If you’ve spent any time in Google Ads recently, then you’ve probably got an invitation to try out the new responsive search ads beta.
They make it sounds great: “You provide the content and Google's machine learning will assemble it into ads that can improve performance.”
In reality, it also pretty easy to set up. You enter up to five headlines and two descriptions.
Then, Google will mix and match them all until it can determine which combination is the best possible way to get your point across to consumers.
If that sounds cool it’s because it is cool. Machine learning is awesome AF when it makes our lives easier.
But does it work? Sometimes beta technologies straight up suck, which can end up losing you money or just making you look silly. And that brings us to this week’s story: Machine Learning In Advertising
Being the curious person I am, earlier this year I turned on dynamic sitelink extensions in my Google Ads account.
For the unacquainted, sitelinks extensions are extra links on your Google Search Ads that you can use to promote additional landing pages with the same ad.
Normally, you set these yourself, but with dynamic sitelinks, Google promises to automatically “link directly to conversion-focused sections or pages of your website that promote user action.”
I turned on the feature and went and made myself some nachos.
A month later, I was looking at the performance of one my campaigns and found that dynamic sitelinks had generated about 186 clicks for me to the tune of $112. For context, I spent a total of $369 during that time frame, so about a ⅓ of my budget went towards dynamic sitelinks.
At first I think to myself, way to go Googs! Thanks for going above and beyond to rustle me up some of them clicks.
But then I looked at my landing page and realized that it was a single page website that didn’t link to any other page and that lived alone on a subdomain of my main domain.
Where was Google getting my sitelinks?
I start Googling and sure enough, I see my ad with three sitelinks added with the anchor text, Pricing Information, View Features, and Page Templates.
Now, I don’t have the kind of money where I can just go around clicking my own ads all the time, but I don’t have any pages on my site that remotely resemble the pages featured in my sitelinks. So I clicked it.
Surprisingly, I landing on the pricing page of Leadpages.net.
Apparently, Google’s super fancy robot logic figured that since I was using Leadpages software to build my landing page, I’d probably want to go ahead and just send them a third of my paid traffic, too. P.S. You’re welcome, John and Jason.
I thought myself, I gotta tell Google about this.
So I submit a ticket with the screenshot I just showed you all and get a response saying my request would be escalated.
Awesome. I feel important. I’m escalated.
The next day, Mr. AdWords Support writes me back and says that the screenshot wasn’t enough and that their engineers need the click string of the ad to investigate the issue further.
Again for the unacquainted, the click string is the tracking URL used in ads and you can only get them if you spot your ad out in the wild.
So I sat there and Googled my click strings for a while. I’d already put this much time into getting $112 back. Why stop now?
As luck would have it, I was able to trigger the suspect sitelinks again and sent Google my findings.
Wait another day. The response comes, and apparently now the click string is also not enough. He says, “To escalate this issue further to the engineering team I would need a screencast/ video confirming this behavior.”
Now, I’m a pretty reasonable guy, but at this point, I start to think of the millions of marketers out there who helped Google earn more than $28 billion dollars in ad revenue in Q2 of this year alone.
If I lost a ⅓ of my ad dollars to a bug, What does that make, $9 billion in revenue from some bogus dynamic sitelinks? Admittedly, I was losing my patience.
So I shot back an email, again asking for my money back.
Only this time I didn’t just complain about a system glitch that they couldn’t do anything about. This time, I kind of suggested that Google defrauded me and I feel like I showed them enough evidence to get a refund.
What else do you want, a picture of me holding today’s newspaper and posing in front of my adwords account?
Well my little mention of the f-word must have gotten someone’s attention. Not more than a few hours later and I got a final email saying that my refund was approved.
The moral of the story? While it’s tempting to let machine learning take over your advertising, turns out, they’re still pretty bad at it and might end up sending a third of your budget to someone else’s website if you’re not careful.
So please, sign up for all the betas, just keep a close eye on your campaigns. And don’t be afraid to stand up for yourself if you have the data to show you're right.