The Facebook Ads Tip That Changed Our Results

Only change one thing per test. Most ad accounts we get sent have four ads running that differ in four ways at once, so when one of them wins, nobody can say why. Two identical ads and one different headline is a test. Everything else is just spending money in an interesting pattern.

Two ads, one difference, or you learned nothing

We published the video above in 2021 and the rule in it has not aged a day, because it isn't a Facebook rule. It's an arithmetic rule.

Say you run two ads. Ad A has your original headline and image. Ad B has a new headline and a new image. Ad B wins by a mile. Congratulations, you know nothing. Was it the headline? The image? Would the new headline have won harder with the old image? You paid for that data and it answers nothing you can act on next week.

Now run the same two ads with identical images and only the headline different. This time you own a fact: that headline sells better to that audience. Facts stack. You carry the winner into the next test and put it up against a new image. Six weeks of that and the ad you're running is nothing like the one you started with, and every piece of it beat something.

It sounds too simple to matter. It is simple, and almost nobody does it, because making three near-identical versions of an ad and three matching pages feels like admin. That's a software problem, and we removed it.

What we meant by "test time"

We had a name for this internally. Test time, because 100 percent of the time you should be testing something. Not a phase you enter when an ad stops working. A permanent condition of the account.

The reason is unflattering: what we think will work best usually doesn't. We've spent over $3 million across Facebook, Instagram, YouTube and Google, and we're still wrong about our own ads regularly. A headline we're certain about goes up against two throwaways written in ninety seconds, and the throwaway wins. That happens often enough that we stopped having opinions and started having tests.

The clearest one we ever ran was three headlines on one ad, everything else locked. Headline two came back with the cheapest clicks, the most leads and the most purchases. The headline we were confident in, the one we'd have run alone, was on pace for less than half the sales. After that, the first thing we split tested on every launch was the headline, three or four variations, and it kept paying: on one test the two winners came in at half the cost per click of the rest.

If you want the wider set of lessons that came out of that spend, they're in what a million dollars of Facebook ads taught us. And the free one-hour training we made walks the entire funnel these ads pointed at, so if you want the version with the funnel attached, it's the whole seven-figure funnel, free.

What one test should cost you, and how long to leave it

People hear "split testing" and picture an agency budget. The numbers are much smaller than that, and being small is the point.

Two to three days, or about $20 to $30 total, is enough for a first call on most tests. We've reviewed student accounts that opened at $100 a day on something brand new, and that is way too much money for a question you haven't asked properly yet. Start at $10 or $20 a day. You'll know inside about $10 whether an ad is wildly off or roughly on track. On track means let it run and gather real numbers. Wildly off means you just saved the other $90.

The headline test above ran about a week and cost roughly a hundred dollars, because we wanted purchase data and purchases are slower than clicks. That is the ceiling, not the floor. A hundred dollars to stop running an ad at half strength for a year is the best trade in this entire business.

One caveat: testing does not rescue an offer nobody wants. If you have never sold the thing to a human being, no headline fixes that. Build the funnel, send it to your DMs, take the money by hand, then buy traffic for something you already know converts. That's why the funnel templates sit in the same account as your course and your checkout, so the sold-by-hand version and the paid-traffic version are the same pages with more people pointed at them.

The order we test in

There's no legally correct order. This one has worked for years, and it's ranked by how much the variable moves the number.

Test	Versions	How long	Why it's in this position
Headline	3 to 4	2 to 3 days, then a week for purchase data	Cheapest change, biggest early swing. Nothing else moves until it's settled.
Image, or thumbnail on a video ad	3	2 to 3 days	First thing the eye lands on. Winning headline stays locked.
Hook, the first 3 to 5 seconds	4 to 5	3 to 4 days	On video this replaces the thumbnail test, and usually swings hardest.
Ad copy	2 to 3	3 days	Only matters once someone has already stopped scrolling.
Audience	2 to 3 ad sets	Week	Fund each one equally by hand, or you won't know if it lost or was never shown.
Call to action and landing page	2	Week	Last, because a page test needs real traffic, and now the ad sends it.

Run it top to bottom and each round starts from a stronger ad than the one before. Run it in a random order and you'll keep re-testing things against a losing headline.

On a video ad, test the hook instead of the thumbnail

Here's a detail most people get backwards. Headlines flash on screen for a split second before the video starts playing, because nearly everyone has autoplay on. So the thumbnail barely gets seen, while the first three to five seconds gets seen by everybody the ad reaches. That is the hook, and on video the move is to skip the thumbnail test and put four or five hooks against each other instead.

Same rule applies inside the test: same video, same everything, only the opening seconds recut. If you want the anatomy of the rest of the ad after that opening, we broke it down in the five parts of a video ad that converts.

