Build a Scroll-Stopping AI Carousel in One Afternoon.

Why most carousels stall at slide three

The pattern is familiar: a strong hook slide, three slides of generic filler, and a CTA that asks for nothing in particular. Instagram carousels average around 0.55% engagement, the highest of any format on the platform, but the drop-off point is consistent. Audiences swipe past slide one on the strength of the hook, then bail by slide three or four when the content stops delivering anything new.

Carousels live or die on pattern interruption, not polish. A beautifully designed slide with a generic phrase still loses to a rougher slide that says something specific. The fix isn't more design time. It's a tighter structure built before a single visual gets made.

This guide runs the whole thing in an afternoon: research, structure, design, refine, export. Four hours, one sitting, all of it inside Marketing Studio. The target isn't a rough draft to tidy up next week. It's a 6-8 slide carousel ready to publish by the time you close the laptop.

Hour one: ground the concept in real research, not vibes

Skip the blank-page opener. Before writing a hook, use Research to pull current data, competitor angles, and trending formats in your niche. What are competitors in your space oversimplifying or skipping entirely? Educational, value-driven carousels and data visualisations consistently outperform generic inspiration posts, so the research step is what separates a carousel with substance from one that just looks nice.

Once you've got a handful of data points, angles, or gaps worth exploiting, bring them to Ray and ask for a hook structure based on what you found and the platform you're publishing to. An Instagram carousel and a LinkedIn document post reward different openers: Instagram favours a sharp visual pattern interrupt, LinkedIn rewards a framework or data claim stated plainly in the first line.

This mirrors how a performance marketer builds a paid ad angle and how a content creator builds an organic educational series. Both start from the same place: research first, hook second. A marketer might be digging for the ad angle that converts; a creator might be looking for the myth their audience keeps repeating. The research step is identical even though the output differs.

Before moving to copy, lock a one-line thesis for the carousel. Something like: "Most exhibition stands fail because the walk-through plan comes last, not first." Every slide after this point has to serve that line. If a slide doesn't advance the thesis, it doesn't make the cut, no matter how good it looks on its own.

Hour two: draft copy and structure slide by slide

With a thesis and a hook direction in hand, move to Write and use the model picker to draft slide-by-slide copy. Gemini Flash is the fast pass: use it to generate a wide spread of hook options or to quickly rewrite a CTA for LinkedIn versus Instagram without waiting around. Claude Sonnet 5 earns its place when you need sharper hooks and tighter, more structured multi-slide narratives, particularly once you've pasted in brand context or a swipe file and want the copy to actually sound like your brand rather than a generic template.

The logic here is the same hook-body-payoff structure a graphic designer uses when drafting a brand one-pager, or a game studio uses when writing a Steam capsule description. Slide one states the stakes. The middle slides deliver the substance. The last slide pays it off. A capsule description that opens with "The co-op horror game where the map fights back" doesn't waste slide two explaining art style, it moves straight into mechanics, because the hook already did its job.

Keep each slide to one idea and, where possible, one sentence of copy. Carousels reward restraint. A slide crammed with three points reads as effort, not clarity, and effort is not what gets a save or a share. Specific stakes beat vague inspiration every time: "The ads mistake that cost me £50,000" outperforms "5 mistakes to avoid in your ad strategy" because it commits to a number and a consequence.

Draft the final CTA slide early, before the middle slides. This sounds backwards but it isn't. Ad copy and conversion funnels are built from the goal backwards, and a carousel is no different. Once you know exactly what the last slide asks for, whether that's "wishlist the game," "book the stand walkthrough," or "swipe up for the full spec sheet," every prior slide can be edited to build toward it rather than toward nothing.

Model	Best for in this workflow	Why
Gemini Flash	Fast iteration, hook lists, platform rewrites	Quick turnaround for volume drafts
Claude Sonnet 5	Sharper hooks, structured narratives, tight CTAs	Holds brand context and long swipe files well

Hour three: generate and lock the visual system in Marketing Studio

Copy and visuals live in the same place in Marketing Studio, so there's no toggling between a design tool and a separate caption document while you try to keep the two in sync. That matters more than it sounds: half the reason carousels drift off-brand by slide five is that the copy gets finalised in one tool and the visuals get built in another, days apart, by which point the thesis has quietly shifted.

Before touching slide six, establish a visual anchor: one colour system, one type treatment, one layout grid, repeated across every slide. This is what makes a carousel read as one piece rather than six loosely related tiles. The grid can flex slide to slide (a full-bleed image on slide one, a two-column data point on slide four) but the colour and type language should stay locked.

This applies differently depending on discipline. An automotive brand announcing a colourway launch might structure the visual system around one hero shot per slide, each cropped to the same feature area (exterior line, interior texture, dashboard detail), ending on a CTA to build a spec. An exhibition design team teasing a stand reveal might run concept sketch, then 3D render, then materials, then visitor flow, each slide using the same corner treatment and caption style so the progression feels intentional rather than assembled after the fact.

Generate two or three visual directions before committing to one. This is the cheap moment to explore alternatives. It gets expensive fast once you're four slides into a direction and realise the type treatment doesn't hold up against a busy background image. Comparing directions side by side, before locking, saves the rebuild.

Hour four: refine, caption, and export for the platform

Before exporting anything, sense-check every slide against the one-line thesis from hour one. Anything that doesn't serve it gets cut, even if it's a slide you're personally attached to. This is the point where most carousels quietly bloat back up to ten slides of "nice to have" content that dilutes the thesis instead of sharpening it.

Use Ray as a second pair of eyes here. Paste the full slide sequence and ask where attention is likely to drop. It's a useful gut-check against the slide-three problem, because it's easy to lose objectivity after staring at your own sequence for three hours.

Match export format and aspect ratio to the destination. Instagram carousels are best built at 4:5 portrait (1080x1350px), which occupies roughly 20% more feed real estate than a square crop, still the recommended ratio for reach. Instagram supports up to 20 slides technically, but 8-10 tends to perform best there, while LinkedIn favours 6-10 with a more data-forward, document-style structure. Mixed media inside the set, a short video clip alongside stills, can help hold dwell time, but it's not a requirement for the format to work.

Caption strategy matters as much as the slides themselves. Lead with the target phrase in the first sentence for social search, write the caption as a short standalone piece that extends the carousel's value rather than just teasing it, and close with a participation prompt. Measure saves, shares, and meaningful comments, not link clicks, since that's what carousels are actually built to earn.

Final check before publishing: does slide one work as a stand-alone thumbnail, with no context, no caption, nothing else around it? If it doesn't stop a scroll on its own, no amount of polish on slides two through eight will save it.

What to reuse next time

The visual system built in hour three doesn't need rebuilding from scratch next time. Save it as a project template inside Projects, so the next carousel starts at hour two, not hour zero. The colour, type, and grid decisions are already made; only the research and copy need fresh attention.

Keep a swipe file of hooks that tested well, whether that's a specific stakes-driven opener, a myth-versus-reality structure, or a social proof funnel that drove strong saves. Feed that swipe file back into Write as reference material the next time you're drafting, so Claude Sonnet 5 or Gemini Flash aren't starting cold either.

The real time saving isn't the generation step. It's not having to re-explain your brand voice, visual system, and past performance notes every time you open a new tool. A brand context doc, a swipe file, and a saved visual template mean the next carousel skips straight to research and drafting, and an afternoon becomes an hour and a half.

One afternoon, one carousel, ready to publish. The workflow holds whether you're an automotive brand teasing a colourway or a game studio building hype ahead of a wishlist push. The discipline changes. The four hours don't.