---
name: carousel-copy
description: Specialist drafter for Instagram carousels — picks the optimal slide count + playbook for the topic and writes copy engineered for virality (saves, shares, completion rate). Use when Artem says "carousel", "draft a carousel", "make this a carousel", "turn this into slides", or when the source material clearly fits the format (lists, frameworks, tutorials, roadmaps, "X tools/skills/steps", systems). Pulls from content/carousel-bible.md (10 commandments, 10 playbooks, magic-8, slide-3 danger zone, house style) and content/my_carousels.md (Artem's top performers ranked by views). Outputs full carousel copy with hook variants, per-slide intent, virality design, and a tone check.
---

# Carousel Copy

The specialist for IG carousels. Carousels in Artem's data have **74% higher ER than reels** (10.9% vs 6.3%) and are massively underused. Every carousel that ships should be engineered — not drafted.

For reels/tweets/threads → use `short-form`. For YouTube/articles → use `long-form`. For Skool/Discord/courses → use `community-and-courses`.

## Source of truth

Load these in order, every draft:

1. **`../../../content/carousel-bible.md`** — the synthesized operating layer. 10 commandments, 8 universal agreements, 10 carousel-type playbooks, slide-3 danger zone, magic-8, screenshot test, How-I vs How-To, hook formulas, content slide templates, captions, hashtag strategy, pre-publish checklist — **plus the "Artem's house style" section extracted from his actual top-performing carousels.** Single biggest source of truth.
2. **`../../../content/my_carousels.md`** — Artem's actual carousels ranked by views (100k → 25k). This is the ground truth for what already worked. Match the patterns documented here (CAPS section labels, two-line headlines, bottom-arrow continuation cues, named-tool credibility, dollar/time anchors, "I [result]" or "You [hidden mechanism]" hooks).
3. **`../../../content/transcripts.md`** — Artem's reel transcripts. Useful for hook mining — top reels often convert into carousel covers (Bible §5.1).
4. **`references/slide-count.md`** — slide count picker mapped to the Bible's 10 carousel-type playbooks.
5. **`../../../content/carousel_bible_from-pdf.md`** — the raw 49-page meta-review markdown. Consult for deeper nuance or source quotes when carousel-bible.md is ambiguous.

**Do NOT pull voice rules from `content/DNA.md` or `content/feedback-log.md` for carousels.** Those were tuned for reels — the cadence is different. The carousel voice ground truth lives in `my_carousels.md` and is summarized in the "Artem's house style" section of `carousel-bible.md`.

## Process

1. **Read carousel-bible.md + my_carousels.md.** Don't skip — the Bible's 10 commandments and Artem's actual top-performer patterns are load-bearing.
2. **Pick the carousel-type playbook** from the Bible's 10 (Tip, Story, Curation, Same-Meme, Quote, Postcard, EDC, Two-Slide, Brand Story, Gamified). State which and why. Each has a slide-by-slide skeleton.
3. **Pick slide count.** Default to 8 (Magic 8: hook + 6 content + CTA). Use `references/slide-count.md` for the topic-type override.
4. **Decide TOFU or BOFU.** TOFU (broad, growth, "Follow for more") vs BOFU (narrow, problem-solving, "Comment KEYWORD"). State which. Affects CTA + hook angle.
5. **Pick the hook category** from the Bible (Challenge / Mistake / Listicle / Story / Curiosity Gap / Authority) AND apply the **How-I > How-To** rule when credibility allows.
6. **Draft the cover (S1)** with **3 hook variants**. 3-7 words each. Run the hook checklist on each. Pick the strongest, but show all three so Artem can swap.
7. **Draft S2 — second-chance hook.** IG re-shows S2 to people who skipped S1, so it must stand alone AND continue the curiosity. Don't restate the cover. Reveal something. Most carousels die between S1 and S2.
8. **Front-load value at Slide 3.** This is the algorithm checkpoint. Don't hide the densest tip on S8. If they read S3 and bail, they got the value, the algorithm got the dwell signal.
9. **Draft middle slides.** One breath, one idea per slide. Max 2-3 short lines. Each slide must pass the **screenshot test** — could someone share JUST this slide to their story? If no, revise.
10. **Draft penultimate (the payoff).** The "oh shit" / aha moment. This drives shares.
11. **Draft last slide (CTA + reward).** ONE specific action. "Comment [KEYWORD] and I'll DM you [deliverable]" beats "follow for more". Match CTA to goal (see Bible's CTA-by-goal table).
12. **Write the caption.** Pick ONE strategy: Inside Joke / Expansion / Question / Story Hook. Length matches type (short for value, medium for story). Don't repeat the slides.
13. **Add 5-10 hashtags** with varied sizes (3-4 niche, 2-3 audience, 0-1 branded). Avoid hashtags with 1M+ uses.
14. **Flag the audio requirement** in the output. The carousel-design skill handles the actual attach, but copy output should remind: *"Add audio post-export — without it, only existing followers see this."*
15. **Run the virality design check** (template below) — name which slides drive saves, shares, comments, completion.
16. **Run the anti-AI filter.** Cut every line that trips the house-style red flags from `carousel-bible.md` ("Voice rules" + "Dial down" list).

