Skip to content

Ad Platforms · AppLovin

Last Updated: August 11, 2026

Creative

How Tierra builds, iterates, and gets AppLovin creative approved: video, the Interactive, the end card and catalog, and compliance. Read alongside the playbook for the platform model and vocabulary. Not covered here: which creative to test and when to relaunch (strategy), pausing and budget moves (media buying), and reading performance or catalog contribution (reporting and tracking). A client's own compliance word lists and rejection history live in that client's project.

A creative set plays three steps in one impression: Video → Interactive → End Card. A node is Tierra's term for one video paired with one Interactive. Build for the combination, not the single asset.

Video (Step 1)

Specs, from AppLovin's ad specs and requirements:

  • Portrait, 9:16. Minimum 270×480; we produce at 1080×1920.
  • MP4, MOV, or M4V. Up to 1 GB.
  • 60 seconds is a hard maximum. Export at 59 seconds to be safe, because a video that reads as 60 can round just over from encoding, and anything over 60 gets rejected.

Our production defaults (not platform requirements): H.264 at 30fps, audio AAC around -14 LUFS.

There's no built-in call-to-action button; the platform overlays its own. Keep the bottom of the frame clear for it. We reserve the bottom one-sixth as the safe margin, which is our own guideline rather than a documented spec.

A video that underperformed on another platform can still be worth reworking for AppLovin, but judge it case by case. If the concept itself is weak, a rework won't save it. If the concept is sound and it just wasn't executed well for that platform, re-editing it to AppLovin's format can absolutely work. Don't write off a video because it flopped elsewhere, and don't assume it'll carry over either.

Opening and hook:

  • Skippable (interstitial) ads can be skipped after a short minimum delay (AppLovin's spec is a five-second minimum before the skip appears). Rewarded ads aren't skippable; the viewer chose to watch, so they see the whole thing.
  • The skip control sits in the top-left and appears after about five seconds, and viewers' eyes go there, so put the hook and any callout in the top third of the frame where they're already looking.
  • The opening still has to earn attention, but unlike a scroll feed you often have a more engaged viewer, so there's room to tell a fuller story rather than only chasing a scroll-stopping hook.
  • Assume the opening has to work with the sound off. Around 90% of views are muted, so captions are required. AppLovin-native captions (progressive word highlighting, two to three words per second) beat captions ported straight from Meta. In our own experience captions often read better in the middle band of the frame than at the bottom, since the bottom is the platform's CTA zone.
  • Hook archetypes worth testing as a set: a time-bound promise, a vulnerable confessional, product-as-character, and a reverse-hook curiosity gap. Rotate through them rather than betting on one.

Structure:

  • Use a proven direct-response structure. Two we lean on are Problem-Agitate-Solve and Attention-Interest-Desire-Action. These are examples, not the only options.
  • A strong close in the last few seconds helps: one or two of price, urgency, social proof, scarcity, or a plain instruction, with the eye guided to the button at the bottom. This is general creative craft, not an AppLovin rule, and it's never fully formulaic.

Length:

  • Match length to funnel intent: longer for top-of-funnel (around 50 to 60 seconds), around 30 to 45 seconds for mid-funnel, and around 20 to 30 seconds for a bottom-of-funnel offer. For problem-solution creative, don't name the product until around 15 seconds in.
  • As a working target mix to test (not a rule), aim for contiguous coverage across lengths: roughly a third at 45 to 60 seconds, a third at 30 to 45 seconds, a fifth at 15 to 30 seconds, and the rest under 15 seconds. A large share of rewarded-placement spend tends to land on the 45-to-60-second tier. A high-volume campaign can still underperform if it over-indexes short and leaves gaps in the mid-length tiers, so audit the ladder for dead zones.
  • Our own read: longer videos tend to win here. Treat it as a bias to test into, not a guarantee.

Cutting long-form down to size is covered below.

Interactives (Step 2)

Specs, from AppLovin's ad specs and requirements:

  • An HTML/interactive page, or an image (GIF, JPEG, or PNG). Minimum 270×480, up to 10 MB, GIF up to 60 seconds. Uploaded to the Media library.
  • A looping MP4 (5 to 15 seconds) also works as the input to the Interactive Builder.

Guidance:

  • Video vs static: default to video, it's the reliable performer. Keep testing statics on each account, since neither wins everywhere. Produce both from a winning concept and let them compete.
  • Put the call-to-action at the bottom or the middle, and always include one. Bottom or middle is a strong recommendation, not a hard rule.
  • An advertorial or presell page can work well as the Interactive.
  • AI-generated Interactives get about half the click rate of hand-built ones (smaller tap zones), but conversion and cost per purchase can match or beat them, so judge them on conversion, not clicks. AppLovin's own auto-generated Interactives can show a misleadingly high click rate.
  • Never use deceptive elements to farm taps: a fake close button, a fake system prompt, anything meant to trick a tap. They spike clicks, wreck cost per purchase, and violate AppLovin policy.

