AI video ad guidesPet ads

How to scale UGC ads with AI for an agency

Use AI to get more testable ads out of each creator shoot, not to replace real customer experiences. Shoot swappable pieces, let AI handle briefs, transcripts and rough cuts, then run every ad through one scorecard and a human approval step. Judge the system by cost per winner, not by how many videos you export.

Updated November 20, 20267 steps5 prompts3,228 real ads

Make my video ads →

$1 today. Cancel anytime before day 7.

The steps
  1. 1Shoot swappable creator piecesGive creators a shot list, not a script. Ask for three openers, a demo, an objection answer, and clean B-roll, each recorded as its own take.
  2. 2Turn research into angle briefsFeed an AI assistant client-approved reviews, support questions, and past results. Make it use only an approved claims list and flag anything that needs substantiation.
  3. 3Build a small test matrixPlan 3 angles x 2 creators x 2 hooks, then launch only as many cuts as the budget can support. Change reasons to buy, not caption colors.
  4. 4Use AI for edit grunt workTranscribe footage, find usable takes, draft captions, and assemble rough cuts. A human editor finishes the pacing and the first second.
  5. 5Tag and score every adName each ad by angle, creator, hook, offer, and version. Review weekly against the client's business goal, not just views.
  6. 6Approve before anything launchesA reviewer checks claims, captions, disclosures, usage rights, and likeness permission. No synthetic presenter poses as a real customer.
  7. 7Pilot one client for a monthRun one client, one offer, one platform. Compare hours, approval time, and cost per winner against your old process.
On this page
  1. The steps
  2. Real pet video ads
  3. Shoot creators for swappable pieces
  4. What long-running pet ads open with
  5. Turn client research into briefs
  6. Build a test matrix that stays small
  7. Where AI speeds up the edit
  8. Give production and buying one scorecard
  9. Approve first, then pilot for a month
  10. Prompts to copy
  11. A weekly UGC batch for one client
  12. Mistakes to avoid
  13. Questions

Real pet video ads to model yours on

3,228 video ads from 395 advertisers in the Meta Ad Library, read October 5, 2026. The median one ran 44 days, and an ad still paid for after months is the closest public sign that it works. Live ads first.

  • Ran 8 days
  • Ran 4 days
  • Ran 4 days
  • Ran 4 days
  • Ran 18 days
  • Ran 8 days
  • Ran 12 days
  • +3,216 more pet video ads

    Hooks, landing pages and run length for each

  • Ran 15 days
  • Ran 18 days
  • Ran 3 days
  • Ran 31 days

Every creative, with its copy, format and run length

The openers that ran longest

  • ShawnCo Essential Dog Harness!Shawnco · running 1,024 days
  • {{product.name}}Whunde · running 953 days
  • {{product.name}}ACANA Italia Petfoods · running 921 days
  • Permanently Removes Pet Stains and Odors!FurryFreshness · running 800 days

Shoot creators for swappable pieces

Stop briefing creators with a word-for-word script. Give them a shot list instead, so every piece can be swapped without rebuilding the ad. Ask for three different opening lines, one product demo, one honest reaction or answer to a common objection, and a few seconds of clean B-roll. Have the creator record each piece as its own take, with a beat of silence before and after.

Shoot vertically at 9:16 and keep key text and the creator's face clear of the platform interface. The TikTok ad specs page shows where buttons and captions sit. Film all three openers in the same outfit, the same light, and the same room. That way an editor can drop any opener onto any body and the ad still feels like one continuous clip.

Add a one-page agreement before the shoot. It should name paid usage, platforms, duration, and whether the creator allows AI edits such as captions, translation, or voice changes. Never assume voice or likeness cloning is covered. Ask for it explicitly, in writing, for each creator. It takes ten minutes and saves a takedown later.

  • Three openers, each 2 to 3 seconds long
  • One demo showing the product in real use
  • One reaction or objection answer in the creator's own words
  • 5 to 10 seconds of clean B-roll with no music
  • Raw audio, so editors can swap lines cleanly

What long-running pet ads open with

Look at what ran longest. In the Meta Ad Library, the pet category holds 3228 video ads from 372 advertisers, and the median ad ran 44 days. Against that, FurryFreshness opened with "Permanently Removes Pet Stains and Odors!" and ran 800 days. WINPRO Pet ran "See How Plasma Can Help Your Dog" for 433 days. JustFoodForDogs ran "Your Dogs New Favorite Food" for 342 days.

The pattern is one promise per ad. Each opener names a single result or a single reason to keep watching, in plain words, with no setup. A stain disappears. A dog gets help. A food becomes a favorite. Wild Earth took a different route and led with a creator story, "From the Te Araroa to the PCT: Cam Bostock on Adventure, Content Creation & Community", and ran it 548 days with a Learn more button. Promise-led and creator-led openers both lasted.

