Has AI Killed the Landing Page Optimization Star?

We know video killed the radio star.

And as feverishly as Meta is trying to rip off TikTok and BeReal these days…

… it seems Mark Zuckerberg is worried video is coming for him next.

Anyhoo, new tech is always taking aim at the old way of doing things.

And, when that happens, jobs that used to be common – like town criers, elevator operators and soda jerks – vanish like weekend guests in a murder mystery. 

Well, those who do split testing and landing page optimization for a living may soon be collecting unemployment along with the soda jerks.

Why?

Because it turns out AI / machine learning (ML) and split testing your landing pages don’t play well together.

And while this might not totally crush the job prospects of those who do landing page optimization for a living…

… it does require them to take a much different approach to this critical marketing task.

Unfortunately, not many people understand what’s going on here and the huge impact split testing has on ML-driven marketing campaigns.

But you’ll be one of the few in the know if you stick with us for the next couple of minutes…

To set the stage, let’s start here:

Relentlessly A/B split testing landing pages has been key to unleashing the true money-making mojo of many a website.

Quick refresher for those who need it… with split testing, you send some traffic to landing page A, and some to landing page B, to see which one performs better.

You keep the winner. Get rid of the loser. Rinse and repeat.

Over time, it’s the best way to improve the performance of your landing pages.

Simple, right?!

It is… except if you’re sending traffic to your landers from campaigns that use machine learning (as most paid ad platforms do these days).

To understand how ML is changing the split testing game, it helps to understand how these campaigns work.

Essentially ML campaigns work based on a past history of success as, over time, the machine learns what converts and what doesn’t.

The machine uses this historical data to optimize for the keywords, audiences, traffic sources, etc., that drive conversions to improve your campaign’s performance. (Yes, an oversimplification, but do you really want the Ph.D. level seminar here? Yeah, me neither!)

So you’re running your campaigns, and ML is doing its thing… happily optimizing for conversions, driving more and more leads your way.

All is right with the world.

Then you think, “Hey, I bet we could be getting even MORE leads if we improved our landing page.” Which, traditionally, is exactly what you should be thinking (so good on ya… give yourself a Gold Star!).

This leads you to head to your landing page software of choice (our personal preference is Convertri), kick out a new landing page variation, and let the split-testing games begin.

Now here’s where not understanding the ML beast becomes an issue…

… Your machine learning campaign has been sending traffic based on its past history of success with landing page A.

When you insert split testing into the equation, the traffic may go to landing page A, but it could also go to landing page B (or, for you go-getters, even landing pages C, D or E!).

The problem here is machine learning doesn’t know the traffic is going to different versions of the landing page.

All it sees is the traffic it’s now sending to your site isn’t converting in the same, relatively predictable way it had been.

This can cause the ML algorithm to freak out because it doesn’t understand what’s going on. The model it had been using, based on its past history of success with landing page A, now looks broken.

This causes it to start shutting things down in the campaign it shouldn’t, and, as a result, your once high-flyin’ campaign gets Humpty Dumpty’d.

So instead of improving your campaign like it did back in the good ‘ol days, split testing ends up confusing the ML, and a once profitable campaign becomes a heaping mess.

But fear not, dear marketer, this does not mean split testing is dead!

It just requires a different approach.

Here’s how we recommend handling split testing these days when sending traffic from paid campaigns…

1. If your number one goal is to split test landing pages, use campaigns with manual bidding.

You don’t even want to use Enhanced CPC here because that uses ML (sorry to get a little technical, but you fellow PPC Nerds know what I’m talking about).

This isn’t an ideal long-term scenario because there are consequences to not using machine learning bidding strategies.

But if you want to just do an honest head-to-head test of some landers, this is the way to do it.

2. Use Experiments (at least in Google Ads) to test new landing pages.

Using Experiments, you can siphon 10% or 20% of your campaign traffic to a new landing page variation to see how it performs vs. your original lander.

This approach leaves most of your campaign untouched so you can test things without the risk of freaking out the ML and totally destroying your campaign.

3. If Experiments are not an option, then you can set up a new campaign (with a limited budget to start) and use it to send traffic to a new version of your landing page.

Ideally, you want to keep the other parameters of the campaign as close as possible to the original. This won’t give you a perfect apples-to-apples comparison, but none of these options are perfect.

But, unfortunately, until ML gets advanced enough to factor in split testing of landing pages, these are the best options we’ve got.

Our advice… keep split testing. It will always be a critical part of improving your marketing results.

But, when you split test, understand where the traffic is coming from and, if machine learning is involved, use one of the strategies above.

Otherwise, you may destroy a once profitable campaign and soon find yourself standing next to the soda jerks and town criers on the unemployment line.  

Share this article
Adam K
Adam K

Adam has been fascinated with online marketing, particularly PPC, since 2004 and opened his own PPC management company in 2006. Over the years he's written extensively about Google AdWords and online marketing on his own sites as well as partnered with/written for Perry Marshall, Ryan Deiss of DigitalMarketer.com and Neil Patel.