The task
Your last carousel got 18k reach but only 14% completion. People scrolled past slide 2 and never saw the actual content. Slide 1 was the title slide (“AI Prompts 101”). Slide 2 was a “here is what we will cover” agenda. Slide 7, where the real payoff was hiding, got viewed by 1 in 8 people. Saves were terrible because most people never saw the save-worthy slide. You want a 7-slide script that earns the swipe through structure — slide 1 hooks, slide 2 delivers a concrete fact (not setup), slide 6 lands the surprise, slide 7 asks for the action. Pull-through engineered into the content; no “swipe →” arrows pretending to do the work.
Two facts make this worth getting right in 2026. Carousels pull roughly 1.4x more reach than a static post (Instagram head Adam Mosseri’s own figure), and they post the highest engagement rate of any format — about 1.92% average, versus ~0.50% for Reels and ~0.45% for static images (industry benchmarks, as of June 2026). The format rewards a tight script more than any other.
Where AI helps — and where it does not
AI is good at structuring the carousel along an attention curve — placing the hook, the build, the surprise payoff, and the CTA at the right slides. It is also useful for writing slide copy at the right density per slot (slide 1 needs 4-6 words, slide 4 can carry 25 words, slide 7 needs 8 words and a clear ask). Any current flagship handles this: GPT-5.5 (ChatGPT), Claude Sonnet 4.6, or Gemini 3.1 Pro all return usable scripts. Claude tends to write the tightest short-form copy and respect word budgets most reliably; GPT-5.5 is stronger at generating hook variants in bulk. If you are unsure which to use, see ChatGPT vs Claude vs Gemini.
Where AI fails: designing the visual. Carousel is half copy and half design; the script tells you what to write, but the slide layout, type hierarchy, and color decisions belong to a designer or a Canva template. Carousels that lose at slide 2 sometimes lose because the type is too small to read on a phone, not because the copy is weak.
A common failure mode: the model writes slide 1 as a title and slide 2 as an agenda. Both are wasted. Slide 1 must be a hook with a curiosity gap that slide 2 closes; slide 2 must deliver one concrete fact, not “here is what we will cover.”
Why slide 2 is now a second hook, not an afterthought
The carousel “re-serve” mechanic makes this concrete. In Mosseri’s words: “If someone sees your carousel post but they don’t swipe, we’ll often give that carousel a second chance and automatically move to that second piece of media for the viewer.” In practice, a follower who scrolls past slide 1 can get the same carousel served again later with slide 2 as the preview — a second free impression no other format gets. (Mosseri shares these mechanics directly on his @mosseri account and in Instagram’s creator guidance.) So slide 2 has two jobs: close the slide-1 curiosity gap for people who swiped, and stand alone as a fresh hook for people who never saw slide 1. An agenda slide fails both jobs. Tell the AI this directly in the prompt and it stops writing throwaway slide 2s.
What to feed the AI
- The single concept the carousel teaches in one sentence — narrower is better; “5 AI prompt patterns” beats “how to use AI”
- Audience level — beginner / intermediate / expert; the same concept needs different examples per level
- The specific action you want at slide 7 — save, comment, click bio link, share to story; each has different copy
- One past carousel of yours that performed well (tone anchor) — paste the actual slide text
- The format constraint — 4:5 vertical (1080×1350px), square, or 9:16 reels-style carousel
- The number of slides — Instagram allows up to 20 slides (raised from 10 in 2024), but 7-10 is the practical sweet spot; a 20-slide carousel people abandon at slide 8 beats nothing, but loses to an 8-slide one where 70% reach the end
- The “save trigger” — a specific reason to save (a reference list, a framework, a checklist, a prompt template)
- The hard avoid list — no “swipe →” arrows in the script, no “100% of people,” no emoji unless your brand uses them
The 7-slide spec at a glance
Each slot has a job, a word budget, and a read-time ceiling. This is the structure the prompt below enforces.
| Slide | Role | Headline | Body copy | Job |
|---|---|---|---|---|
| 1 | Hook | ≤6 words | ≤10 words | Open a curiosity gap; never a title |
| 2 | Second hook | ≤8 words | ≤25 words | Deliver one concrete fact; stand alone (re-serve target) |
| 3-5 | Build | ≤8 words | ≤25 words each | Each slide works on its own |
| 6 | Payoff | ≤8 words | ≤20 words | The surprise / save slide |
| 7 | CTA | ≤8 words | ≤15 words | One-tap action (save / comment / bio) |
Every slide should be readable in about 4 seconds on a phone. If a slide exceeds its word budget, cut it before you touch the design.
Copy-ready prompt
Write a 7-slide Instagram carousel script.
Concept (one sentence, narrow): {paste}
Audience level: {beginner / intermediate / expert}
Action at slide 7: {save / comment / click bio / share to story}
Past carousel that performed (tone anchor): {paste actual slide text or "first carousel"}
Format: {4:5 vertical / square / 9:16}
Save trigger: {what specifically would make someone save this — list, framework, prompt}
Banned: {"swipe →" arrows, "100% of people," "this changed my life," emoji-only slides}
For each slide return:
1) Slide number + role (hook / build / payoff / CTA)
2) Headline — under 6 words for slide 1, under 8 words for slides 2-6, under 8 words for slide 7
3) Body copy — slide 1 ≤10 words, slides 2-5 ≤25 words, slide 6 ≤20 words (the surprise lands with fewer words), slide 7 ≤15 words
4) Reason this slide pulls the reader to the next slide — written into the content, not into an arrow
5) Visual idea in one sentence — what the slide image / illustration should show
Structure:
- Slide 1: hook with a curiosity gap. Make the reader feel "I do not know the answer and I need to know it." NOT a title slide.
