I did] + [a number]. Then a second line with the proof or the recipe.
Example of the shape: "I connected [tool] to Claude. Now it [does the job]."

The last slide: say exactly what they get

54 of the top 60 carousels end with a Comment ask. The strong ones list the thing, not "more info":

"I'll send you the links to all 7 and exactly what to charge for each one."
"the audit worksheet, the grading rubric, the folder structure"

The sorting prompt: do it for your own niche ($0)

Pick 50 posts from accounts in your niche that you'd like to be like.
Copy the first line of each caption by hand into one note. No tools to buy.
Below the list, paste the first lines of your own last 10 posts.
Paste this into Claude (or ChatGPT):

Paste this

Below are the first lines of 50 posts from my niche, one per line, and then the first lines of my own last 10 posts. Group the 50 by how they open (the first 3 to 5 words). For each group give me: how many posts, one real example from the list, and the shape as a fill-in template, like Comment "[WORD]" and I'll send you [thing]. Then tell me which shape my own 10 posts never use. Only use lines that are in my list. Don't invent examples.

The honest part

This is 24 accounts in the AI niche in October 2026. Your niche may open differently. That's why the prompt is there.
The score measures likes and comments, not reach or saves. Comments count three times, so "Comment WORD" pushes it up.
Copying a formula without the trust behind it fails. One account copied a top cover almost word for word and got 0.2.
Paid partnerships may be in the data. I flagged the ones Instagram labels.

Source: my competitor dataset (1,666 posts, collected 10 Oct 2026) and my notes on the top 60 carousels (9 to 10 Oct 2026).