## The 4 virality drivers (each carousel must design for all 4)

| Driver | What feeds it | Where it lives |
|---|---|---|
| **Saves** | Density + reference value ("I'll need this later") | Middle slides — high signal, scannable |
| **Shares** | Surprise + hot take + "you have to see this" | Cover + penultimate slide |
| **Comments** | Keyword CTA with specific deliverable | Last slide + caption |
| **Completion rate** | Each slide earning the swipe | All slides, but middle is where drop-off happens |

If a carousel has 0 of these designed in, it's a vanity post.

## Hook checklist (run on cover S1 + every variant)

From the Bible's 5 cover-slide elements + Artem's house-style hook patterns.

- [ ] **3-7 words max.** Above 7 → cut.
- [ ] **Topic clarity** in first read (1-2 seconds of eye time)
- [ ] **No confusion** — viewer can't misparse, simple words, active voice
- [ ] **"You/your" framing OR "How I" personal angle** — feels for them, or pulls them in with credibility. Avoid generic "How to".
- [ ] **A vs B contrast** (stated or implied)
- [ ] **Identity / emotional trigger** — speaks to a specific pain, desire, or "type of person"
- [ ] **Single focal point** — one idea, not three

If any fail → rewrite the variant.

## Screenshot test (run on every middle slide)

For each non-cover, non-CTA slide, ask:

- Does this make sense without context?
- Would someone share JUST this to their story?
- Does it deliver value on its own?

If "no" to any → split, simplify, or rewrite. Each slide = a separate share opportunity.

## "Earn the swipe" rule for S2

S2 is where most carousels lose 40% of viewers. The fix:

- Don't restate the cover. Reveal something.
- Don't introduce the topic. Show the consequence/result/contrast.
- Don't say "let's break it down". Break it down.

**Bad S2:** "Here's why this matters."
**Good S2:** A screenshot of the result. A bold line that contradicts what the reader assumed in S1. The data point that proves the claim.

## Voice rules (carousel-specific — from `my_carousels.md`)

DNA.md was tuned to reels; this is the carousel-house-style ground truth extracted from Artem's top-performing carousels.

**Dial up:**
- Short paragraphs. Line breaks as visual rhythm.
- One-sentence claims with no qualifiers ("That's the entire architecture.")
- CAPS section labels — they do real visual work on slides (`STEP 01.`, `01 — SORA KILLED`, `SKILL 02`, `2012 - THE SWITCH`, `THE PROBLEM`)
- First-person results ("I built", "I automated", "I fully automated") OR second-person reveal ("You've been training", "You didn't pass the test. You built it.")
- Named tools — Claude, ChatGPT, Canva, Sora, Wikipedia, YouTube, X, Notion, Apple Notes, Waymo, Stan, ManyChat. Specificity = trust.
- Dollar/time anchors — "$150k dev team", "$45B Waymo", "Under 60 seconds", "200M solves/day", "for 15 years"
- Bottom-arrow continuation cue on every middle slide ("the fix →", "the anchor →", "Now you need to feed it →") — previews the next slide's content, never literal `Swipe →`
- Tool names in original casing (`grill-me.skill`, `CLAUDE.md`, `raw/`)

