How to make a YouTube Shorts ad in 2026

The Shorts brief looks like a TikTok brief with three additions: a loop angle, a written voiceover script, and an explicit end-screen CTA target. Skip any of the three and the ad will underdeliver against placements that include them. Shorts is the only mainstream short-form surface where the loop is a metric in its own right, the voiceover is sound-on by default, and the CTA is a tappable card rather than a button overlay. Brief for the platform, not for short video in general.

The loop angle is the single most important Shorts-specific decision. Each loop counts as a watch in YouTube's ranking model, so a 15-second Short that loops cleanly three times reads to the algorithm like a 45-second watch — at no extra production cost. The brief should state, in one sentence, how the closer hands off into the opener. “Last frame is the same product reveal that opens the ad” or “final voiceover line is the same question we opened with” — concrete, mechanical, written down before the shoot.

The voiceover script is the second addition. Shorts viewers are roughly 75% sound-on (compared to TikTok's 60% sound-off feed behavior), so the voice channel carries the message. Write the script line by line in the brief, with timestamps. The third addition is the end-screen CTA target: which Demand Gen card appears in the last three seconds, what the action is (Shop, Sign Up, Learn More), and which landing page it points at. A Shorts ad without a planned end-screen card is leaving the highest-intent three seconds of the ad as dead space.

Hooks on Shorts have a second job that TikTok hooks don't: they have to survive being seen multiple times. A pure pattern-interrupt hook works once, then becomes noise on loop two. A question hook or an open-loop hook gets stronger each pass because the viewer is now watching to see whether the closer actually answered the question they were promised an answer to.

The mechanical rule: the first two seconds of the loop should match the last two seconds of the cut. That match is what makes a loop invisible — the viewer doesn't realize the video restarted, so watch time accrues silently. Question-and-answer is the cleanest structure: open with “why does X happen?” and close with “...and that's why X happens,” with the same visual framing on both ends. The Ad Bench database shows question-hooked Shorts loop 2.3x on average; pattern-interrupt-hooked Shorts loop 1.1x — almost no loop value at all.

Open-loop hooks are the second strong pattern. “You've been doing this wrong your whole life” opens a curiosity gap the closer must pay off. The pay-off in the last two seconds is also the setup for the next loop's opening line, which is why the same visual carries both jobs. Pick the hook structure in the brief, not on set. Walking onto a Shorts shoot without a known loop structure means editing one into existence later, which rarely works as cleanly as designing for it up front.

The technical floor is 9:16, 1080×1920, 30fps, H.264, MP4, under 500MB — same as TikTok. The aesthetic floor is different. Shorts tolerates and even rewards a slightly more produced look than TikTok: cleaner audio, a real microphone for the voiceover, a single-subject framing that holds for longer beats. Selfie-cam still wins on native-feel, but the production bar above which an ad starts reading as “commercial” is higher on Shorts than on TikTok.

Voiceover is the central craft decision. Roughly 75% of Shorts impressions are sound-on, the inverse of TikTok's 40% sound-on baseline. That single number changes everything about how the ad is built. The voiceover is the message; captions are the redundancy layer, not the primary channel. Record the voiceover separately on a real mic if you can, layer it over the visuals in the edit, and script it line by line in advance so timing is locked. Improvised voiceover wanders; scripted voiceover hits the CTA on the beat.

Why the 75% sound-on number matters: it means audio quality is a ranking signal, not an accessibility nice-to-have. A muddy phone-mic voiceover gets the same algorithmic treatment as a muddy stock track, and the algorithm has learned that high-quality audio correlates with longer watch time. Captions still go on every line for the remaining 25% of sound-off impressions, but they're no longer doing the heavy lifting. The voice is. Cast for a voice that can hold a viewer's attention for 30 seconds without visual tricks, then build the visuals around what the voice is saying.

The Shorts edit is built backwards from the loop. Cut the closer first — the last two seconds — and then cut the opener to match it frame-for-frame. Same framing, same lighting, ideally same audio tone. When the loop restarts, the seam between closer and opener should be invisible to the viewer. That invisibility is what converts a 20-second Short into a 60-second equivalent watch in YouTube's metric.

Cut rate runs denser than TikTok. The Shorts feed scrolls faster than the TikTok For You feed in 2026 average session data, so the edit has to keep pace. A cut every 1-2 seconds for the full duration — not just the opening five seconds — holds attention through the loop transitions where the viewer might otherwise notice the restart and swipe. Hard cuts only; dissolves break the loop illusion because they read as “ending” rather than “continuing.”

Reserve the last three seconds for the end-screen CTA card. The card lives in the YouTube UI, not the video pixels, so the visual underneath it has to be readable behind the card overlay and the voiceover has to call out the action explicitly. “Tap the card” or “link below” spoken aloud lifts CTA click-through 30-40% versus relying on the card alone. The closer serves two jobs: it sets up the loop's opener, and it surfaces the CTA. Both have to land in the same three seconds.

Paid promotion on Shorts pulls audio from the YouTube Demand Gen audio library, which is narrower than TikTok's Commercial Music Library (~1M tracks) or Meta's catalog. The Demand Gen library sits closer to 150K cleared tracks in 2026 — usable but not deep, and skewed toward instrumental and licensed-mood tracks rather than the pop and trending sounds that dominate organic Shorts. The trending audio on the organic feed is not available for paid use, same constraint as TikTok's CML.

Voiceover-first creative makes the audio library constraint less painful. If the voice is the message, the background track only needs to support — not carry — the audio channel. A licensed mid-tempo instrumental at -18dB under a clearly mixed voiceover performs better than a louder trending-style track that fights the voice. Match BPM to cut rate: ~120 BPM for 1-2 second cuts holds rhythm without feeling rushed. Slower tempos read as broadcast pacing on Shorts and underperform the same way they do on TikTok.