- Slide 2: deliver one concrete fact that closes the slide 1 curiosity gap and opens a new one. It must ALSO stand alone as a hook, because Instagram re-serves the carousel with slide 2 as the preview to people who scrolled past slide 1. NOT an agenda slide.
- Slides 3-5: build the argument or list. Each slide must work on its own; readers who skipped slide 4 should still get value from slide 5.
- Slide 6: the unexpected payoff or surprising counterpoint. This is the save slide.
- Slide 7: CTA. The action must be answerable in one tap (save), one sentence (comment), or one click (bio).
End with: which slide is the most likely drop-off point in this draft, and what you would change in design (not copy) to reduce it.Shorter variant — single-slide rewrite
Below is slide {N} of my carousel: {paste copy + role in structure}.
Its job in the carousel is: {hook / build / surprise payoff / CTA}.
The audience drop-off at this slide is {N%}.
Rewrite the slide to better serve its job. Return 3 versions ranked by strongest pull-through. Do NOT add "swipe →" — the pull is in the content.Sample output
A strong slide 1 that earns the swipe: “Why 90% of carousels die at slide 2.” — curiosity gap (why?), specific number, the reader genuinely does not know the answer.
A useful slide 2 that closes the gap: “Slide 1 promised a payoff. Slide 2 usually delivers an agenda. Brains skip agendas. The fix: deliver one concrete fact on slide 2, not ‘what we will cover.’” — closes the slide-1 curiosity, opens a new question (what fact for my concept?).
A useful slide 6 (surprise payoff): “The carousel that performs best in 2026 has only 5 slides and a paragraph CTA on slide 5. Long carousels are 2023 thinking.” — counterintuitive, save-worthy.
A useful slide 7 CTA: “Save this if you publish carousels weekly. The slide structure works on any topic.” — direct action, names the audience, justifies the save.
How to refine
- If carousel completion is below 30%: “Slide 2 is the problem. It must deliver one concrete fact, not a setup. Rewrite slide 2 to close the slide-1 curiosity gap and open a new question for slide 3.”
- If saves are low but completion is high: “Slide 6 or slide 7 is the problem. Slide 6 needs the save trigger (a reference, a checklist, a framework); slide 7 must explicitly tell the viewer why to save.”
- If slide 1 reads as a title: “Rewrite slide 1 as a question or a contrarian statement, not as the topic of the carousel. The reader should feel they cannot guess the answer.”
- If the “reason to swipe” feels artificial: “Rewrite the slide so the reason to swipe is embedded in the content. ‘Curious why? Slide 3 →’ is artificial; ending slide 2 with a half-truth that slide 3 completes is embedded.”
- If slide text is too dense: “Each slide must be readable in 4 seconds on a phone. Cut any slide where body copy exceeds the target word count.”
Common mistakes
- Slide 1 as the title slide: wastes the most valuable real estate; readers see “AI Prompts 101” and skip because they cannot guess what is new about this version.
- Slide 2 as an agenda: brains skip agendas; if slide 2 does not deliver a fact, the carousel is over for 60% of viewers.
- Same text density on every slide: the brain craves rhythm; alternating dense and light slides keeps attention more than uniform density.
- CTA on slide 6, not slide 7: never assume viewers see the last slide; CTA placement should match the slide most people actually reach, which is slide 6 on poorly-built carousels but slide 7 on well-built ones.
- “Swipe →” arrows doing the work: the arrow is a sign of weak content; if you need an arrow to make people swipe, the slide failed.
- No save trigger anywhere: a carousel that does not deserve a save will not get one; engineer the save into slide 6.
- Same hook pattern as your last 4 carousels: viewers learn the pattern and the algorithm shows them less; vary the hook structure every 2-3 carousels.
- No mobile readability check: body text at 24pt looks great on desktop, unreadable at 4:5 on a phone; design for the smallest screen the carousel will be viewed on.
FAQ
- How many slides maximum?: Instagram allows up to 20 slides as of June 2026 (the cap rose from 10 in 2024). But the practical sweet spot is 7-10; engagement tends to dip in the middle and recover near the end, so a 20-slide post people abandon at slide 8 underperforms a tight 8-slide one where most viewers finish. Use 10+ only when the content genuinely requires it (a step-by-step framework, a multi-part case study).
- Square or vertical carousels?: Vertical 4:5 at 1080×1350px takes ~25-33% more feed real estate than square, gets more saves, and outperforms square in nearly every category. Use vertical unless brand visuals demand square. Keep every slide the same aspect ratio — mixed ratios get cropped.
- Can I mix images and video slides?: Yes, and it pays. Mixed-media carousels (images + video) average roughly 2.33% engagement versus ~1.80% for image-only, per 2026 benchmarks. Export video slides at the same pixel dimensions as your images (1080×1350 for 4:5).
- Should the cover slide have the brand logo?: Tiny, in a consistent corner. Big logo on slide 1 reads as a stock-photo template and hurts swipe rate.
- What about reels-style 9:16 carousels?: Newer format, currently underused, gets a temporary algo boost on some accounts. Test against your 4:5 baseline; do not assume the win transfers across audiences.
- How do I measure if a carousel worked?: Completion rate (last-slide views / impressions) and save rate (saves / impressions). Comments are a bonus, not the primary metric; the algorithm currently weights saves and shares highest for carousel reach. Watch slide-2 retention specifically — the re-serve mechanic means a weak slide 2 leaks both your swipers and your second-impression viewers.
Related
- AI Xiaohongshu Series Content
- AI Viral Opening Lines
- AI Audience Engagement Reply
- AI Creator Collab Pitch
- Instagram Carousel Outline Prompts for 7-10 Slide Stories
Tags: #AI writing #Social media #Workflow #Instagram #Carousel