Instagram research

Find posts that beat their own account's normal performance, not posts that are merely big. A 3x post from a 5k account teaches more than a 1x post from a 5M account.

Limits (account safety)

• At most 12 accounts and 50 posts per account per run. The CLI enforces both and spaces requests at least 1.5 s apart.
• One research run per session is plenty. On rate_limited or checkpoint, stop, report the hint, and do not continue with the remaining accounts.
• Research is read-only: never follow, like or comment on researched accounts.

1. Pick the accounts

Ask for 5 to 12 accounts in the same niche, ideally with similar or slightly larger audiences. If the user has none, find candidates (run from ../instagram):

npx -y bun scripts/main.ts search "<niche keyword>" --json
npx -y bun scripts/main.ts hashtag <tag> --limit 24 --json     # top and recent posts for a tag

Confirm the list with the user before pulling posts.

2. Build the swipe file

npx -y bun scripts/main.ts swipe user1,user2,user3 --limit 24 --min 3 --json > "<project-dir>/instagram-swipe.json"

The output has outliers[] (username, url, format, multiple, engagement, and hook, which is the caption's first line), posts read per account, and skipped (private or missing accounts). An account with fewer than 6 posts read is too thin to judge. Say so.

3. Reverse-engineer the winners

For each outlier, highest multiple first:

1. Classify the hook against ../instagram-content/scripts/hooks.json (formula id and name), and score it with npx -y bun ../instagram-content/scripts/hook-score.ts --hook "<line>".
2. Note the format, caption length, call to action and topic.
3. When needed, read the post (npx -y bun scripts/main.ts media <url>) and its top comments (npx -y bun scripts/main.ts comments <url> --limit 20), from ../instagram. Comments show why it spread: questions, friends tagged, disagreement.

Then aggregate: which formulas and formats appear most among the outliers, and which topics repeat across different accounts. A topic that wins on 3 or more accounts is a strong signal.

4. Output

SWIPE FILE · 8 accounts · 176 posts read · 14 outliers (>= 3x own median)

PATTERNS
- Hook formula #5 Time Collapse: 5 of 14 outliers, across 4 accounts.
- Carousels: 9 of 14 outliers, though only 40% of posts read were carousels.
- Topic "pricing mistakes" wins on 3 accounts.

TOP OUTLIERS
6.2x  @acct  reel      "Proposals used to take me five hours..."   <url>
...

IDEAS FOR YOU (adapted, not copied)
1. ...

Save it as instagram-swipe.md in the project when the user wants to keep it. instagram-content reads that file. Adapt the structure, never the words: never copy captions, scripts or footage.

Additional resources

• ../instagram-content/scripts/hooks.json: the 26 hook formulas used for classification.
• ../instagram/references/endpoints.md: what the read commands call, and their limits.