When the numbers disagree, purchases break the tie

Three numbers tell you who won: cost per click, cost per lead, cost per purchase. Most of the time they agree and the decision takes four seconds.

Then there's the test where the cheapest clicks come from one ad and the sales come from another. Confusing, annoying, and the most valuable test you'll run all quarter, because it just showed you your cheap traffic is the wrong traffic. Give it two or three more days to clarify. If it doesn't, keep the ad producing purchases and let the pretty click number go. Leads don't keep you in business. Buyers do.

This only works if you can actually see which ad produced a payment, and that is where most course businesses go blind. Pages on one tool, email on another, checkout on a third, and no single place that connects the ad to the charge. Ours all live in one account with the funnels, the emails and the checkout together, so the winner of a test is a fact and not a vibe. If you'd rather see that wired up before you commit, grab a walkthrough with our team and bring your ad account with you.

Why almost everybody quits after the first test

Rule two in that 2021 video was don't stop testing, and it's the rule people break within a month. Not out of laziness. They stop because round two is where the admin shows up.

Testing headlines is easy, it's one field in Ads Manager. Testing three hooks means three edits. Testing landing pages means building three pages, and if your page tool, your email tool and your checkout are three subscriptions, that's nine things to keep in sync and a real chance you break the one that was working. That's the moment the habit dies.

So we killed the admin. Premium comes with 50-plus funnel and website templates, so a variant page is a duplicate plus one edited headline. Automations and emails duplicate the same way. Small thing, and it decides a big one: whether you're still testing in month four. If you want it without the do-it-yourself part, our vetted experts build the funnel for a flat fee and hand it back ready to test.

One more failure mode we see constantly, and it has nothing to do with ads. People launch a campaign before their sending domain is set up, every lead they buy goes to spam, and the ads get blamed for a deliverability problem. The onboarding call included on Premium and Elite sets up your domain, DNS and email deliverability before you spend a dollar on traffic, which is not glamorous and has saved more ad budgets than any headline ever written.

The one test to run this week

If you're running ads and you have never done this properly, don't rebuild anything. Take the ad you already have and clone it twice.

Change nothing but the headline in each clone. Write one that names the outcome, one that names the pain, one that asks a question. Put all three in the same ad set, $10 to $20 a day, and leave them alone for three days. Read cost per click, then cost per lead, then wait out the week for purchases before you crown anybody. Kill the losers, keep the winner, run the next test against a new image. That's the whole difference between an account that improves and one that just ages.

Do that four times and you have a proven headline, a proven image, a proven hook and a page you can trust, for roughly the cost of one week of unstructured spending. If the page is the part you're least sure about, start with the mistake that kills more landing pages than any other before you test it. Meta's own Ads Manager documentation covers where the buttons live; the discipline above is the part it will never teach you.

And the free training is still the fastest way to see the whole machine these tests feed, front to back: watch the one-hour workshop, then go run your three headlines.

Frequently asked questions

What is the most important Facebook ads tip for beginners?

Test one variable at a time. Two identical ads with one different headline tell you something you can use forever. Two ads that differ in three ways tell you nothing, no matter how big the winner's margin looks.

How many ad variations should I test at once?

Three to four versions of one element. On headlines we always launched three or four. On video hooks, run four or five, because the opening seconds swing results more than anything else in the ad.

How long should I run a Facebook A/B test before deciding?

Two to three days, or about $20 to $30, for a first read on clicks and leads. Give it a full week before you judge purchases, since sales come in slower than clicks. If an ad is wildly off, you'll usually see it inside the first $10.

Which metric should decide the winner of a split test?

Cost per purchase, every time. Cost per click and cost per lead are early signals and they're useful, but the ad that brings the cheapest clicks is not always the ad that brings the sales. When they disagree, give it a couple more days and then back the buyers.

Do I need a big budget to split test Facebook ads?

No. $10 to $20 a day is enough to run a real test, and $100 a day on something brand new is a way to lose money faster, not learn faster. Small budgets are what make testing sustainable enough to keep doing.

What tools do I need to actually run these tests?

Ads Manager for the ads, and one place behind it that holds your pages, emails, checkout and contacts so you can tell which ad produced a paying customer. That's the whole reason we built Course Creator 360 the way we did: duplicate a funnel template, change one headline, and the tracking follows the money instead of stopping at the click. Premium and Elite come with a 30-day trial and an onboarding call that gets your domain and email right first, so set the funnel up there and put your test budget into the test.

Related

• What $1 Million in Facebook Ads Taught Us About Selling Courses
• 5 Steps to a Video Ad That Actually Converts
• The #1 Landing Page Mistake (And The One-Line Fix)