Treat run length as a signal, not proof of performance. It tells you which angles advertisers kept paying for. For an agency, the lesson is to build a durable angle library, then refresh the hook and the creator around each angle. The Sprites ads library shows real ads from a category, so pull your client's category before you write briefs. For pet clients, start with pet ads.

Want pet video ads like these for your product?

The pet video ads that kept running, read from the ad libraries. Every change waits for your approval.

Turn client research into briefs

Feed the AI assistant only material the client has approved: customer reviews, support questions, past ad results, and the product page. Ask it to cluster everything into reasons to buy. A travel backpack brand might get three angles: fits under the seat, easier to pack, and quick access at security. Each angle becomes its own brief.

Give the assistant an approved claims list. Tell it to use only those claims and to flag any idea that would need substantiation, rather than inventing proof, numbers, or results. Ask for one brief per angle with three opening lines, a demo beat, an objection answer, and a call to action. Review the output as a draft, never as final copy.

Keep the creator's voice in the final lines. AI briefs tend to produce polished slogans, and polished slogans kill UGC. Pass review phrases to creators as prompts, not as lines to read, then keep the take that sounds like a person talking. Customer language from real reviews often makes the best opener.

Build a test matrix that stays small

Use a simple grid for the first batch: 3 angles x 2 creators x 2 hooks = 12 possible cuts. Those twelve are a menu, not a launch list. If the client's budget can't give each ad enough spend to teach you something, launch fewer. TikTok and Meta both recommend diversifying creative, and a different reason to buy is real diversity. A new caption color is not.

When you want to know whether a hook made the difference, keep the creator, body, and offer identical and change only the opening. Comparisons across different creators or offers are directional at best. Say so in the weekly review so the client doesn't read a creator difference as a hook difference.

Once an angle shows promise, build follow-up tests around it: new hooks, a second creator, a different demo. This is where AI saves the most hours. Use it to transcribe footage, find usable takes, draft captions, and assemble rough cuts. Polish only the cuts that earned a second round.

Where AI speeds up the edit

Put AI in the slow, repetitive steps. Transcribe every raw clip and search the transcripts for the lines you want. Let it flag the cleanest takes, draft captions, and assemble a rough cut from your tagged pieces. A human editor then trims, fixes pacing, and checks that the first second is clear on a phone with the sound off.

Use generated video for supporting shots only, such as abstract B-roll, a product-in-scene insert, or a background for a text card. Keep it 9:16, keep each clip to a few seconds, and avoid real people and brand logos. Don't use a synthetic presenter as a stand-in for a real customer. For the wider format, see the guide to UGC ads.

Name every file with the same tag pattern: angle, creator, hook, offer, version. For example, fits-under-seat_creatorA_hook2_offerFreeShip_v1. Tags let you reassemble any ad in minutes and let the media buyer trace results back to one piece. Without tags, AI just gives you a bigger folder to dig through.

Give production and buying one scorecard

Use the same tags in the ad name inside the platform. In the weekly review, use early-view metrics and click-through rate to diagnose where attention drops. If viewers leave in the first seconds, change the hook. If they watch but don't click, change the demo or the offer.

Judge winners on the client's business goal, such as purchase CPA or qualified-lead cost. Don't declare a winner off a handful of conversions. Also track two internal numbers: cost per approved concept and cost per winner. They show whether AI is improving your margin or just increasing output. If output doubles and cost per winner doesn't fall, you bought more ads, not better results.

Approve first, then pilot for a month

Put a human reviewer between every cut and the launch button. They check claims against the approved list, captions for accuracy, disclosures, and paid usage rights. They also confirm the ad against each platform's current rules on AI-generated content, since those change. Get explicit permission before cloning a creator's voice or likeness.

The U.S. FTC's rule prohibits fake or false testimonials, including AI-generated ones that misrepresent who had the experience or what happened. So never script a review the creator didn't live, and never present a synthetic presenter as a real customer. Real experiences stay real. AI handles the packaging around them.

Run the whole workflow for a month on one client, one offer, and one platform, using your existing editor, a shared asset tracker, and an AI assistant. Compare production hours, approval time, and cost per winner against your previous process. Then automate only the step that is actually slowing you down. If you'd rather hand off the weekly output, Director by Sprites is a full creative team for your startup. Every Monday you get finished, on-brand video ads, reviewed by a human.

Prompts to copy

Paste each one as written, then change the product, the setting and the light to match your own footage.

