Ad Creative Velocity on Meta Ads: How Many Creatives You Should Actually Ship Per Week (2026)

Why ad creatives are the second most important thing in Meta Ads

If you've been living under a rock, ad creatives aren't that big a deal. If you haven't, you already know they're basically the whole game now. Once your conversion tracking is clean and your campaign setup isn't broken, all that's left is creatives and landing pages. That's it. That's the entire industry at this point.

My priority order in 2026: conversion tracking first (covered that here), ad creatives second, landing pages and funnels third. That order matters. Tracking is what makes everything else actually measurable. Without measurement your creative tests are just vibes.

A few years ago this wasn't the case. Clients would hire me to run the media buying, hand me their creatives, and I'd go do my thing. That doesn't cut it anymore. Now I'm involved in every creative that gets produced. Angles, scripts, offers, setting, location, lighting, wardrobe, pacing. That's just what it takes to make this work in 2026.

One Sora ad, $1.2M in revenue, 49 flops before it

A few months ago I made a client $1.2M in revenue from a single Sora ad. Before that winner, I ran 49 other Sora ads for the same client. Most were fine. A few were decent. None of them moved the needle like the one that hit.

This is how it actually works. You ship a lot, most don't work, one does, and the one that does pays for everything. People who've never done this at scale think creative is about making "the right ad." It's not. It's about making enough ads that one of them happens to be right.

What changed with AI video is the cost of trying. Each of those 50 Sora ads cost maybe $4 in API credits. The script was mine. The prompt took 5 minutes. If I had produced those same 50 ads with human actors, it would have cost the client $50,000 and taken 6 months. With Sora it cost about $200 total and took a week.

That's the shift. Not "AI video is better than human video," because it isn't yet. The shift is that AI video makes creative volume cheap enough that you can actually find the winner in the first place.

How many creatives should you actually ship per week?

Depends on your spend. I know that's the answer you didn't want to hear, but it's the real one.

Here's roughly what I run across my client roster:

• $150K/month on Meta: 15+ new creatives per week
• $50K/month: 8-10 per week
• $20K/month: 4-6 per week
• $5K/month: 3-5 every two weeks
• Under $5K/month: honestly, 2-3 per week is fine, but you probably have bigger problems than creative volume

The variables that push these numbers around: how fast your ads fatigue (function of your audience size and budget), how big your addressable market is, how concentrated your spend is on winning ads, and whether your targeting is narrow (fatigues fast) or broad (fatigues slow).

For most service businesses spending $10K-$50K/month, shipping somewhere between 5 and 15 new creatives a week is the right range. Below that you're fatiguing faster than you're replacing winners. Above that is waste unless you're already spending a lot and scaling fast.

Static image ads vs video ads for service businesses

Video generally wins for my clients. I know it's supposed to be a hot take that "it depends," but honestly in 80% of my accounts video outperforms static. And not just on CPL. On cost per sale and ROAS too, which are the metrics I actually care about.

This isn't a blanket rule. Static image ads still work. They're faster to produce than any video. A lot of the leads and sales I've made my clients over the years came from static ads.

What static ads can't do: expand on a topic, demonstrate expertise, or walk through "how it works." A static ad's job is to hook for a second, grab attention, and get the click. Then your landing page does all the education and trust-building.

Video does the harder job. It can tell a story, build credibility, show a face, walk through an offer, explain a result. For service businesses selling anything that requires trust before the call, video is almost always the bigger mover.

If you're stuck picking one: test both, but spend most of your creative budget on video. Use static for fast tests and broad-reach plays.

How I prompt Sora (handheld iPhone, no text, natural language)

Two rules that make Sora actually useful for ad creatives.

One: always tell Sora "handheld iPhone video." The default cinematic look Sora produces screams AI. Handheld iPhone footage doesn't. That single instruction fixes 80% of the "this looks like AI" problem.

Two: never ask Sora to put text on screen. It hallucinates. The text comes out garbled, misspelled, or nonsense. If you need on-screen text, add it in post. Sora handles speech and visuals well. It cannot handle typography.

A few other things that help:

• Use natural language, not prompt-engineering jargon. "A guy in his 30s sitting on his couch talking to the camera about how stressed he was before he found X" works better than structured templates.
• Specify setting, lighting, wardrobe, vibe. "Late afternoon, soft natural light from a window, casual t-shirt."
• Keep individual shots under 10 seconds. Longer shots drift and the character starts morphing.
• If a shot has a specific person, describe them in enough detail that Sora gives you the same person across multiple takes. Otherwise you'll get a different face every time.

Quality with these rules is honestly insane. The videos look good, they sound good, and most people can't tell they're AI unless they're looking for tells. And the cost is maybe $4 per finished ad in API credits.

Do AI video ads beat humans?

No. Humans still win on CPL, cost per sale, and cost per qualified lead. Every single test I've run confirms this. AI video has closed the gap a lot, but it hasn't beaten real humans yet on performance.

The catch is speed and cost. A real human actor ad takes me 5 days minimum from script to delivered video. Going through a UGC creator network, 2-3 weeks. Casting a specific face, longer. AI video takes an hour.

