How should I test new Meta ad creative for a newsletter?

I think in concepts, not individual ads. My durable winner is the local event angle. We change the terminology and the actual video, but it's always the same general concept, and I refresh it with small variants rather than replacing it.

What's the testing ladder?

1. Tier one: tiny tweaks. If it looks visually the same, change the copy, what's highlighted, or the order. Changing one highlighted word from blue to orange and underlining it moved real numbers.
2. Tier two: bigger swings inside the same concept.
3. Tier three: something completely different, not even about events. I haven't tested those in any real volume.

Change one small thing at a time before you scrap a concept. Even if all you changed is one little thing, put it in as its own ad. Another newsletter operator I know runs 14 creatives in one account with the same text on top and a different background video on each. He was still around 35 cents a subscriber after adding 12,000 in three or four months.

What counts as a winning test?

About the same as the main ad. I don't need a new ad to beat my best one. I need stuff that works similarly enough, so the account has more than one ad that works.

When a test came in close to the main ad, I told my manager that was actually good news, because some of the ideas we've tried have been noticeably worse. Finding a strategy that's different enough but still successful is great. If the main ad creeps from 40 to 50 cents, it's still good.

What I don't want is one cheap hero ad and everything else sucking. That means we're one-trick ponies. For consulting clients it matters even more. If all I do is hand them one or two ads, they'll figure they can let those run on their own.

Why not just try wild new ideas?

"Most people doing this just keep doing variations of the same thing. It's working. I could try something more creative, but if it costs me 40% more money, I'm just going to go to what works."

I keep making variations of the winner until the market is saturated, and test wilder ideas in a small separate campaign. The main campaign stays untouched.

How much should a test cost?

Not much. A quick test: two different videos, different text on each, $5 a day apiece, and see which one works.

When I added two new variants to one account, I kept the existing ad running because it was working relatively well, dropped it to $10 a day, and gave each new test $10 a day. That was more than I wanted to spend, so the plan was to pull it back by Monday at the latest. I didn't obsess over matching every line of copy. As long as the text fit the audience, employee or entrepreneur, any decent option was fine, on the same background video.

• Test two copy variants at once on the same background video, so only the words differ.
• Test first on the brand that cares least, or a client brand where learning matters more than squeezing every dollar. There, more creatives, more text, and more locations beat any single perfect test.
• Only bring proven variants into your main brands.

How long should a test run?

Long enough to read it. Day to day, I review the tests the next day. Whatever is working, leave it. Anything running at more than double the cost, turn it off.

Close calls get more patience. On one account I wanted to change an underperforming test ad, but agreed to wait until it got past the roughly 50-conversion learning threshold first. My media buyer staggers creative changes about ten days apart, because changing too often hurts performance. On a small budget, too many creatives at once means none gets enough delivery data.

Early on, my assistant held off launching a better background video because he didn't want to mess with a working ad. A Meta consultant told us to duplicate the ad set and test the new creative there, so the original keeps running.

What's on my test list?

• A fake text message. Two friends: "Did you hear about that event Friday? It was awesome." "No, what do you mean?" "I found it in [the local newsletter]." "What do you pay for that?" "It's free."
• New national brands: start with one short video and one image ad, using my usual model. A person talking to camera comes later, and probably fits a homeschool audience better than insurance.
• Running ads from a person's profile instead of a logo page. Enough people in my space say it's 20% to 40% cheaper that it's worth a test, but I haven't tried it. Launch the normal way first.
• Concepts with no events at all for event-driven brands.
• Real-person UGC versus our AI-presenter versions.

For when to refresh in the first place, see how long a Meta ad lasts, and for the campaign setup, main and test campaigns.