Full build patterns (hero layout, quizzes, refresh cadence, offer alignment) are in interactive production.

End card and catalog (Step 3)

The end card shows in every impression. Always run a catalog, even for a single product. There are two types:

  • Product Catalog: personalized product recommendations that link to product pages, for brands with roughly five or more products. Fed by a Shopify sync (preferred), inferred from the pixel, or a CSV/feed (the only method that lets you edit fields like title and image).
  • Offering Catalog: you build it by hand with your own images and landing-page URLs. Good for few-product accounts, or for benefit tiles.

For a single product, build an Offering Catalog of benefit tiles: one tile per benefit or pain point, all pointing to the same product, each with different copy. Think Google sitelinks.

The catalog is the smallest lever for clicks and sales, since reaching it takes an extra tap, but it's in every impression and worth building. Image and feed specs, the tile-design framework, and the per-type details are in catalog production. Contribution and lift numbers are in reporting and tracking.

When there's no catalog, branding fills the end card (logo, name, description, button), set in Brand Settings. The field list is in the playbook.

When the funnel is unbranded (a product-blind VSL, or video sales letter, funnel), keep the end card product-blind and logo-free so it doesn't break the funnel's framing. Build the end card in-house either way: agency-built end cards tend to beat client-handmade statics on CTR.

Iterations

  • Iterate a proven winner by changing one element at a time: the hook, the call-to-action copy, the on-screen text, or sound on/off over the opening. That's different from shipping a whole new concept.
  • Always ship an iteration as a new set. Never append variants to a live set, because it disturbs a set the model is already learning on.
  • Hook changes are the highest-leverage. Build a hook ladder: pick source videos that share the same concept and sit on adjacent hook numbers, and keep the same spokesperson, so the only thing moving is the opening.
  • A cheap way to spin new hook variants: prepend a short six-to-twelve-second AI-generated intro onto a proven winner, then cut straight to the winner. New hook, same proven body.
  • Spread iterations across two or three proven winners rather than making seven variations of one. Diverse signals from several winners beat depth on a single winner.
  • When picking a batch of variants to test, weight the source video most heavily, or the batch collapses into variations of a few sources. A batch of 100 picks should span at least 50 distinct source videos.
  • Changing several elements at once is a deliberate new test, not an iteration.

How much creative to make

  • Match creative volume to spend and account maturity: more as you scale. Current guidance to apply and adjust per account: roughly one new asset per $5,000 to $10,000 of spend. On a seasoned account (over about $200,000 spent), run about 70% iteration and 30% net-new. Under about $50,000 spent, flip it to about 30% iteration and 70% net-new.
  • Don't flood the account. A big single-day drop (20 or more videos) causes a dip while the model rebalances; steady daily additions are smoother. Exact cadence varies.
  • Relaunch regularly. About once a month, build fresh creative sets from the current top performers (roughly the top three to five spending videos paired with the top four to eight Interactives by click rate over the last seven days) and launch them as new sets. This keeps proven winners in rotation before they go stale.

Compliance

We always follow AppLovin's ad content policy, and we never alter an ad to get it past review. We run only what genuinely gets approved.

Disclosures:

  • If an ad includes a required disclaimer (for example hemp or health claims) or AI-generated content (images, audio, or video), include the required disclosures or labels. For AI-generated content, also make sure we have the rights to use it.
  • Never hide a disclosure behind a reveal, a darkened overlay, or tiny illegible text.

How the review behaves:

  • Any edit triggers a fresh review, which can flip a previously-approved ad to disapproved.
  • The review isn't perfectly consistent. A small change in element position can change the result.

Heuristics Tierra uses, from experience getting ads approved:

  • The coffee-brand test: if the script could pass as an ad for a coffee brand or a meditation app, it should clear; if it sounds like a drug or a medical claim, it won't. Apply it to the script, the on-screen text, and any reply-comment bubble.
  • Anchor any suggestive or borderline visual to clear lifestyle context in the first couple of seconds, and keep a running per-client list of words that get flagged alongside safer ways to phrase them. Those client-specific lists live in the client's project.
  • The review reads on-screen text and emoji the same as spoken words, and text on a whiteboard the same as a voice-over. Assume every surface is read.

When a rejection reason isn't obvious, use the chop test: split the video into segments, upload each one, and find exactly which part is triggering the rejection. Then run only the version without it. The point is to understand what's approved, not to sneak anything through.

Rejection reasons you'll see by name: Unsupported Category, Legal Hemp Disclaimer, Misleading Claims, Harassment and Profanity, Grotesque Imagery, and Restricted/Adult.

Disciplined intake gets rejections down to around 5%. Sloppy health and wellness intake runs 50 to 60%. The full framework and recovery steps are in compliance framework.

Cutting long-form down to size

When a client has long-form creative (a video sales letter, a webinar, a podcast) that runs longer than the 60-second maximum, cut it into standalone 60-second pieces that each work on their own. The detailed workflow (tooling, how many to cut, and saturation limits) is in cutdown workflow.