So the real math isn't "which performs better per unit." It's "which performs better per unit of time AND money." On that math AI video wins for creative volume, and human video wins for the winning concepts you want to scale hard.

What I do across every client now: most of the creative testing runs on AI (Sora for video, Midjourney for static). When something hits, I commission a human version of the same concept to go deeper on the winner. AI finds the winner fast. Humans let you scale it further.

What happens when Sora shuts down April 26, 2026

Sora is shutting down its consumer product on April 26, 2026. Bad news if you've been using the ChatGPT-Sora interface to make ads like I have.

The good news: the Sora API stays live until the end of September 2026. About 6 months from when this article was published.

I built my own tool with Claude Code that talks directly to OpenAI's Sora API. A simple script that takes my prompts, hits the API, and pulls the videos. Works the same as the ChatGPT interface, except it keeps working after the consumer product dies. And it was stupidly easy to build. An afternoon of work.

This is a recurring theme for me. A lot of the tools I used to pay for, stitch together, or wait on vendors to build, I now just build with Claude Code in a day. The first-party tracking pixel from the conversion tracking guide came from the same place. Problems that used to be complicated are now trivial.

After September 2026 we'll need a new AI video model. By then there will probably be 10 of them. The workflow stays the same.

How I structure creative testing

I dump everything into the same ad set. Let Meta decide what to spend on. If I hit the 50-ads-per-ad-set limit, I spin up a new ad set under the same CBO campaign. That's the whole structure.

I know this goes against what a lot of media buyers teach. Separate campaigns per angle, structured testing frameworks, strict budget allocation per test, ISO/CBO hybrid whatever. I've tested those approaches. They don't beat "dump everything in one ad set and let Meta decide."

Why it works: Meta's algorithm is better at allocating spend across creatives than you are. The more creative variety you put in front of it, the faster it finds the winners and kills the losers. Manual media buying with 20 campaign edits a day is dead weight in 2026. I'm not saying it doesn't work. Some people I respect still make it work. It's just not my preference and not what I've seen scale.

The time I save from not manually reshuffling campaigns goes into the things that actually matter: offers, angles, personas, scripts, landing pages, tracking, sales. The stuff the algorithm can't do for you.

When you find a winner, ship more (the fatigue trap)

Here's the trap most people fall into. You find an ad that's working. Great results. You scale it. You sit back and enjoy the ride. Then 3 weeks later the CPL starts climbing and you panic.

What happened: the ad fatigued. Every ad fatigues. No exceptions.

Ad creative fatigue works like this. Think about Instagram or Facebook Reels. Meta learns what kind of content you respond to. It shows you more of it. Eventually it's already shown you (and everyone like you) the specific ad you were most likely to convert on. Everyone who would have converted from that ad already has. So the pool of unconverted-but-would-convert people shrinks, and CPL starts climbing.

The only way to combat fatigue is to keep producing different ads. New angles, new faces, new hooks. Give Meta more options so it can find new pockets of people who'll convert.

The rule I tell every client: if an ad is working, that's exactly when you should be shipping MORE creatives, not less. Your winner has a shelf life. Your replacements need to already be in testing before the winner dies, not after.

Scaling without a creative pipeline running in parallel is how people end up with one great month followed by three terrible ones.

The ethnicity variation trick

A trick that's worked for me consistently: when I find a winning video ad, I produce the exact same ad with the same script but a different presenter ethnicity. White, Black, Asian, Hispanic. Same message, same setting, same hook, different face.

Why it works: Meta's algorithm segments audiences in ways you can't see. A winning ad with a white presenter might be converting really well with one segment while completely failing to land with another. Swap the presenter, and you unlock the segments the first version couldn't reach.

In practice this often doubles or triples the addressable audience for a winning concept without changing the concept itself. Same hook. Same script. Just more of the market can see themselves in the ad.

This is way easier with AI video than human video. Swapping ethnicity with Sora is a one-word change in the prompt. With human actors it's a new shoot, new talent, new schedule, new budget. Another reason AI video has changed the economics of creative testing.

Where to find real creators if you go the human route

When I need actual humans, a few places I source:

• Fiverr. Fine for basic spokesperson videos. Quality varies. Cheap.
• UGC creator networks. Billo, Insense, JoinBrands. Better quality than Fiverr. Takes 1-2 weeks.
• Recruit on Meta itself. Run a small Meta campaign targeting content creators offering paid UGC gigs. This gets you the best quality creators for your niche, because they're already on the platform creating content.
• Cast specific faces. If you have budget, use a casting agent. Most expensive, highest quality, longest timeline.

For most of my lead gen clients, UGC creator networks are the sweet spot. Quality is good enough to scale a winner. Cost is manageable ($200-$500 per video). Turnaround of 1-2 weeks fits a normal creative production rhythm.

Last thing. Please, please, please, please make sure you're producing a ton of creatives. Every single week. Whatever cadence your budget supports. This is the thing that will make or break your ad account more than almost anything else you do.

FAQ