Audio licensing rules are stricter than the organic side. Music uploaded to a Demand Gen creative must come from the in-platform library, or be cleared via a Content ID claim with the rights holder before launch. Lift a track from anywhere else and the ad either gets rejected or runs briefly and then mutes — Demand Gen will keep serving impressions on a silenced creative, which destroys performance without flagging the cause. For deeper treatment of how Demand Gen audio rules interact with monetization policies, see /learn/shorts-demand-gen and the policy boundary notes in /learn/shorts-monetization-vs-ad-policy.

Before publishing, run the cut through The Ad Bench with the platform toggle set to Shorts — not TikTok. The rubric shifts under the hood: hook scoring weights question and open-loop structures higher, native-feel allows a more polished aesthetic before flagging “commercial,” pacing expects a denser cut rate, and a Shorts-specific loop-design audit replaces the TikTok-only audio-trend score.

The loop-design audit checks whether the first two seconds match the last two seconds tightly enough to make the restart invisible. A loop-design score below 70 means the seam is visible — viewers will notice the restart and swipe. Above 80 means the loop is clean enough that repeat-view rate becomes a function of the hook and the offer, not the edit. The repeat-view forecast is the single number to optimize for: The Ad Bench projects what percentage of viewers who reach the end will loop at least once, based on the loop seam quality, voiceover pacing, and closer match.

The end-screen CTA score is the Shorts-specific addition that catches the most launch-ready cuts. CTA below 70 on Shorts usually means one of three failures: the voiceover doesn't name the action, the visual under the card is too busy to read behind the overlay, or the CTA card type (Shop vs Sign Up vs Learn More) doesn't match the funnel stage of the audience signal. Fix all three before launch. A 30-second rewrite of the closer line — “tap the card to get started” spoken aloud — typically lifts CTA score from the 60s to the high 70s with no reshoot required.

Demand Gen is the campaign type that replaced Video Action Campaigns in 2024 and now serves Shorts, Discover, and Gmail from a single creative pack. One vertical video uploaded once gets distributed across all three placements with the Demand Gen machine learning deciding the mix. This is the default path for most advertisers in 2026, but it is not always the right call.

Run Shorts-only when the creative is genuinely loop-built — same opener/closer, voiceover-forward, end-screen CTA — and you want to isolate Shorts performance for a clean read. Run Demand Gen cross-placement when the creative also works as a still-image card (Discover) and a static promotion (Gmail), or when budget is tight enough that the cross-placement reach matters more than attribution clarity. Most accounts land at 60% Demand Gen cross-placement / 40% Shorts-only campaigns once they have a performance baseline.

Brand safety has a different shape on Demand Gen than on TikTok. The Demand Gen brand-safety risk bar is set inside Google Ads under content suitability — Expanded, Standard, or Limited. For most direct-response advertisers, Standard is the default; Limited cuts inventory by 30-40% and only makes sense for regulated verticals or sensitive brand contexts. Expanded unlocks more inventory but includes content categories most advertisers would rather not appear adjacent to. Pick once at campaign setup and document the choice in the brief so the decision is reproducible across launches.

Campaign objective decides the optimization signal. For Shorts advertisers in 2026, the working objectives are Conversion (or Sales) for direct-response funnels and Awareness for top-of-funnel brand work. Pick based on the funnel stage in the brief — objective changes mid-campaign reset learning and waste the audience signal Demand Gen has built. Conversion requires a configured conversion action in the Google Ads conversion column; set it up before campaign creation, not during.

Targeting on Demand Gen uses audience signals rather than explicit targeting. You give Google Ads a seed — a customer-match list, a website-visitor segment, a lookalike, or an interest cluster — and the system uses it as a signal of who to find, not a wall around who it's allowed to serve to. Narrow audience signals can starve delivery; broad signals plus a strong creative consistently outperform tight targeting plus a weak creative. Start broad, watch the audience insights report, and let the system find converters before you constrain it.

Daily budget floor is $30-50/day per ad group to exit the Demand Gen learning phase within a week. Below that, delivery is too sparse for the model to find consistent signal and the ad group spins without learning. Creative diversity beats budget size: 3-5 distinct video assets per ad group lifts CPA performance roughly 40% over single-creative ad groups in the Ad Bench database, because the model rotates assets based on which one the current audience slice responds to. Build a library; don't bet on a single cut.

Repeat-view rate is the leading indicator on Shorts, not completion rate. The metric: percentage of viewers who reach the end and then loop at least once. Healthy Shorts ads run 12-18% repeat-view rate in their first 24 hours of delivery. Anything below 8% after 1,000 impressions is dead — kill it, don't wait for CPA to confirm what repeat-view already told you. For the underlying economics of why repeat-views drive the auction, see /learn/shorts-repeat-view-economy.

The three numbers to watch in the first 24 hours: repeat-view rate (creative health), CPC (audience match quality), and CVR (offer and landing-page strength). Don't wait for CPA to stabilize — by the time CPA is reliable, you've spent more than you needed to spend to learn what repeat-view rate told you in the first few hours. A creative with a 15% repeat-view rate and a weak CTR is usually a CTA problem, not a creative problem; rewrite the closer line, relaunch.

Scale winners by 20% per day, not by doubling. Demand Gen's learning resets when budgets jump more than ~20% in 24 hours, and a reset wastes the audience signal the campaign has been building. The operating cadence that compounds: morning review, kill anything below the 8% repeat-view floor, raise winners by 20%, end the day with a cleaner ad-group composition than you started with. Run that loop for seven days and you have a paid Shorts program with predictable economics — not a series of one-off launches that work or don't for reasons you can't reproduce.