AI Carousel Generator.

Hero carousels usually do. The best-known interaction study — Notre Dame tracking its own homepage rotation — found roughly 1% of visitors clicked any slide at all, and the overwhelming majority of those clicks went to slide one. Motion at the top of a page is processed as advertising and routed around, so the slides after the first are effectively invisible. User-driven carousels are a different story: product galleries and manually paged testimonial strips perform well because the visitor asked for the rotation. The pattern is not the problem; uninvited autoplay is.

Default to no. If autoplay has a real job — an ambient logo marquee, a lobby display — wire the safety features in the same prompt: at least 7 seconds per slide, pause on hover and focus, permanent stop after any user interaction with the controls, and no motion at all under prefers-reduced-motion. Never autoplay anything with sentences on it: reading speed varies, and a slide that rotates mid-sentence is content the visitor cannot finish. And never autoplay a hero — that is the setup behind every banner-blindness statistic.

Wrap it in role="region" with aria-roledescription="carousel", give each slide role="group" with an aria-label like "3 of 8", and make Previous, Next and every dot real buttons with names. Keep inactive slides in the DOM but aria-hidden with their links untabbable, and route slide changes through a polite aria-live region so screen-reader users hear what sighted users see. One counterintuitive rule: while autoplay is running, the live region should stay quiet — announcing every rotation is its own assault. MeDo generates this wiring from the prompt; retrofitting it onto a transform-based slider is the painful path.

Let the browser do it: an overflow-x container with CSS scroll-snap gives you native touch momentum, and — critically — a vertical swipe still scrolls the page. This is where transform-based sliders earn their bad reputation: they attach touch handlers that cannot tell a horizontal flick from a vertical one, so users get trapped in the carousel while trying to scroll past it. With snap containers, Previous and Next become a progressive enhancement for desktop rather than a load-bearing gesture layer.

Stop when the visitor is navigating — galleries and testimonial rotators should hit a disabled button at the last slide, because position is information and infinite wraparound turns 'go back two' into a memory task. Loop when the carousel is ambient — a logo marquee nobody tracks. CSS scroll-snap has no native loop, so looping means cloning slides for the seam, which is exactly the kind of fiddly detail to name in the prompt rather than hand-wave.

Usually, yes. If the visitor needs to compare items side by side, a grid wins — a carousel shows one thing at a time by design. If the content is the page's main message, it should not rotate. If the set is three items, just stack them; paging controls for three things are ceremony. The honest checklist for a carousel: a long set (six or more), visual content, user-initiated browsing, mobile-first audience, and a first item that satisfies everyone who never pages. Fail two of those and a static layout is the better component.