  1. Brief: angles from approved research

    You are a paid social strategist for an agency. Using only the approved reviews, support questions, and claims list below, group the material into 3 distinct reasons to buy. For each angle write 3 opening lines in plain spoken language, one demo beat, one objection answer, and a call to action. Use only approved claims. Flag any idea that needs substantiation instead of inventing proof, numbers, or results. [paste research and claims list]
  2. Hook: product reveal B-roll

    Vertical 9:16 video, 4 seconds. A hand lifts a compact gray travel backpack from the floor and slides it under an airplane seat in one smooth motion. Soft natural window light, handheld camera feel, shallow depth of field, no faces, no text, no logos. Keep the top third of the frame uncluttered for captions. Realistic, unpolished, phone-shot look.
  3. Demo: packing insert

    Vertical 9:16 video, 5 seconds. Top-down shot of a packing cube being zipped shut on a bed beside folded clothes, then a quick pull-back to show the bag closed flat. Bright daylight, casual home setting, steady hands in frame, no faces, no brand logos, no on-screen text. Keep the product centered and away from the bottom 20 percent of the frame.
  4. Objection: quick access insert

    Vertical 9:16 video, 4 seconds. Close-up of a zippered laptop pocket on a backpack opening in one pull, then a slim laptop sliding out onto a gray tray. Cool indoor light, shallow depth of field, generic unbranded gear, no faces, no logos, no text. Realistic motion with no warping of zipper teeth or straps.
  5. Edit: transcript to rough cuts

    Here is a timestamped transcript from one creator shoot. Find the 3 strongest opening lines under 3 seconds, the cleanest demo take, and the best objection answer. Return a table with timestamp, quote, and why it works. Then propose two rough-cut orders: opener, demo, objection, call to action. Do not rewrite the creator's words and do not add claims that are not in the transcript. [paste transcript]

A weekly UGC batch for one client

StageOutputAI's jobHuman's job
ResearchAngle briefsCluster reviews and draft briefsApprove the claims list
ShootOpeners, demo, objection, B-roll takesDraft the shot listCreator films the real experience
EditRough cuts of each tagged combinationTranscribe, pick takes, draft captionsTrim pacing, check the first second
LaunchTagged ads in the platformName ads from tagsCheck claims, disclosures, usage rights
ReviewWeekly scorecardSummarize results by tagPick winners on the business goal

Mistakes to avoid

  • Launching all 12 cuts on a budget that can only fund three leaves every ad with too little data to read.
  • Changing the creator, body, and offer at once and then crediting the hook for the result.
  • Letting AI invent proof, numbers, or testimonials instead of limiting it to an approved claims list.
  • Using a synthetic presenter as if it were a real customer, which runs into the FTC rule on fake testimonials.
  • Cloning a creator's voice or likeness without explicit written permission.
  • Declaring a winner after a handful of conversions or judging it on views instead of the client's goal.

Rather have the videos made for you?

Director by Sprites is a full creative team for your startup. Every Monday you get finished, on-brand video ads, reviewed by a human. You never write a prompt.

Apply to Director by Sprites →

Questions

Can an agency use AI-generated people in UGC ads?

Not as fake customers. The U.S. FTC's rule prohibits fake or false testimonials, including AI-generated ones that misrepresent who had the experience or what happened. Use AI for editing, captions, and B-roll, and check each platform's current policy on AI-generated content before you launch.

How many UGC ad variations should I test at once?

Plan from a grid such as 3 angles x 2 creators x 2 hooks, which gives 12 possible cuts. Launch only as many as the budget can support. Start with a few distinct concepts, then run follow-up tests on the ones that show promise.

What should I ask creators to film?

Ask for three opening lines, a product demo, an honest reaction or objection answer, and clean B-roll, each recorded separately. Shoot vertically and keep key text away from the platform interface. Separate takes let editors swap an opener without rebuilding the ad.

Which metrics should I use to pick winners?

Use early-view metrics and click-through rate to see where attention drops. Pick winners on the client's business goal, such as purchase CPA or qualified-lead cost. Wait for more than a handful of conversions before you decide.

Can I clone a creator's voice with AI?

Only with explicit permission. Put voice and likeness use in the creator agreement, in writing, along with paid usage terms. If a creator hasn't agreed, don't clone their voice or face.

How do I know AI is actually saving my agency money?

Track production hours, approval time, cost per approved concept, and cost per winner against your previous process. If output rises but cost per winner doesn't fall, AI is adding volume without adding results. Automate only the step that is slowing you down.

Keep going

Get pet video ads made for your brand

  • The pet video ads that kept running, read from the ad libraries
  • Finished, on-brand video ads, reviewed by a human
  • Every change waits for your approval