The problem with AI suggestions on social ad platforms

Are you considering using the AI-generated ad copy suggested by Meta’s social media platforms, Instagram and Facebook? Here’s why you shouldn’t.

I was recently tasked with recommending copy for an Instagram ad designed to generate leads for a client’s email list. Grabbing my trusty playbook of best practices I reminded myself of the most important parts of writing effective ad copy for driving direct action:

1. Make the offer immediate and apparent
2. Say the important bit in as few words as possible
3. Ensure the part a user would care about doesn’t get truncated by an ellipsis
4. Preserve brand voice as much as possible

Pretty easy task, I thought to myself. I’ve put together dozens of pieces of ad copy for stuff like this. I settled on a functional and straightforward, if a bit bland, offer.

It’s not going to win any awards, but it’s certainly acceptable for the purpose of getting site visitors to subscribe to a newsletter.

Meta then recommended to me what I can only describe as industrial slop. And if I wasn’t happy with the six or so recommendations, the “Refresh variations” button was right there for me to generate a fresh crop of slop to wade through.

How Meta explains its ad copy “variations”

Meta is clearly not very confident in the output of its system, either, evidenced by the disclaimer that they do not “make any warranties regarding the completeness, reliability and accuracy of the suggested text variations.” Here’s another quote from Meta’s documentation:

“The text variations feature in Meta Ads Manager generates up to 5 variations of text based on the primary text inputs for your ad. Text variations highlight keywords from input phrases you may want to emphasize, enabling you to generate diverse text that increases customization with less effort.”

In typical fashion, Meta’s documentation on the subject isn’t correct because it actually generates six variations. Small nitpick? Sure, but as anyone who has tried to get an answer to an obscure error message on Meta Ads can attest, incomplete or incorrect documentation is endemic and abundant.

Meta’s suggestions—and why I’m not using them

Previous issues aside, I want to focus on the phrase “diverse text that increases customization with less effort” because it shines a gigantic spotlight on how the AI industry views people.

I invite you to read the variations presented below and think about how these ads may perform:

“Wow!” I can hear you say out loud to your computer screen. “Look at all these variations, with such diverse and interesting takes on what I had initially suggested!”

Oh? You’re not exclaiming that? Are you thinking that they aren’t substantively different from what I put in at the start, and therefore are creating variation for variation’s sake and are unlikely to improve performance? Plus you think it’s tacky that they tried to invent two hashtags that are so cynically corporate you threw up in your mouth a little?

Yeah, me too.

What’s the issue?

The big issue with variation for variation’s sake is that you don’t learn anything useful. You can easily tell when one version of an ad does better than another by comparing whatever KPIs you want and seeing which one is higher or lower. 

But what the numbers can’t tell you is why certain ad copy performed differently. That’s inherently a human-intuition problem. It’s our job as marketers to put in variables that can result in actionable insights to the system. When the platforms spit out numbers, we can interpret what those numbers mean, and not just what they say. What the AI-focused approach is forgetting is that marketers are striving to learn more about people. And people are more than just version A or version B.

There’s no one “right” way to structure the copy, either. There are infinite ways to phrase your ads to achieve similar results, but what matters more for testing is figuring out how you want to talk to customers. This will result in more diverse copy and creative compared to what a generative system can pull together based on your current input.

You have to come up with something new.

Our advice: Take the human approach

I’m not saying that AI doesn’t have its uses, because of course it does. But when it comes to instances like this, you’re much better off relying on your experience and strategy to determine what works and, more importantly, what doesn’t.

What if you don’t have that experience yet? Can’t you at least lean on generative AI to point you in the right direction? Sure, but only if you’re confident that you can take control of the wheel and not let AI drive you into a ditch full of terrible hashtags.

To help you feel more in control, I’ve provided this handy list of questions to consider when testing ad copy.

Create ads that resonate

AI-generated ad copy often produces bland, repetitive content lacking any semblance of strategy. While it may offer variety, that same variety will struggle, if not fail, to provide meaningful differentiation or actionable insights. Effective marketing requires human intuition to interpret data, understand audience nuances, and craft compelling messages.

At Ayima, we specialize in Paid Media campaigns with impactful, tailored content that speaks to your audience and delivers tangible outcomes. Some of our core beliefs include:

• Human-crafted ad copy that captures your brand voice
• Strategic insights beyond A/B testing
• Expert interpretation of data to inform future campaigns

Let’s create Paid Media campaigns that truly connect. Email us today to get started.