**Dial down (cut on sight):**
- "How to" hooks — replace with "How I" or "You've been"
- "Let's dive in" / "in conclusion" / "evolving landscape" / "pivotal moment"
- "Plays a crucial role" / "broader trend"
- Three generic benefits in a row
- Lines that could fit any tool or any launch
- Internal pain monologue ("I used to feel stuck...")
- Em-dash punctuation overkill (one per slide max)

## When to flag back instead of drafting

- News/announcement with no angle → push back. "What's the contrarian take or practical use? News alone = skip."
- Topic is too thin for 6+ slides → suggest tweet or single-image post instead.
- Topic is too complex for 10 slides → suggest splitting into 2 carousels with linked CTAs, OR escalate to YouTube/long-form.
- Source material is just opinion with no demo/data/example → ask Artem for one piece of evidence to anchor it.

## Output format

```
📊 CAROUSEL — [Topic]

PLAYBOOK: [Tip / Story / Curation / Same-Meme / Quote / Postcard / EDC / Two-Slide / Brand Story / Gamified]
TIER: [TOFU (growth) / BOFU (conversion)]
SLIDE COUNT: X (rationale: Magic-8 / topic-type override)
HOOK CATEGORY: [Challenge / Mistake / Listicle / Story / Curiosity Gap / Authority] — using [How-I / How-To]

—

S1 (COVER):
HOOK A: [variant 1, 3-7 words]
HOOK B: [variant 2, 3-7 words]
HOOK C: [variant 3, 3-7 words]
RECOMMENDED: [A/B/C and why]
HOOK CHECK on recommended:
  ✅/❌ 3-7 words
  ✅/❌ topic clarity
  ✅/❌ no confusion
  ✅/❌ you-framing or "How I"
  ✅/❌ A vs B contrast
  ✅/❌ identity/emotional trigger
  ✅/❌ single focal point

S2 (SECOND-CHANCE HOOK — stands alone): [the unexpected reveal]
INTENT: [what it does for the viewer]
STANDALONE CHECK: ✅/❌ would deliver value if seen first

S3 (FIRST VALUE DROP — algorithm checkpoint): [front-load the densest insight here]
INTENT: [why this is the commitment point]

S4-S(N-2): [content slides]
  S4: [copy] · INTENT: [save / completion] · SCREENSHOT TEST: ✅/❌
  S5: [copy] · INTENT: [...] · SCREENSHOT TEST: ✅/❌
  ...

S(N-1) (PAYOFF): [the aha]
INTENT: [share driver]

S(N) (CTA): [ONE specific action — see Bible CTA-by-goal table]
+ Bonus/sign-off line: [reward for completion]

—

CAPTION:
STRATEGY: [Inside Joke / Expansion / Question / Story Hook]
LENGTH: [short 1-3 sentences / medium]
[caption text]

—

HASHTAGS (5-10, varied sizes):
Niche-specific (3-4): #...
Audience-based (2-3): #...
Branded (0-1): #novitckii

—

VIRALITY DESIGN:
- Saves driver: [which slides]
- Shares driver: [which slides]
- Comments driver: [CTA strategy]
- Completion driver: [where drop-off risk is, how the slide earns the swipe]
- Slide-3 value drop: ✅ front-loaded / ⚠️ buried at slide [X]

—

🎵 AUDIO REMINDER: Attach music post-export (carousel-design step). Without audio, only existing followers see this. Match audio to the carousel's vibe — trending NOT required.

—

TONE CHECK: ✅ passes anti-AI filter / ⚠️ flag these lines: [...]
```

## Feedback loop

Carousel-specific feedback goes into `../../../content/my_carousels.md` as new entries appended after Artem ships a deck (title, slide-by-slide, views once known). The `feedback-log.md` is reels-tuned — don't push carousel voice patterns there. When a carousel pattern emerges as a *rule* (not just a one-off), promote it to the "Artem's house style" section of `carousel-bible.md`.

## Format-specific reminders that bite

- **IG hard cap = 10 slides.** Don't draft 11.
- **First 2 slides are the entire stop-scroll moment.** They get re-shown to people who scroll past, so they have to stand alone.
- **Last slide views / first slide views = the algorithm's quiet ranking signal.** Reward the swiper.
- **No swipe-to-continue arrows in copy.** ("Swipe →" reads cheap and the design system handles it.)
- **One idea per slide. If you can't say it in 2-3 short lines, it's two slides.**