Can I use AI-generated video in Meta ads for my newsletter?

Background video exists to stop the scroll. AI can make that.

"Nobody's signing up because they see people cheersing at a fun, you know, first Friday thing."

A drone-style shot of a kids' baseball game generated with an AI video tool is indistinguishable from real footage at that size. If the background is just there to catch the eye, AI is fine.

Where do you draw the line?

• Fine: AI background footage.
• Not for me: AI people talking to camera pretending to be real.
• Never: an AI avatar as the face of the brand. AI is fine for ads. For the actual face of my brands, I record myself.

I'm very AI-forward, and the human AI videos are the ones I'm not a fan of, because people can still tell. To be fair, our AI-presenter ads have worked better than you'd think.

Why do most AI clips fail as ads?

Usually it isn't the quality. When I reviewed a batch of AI clips from several video models, a lot of them looked good. They just weren't engaging. Two people walking slowly through a farmers market doesn't stop anybody.

• Not an event. In one batch, only the concert clip was an event. Everything else was people doing random things, which isn't that helpful for an events newsletter.
• Skylines give it away. A downtown with a very specific look tells people this isn't where they live, before they even look at the people. Our best one is just walking around a fair, where there's nothing to see but people at a fair.
• Ad-looking text. Use a casual font instead of professional-ish text that makes it obviously an ad.

What if my audience notices?

Change course. On one event community's ads, the audience was noticeably complaining about AI-generated video. I moved that creative to a photo slideshow and carousel of real event photos.

Keep the complaints in proportion, though. After one Facebook ad with AI background music, I got about six comments about AI, five of them on the ad itself. That was out of roughly 140,000 people who'd seen it. A peer on that call pointed out that the loudest critics are often loyal readers who share the newsletter anyway.

My slideshow recipe:

1. About 10 real event photos, built in CapCut.
2. One to 1.5 seconds per photo.
3. Lead with people, like people hugging.
4. Include women in the first three photos for that audience.

The opener rule holds either way: no empty venue shots and no AI-animated stills as the first two seconds.

Should I start with AI or wait for real footage?

Start with AI. A teammate told me AI videos had gotten worse numbers than real ones so far, and my media buyer has warned about the same thing in past Meta tests. Not proven either way.

My answer: we don't need perfect, we need solid, and then we build amazing on top of solid. AI is something we control and can make today. Once we know which concepts work, we go get real B-roll of those. For paid-subscriber ads that was the plan, and my AI background variants started in the test campaign, not the main one.

How do I make AI footage look less AI?

Ask for handheld camera work and direct cuts, and vary the cast. I wrote that up in how to prompt AI video for ads.