These days, the rate of change in marketing is almost dizzying. I came back to DMi this spring after a three-month maternity leave and found that some products from advertising platforms I’d used for years had become almost unrecognizable – and not always in a good way.

The time away from the keyboard helped me get some perspective about the industry-wide rush to adopt AI – how it helps, how we need to give it guardrails, and where we still need to have our hands firmly at the controls to get great results for our clients.

First, I’ll look at how the major platforms are implementing AI and what marketers need to know about how it’s working. Then, I’ll take a critical look at the options for using AI to build creative and messaging across campaigns.

AI in Google Ads

Google’s AI-powered Performance Max campaigns, which automatically find the “best” placements and users across the Google ecosystem, are what really opened my eyes to the surge of AI. Performance Max, specifically, looked like a different offering altogether when I came back from maternity leave.

Overall, while Google is pushing hard for PMax adoption (for instance, moving people away from Shopping campaigns to PMax), the performance of those campaigns isn’t where it used to be. (We actually had a client come to us having transferred all manual campaigns to PMax and suffering a subsequent drop in performance.) I’m seeing that across the board – B2C, D2C, and B2B. The fact that we’re in an election cycle may be a factor, but I think overall it’s a case where Google has taken controls from advertisers and given them an inferior AI product in return.

With PMax and Demand Gen campaigns, we see AI-created ads producing lower engagement from actual human users – but, often, with higher Quality Score rankings, since PMax is a Google product doing what Google wants. (This is a great argument against optimizing strictly for Quality Score, if you still needed to hear that argument in 2024.)

If you’re just using PMax for Shopping, I don’t see any issue in relying on the targeting (unless you’re seeing horrible performance). High-level shopping campaigns with one or two similar product lines (e.g. you’re just selling hats) are one example where you can just let the AI run – there’s no creative, it’s easy to manage, etc. It’s when you start to segment and drop CPAs that it gets more important to have nuance. That, and you need to make sure PMax isn’t simply finding the easy conversions – for instance, racking up conversions on the lowest-priced item in a big catalog, or converting brand search terms with tons of user affinity already in play. Exclusions and campaign segmentations are your friend when you’re trying to teach PMax how to work for you.

No matter how hard Google is pushing PMax, make sure you’re taking their advice with a big grain of salt. It does have benefits and can be a great addition to what the platforms are offering (for instance, we love integrating first-party data and creating lookalikes for PMax campaigns), but being too reliant on a new product – even a shiny one – can leave your accounts vulnerable. Keep a close eye on performance, and make sure you’re testing and tweaking everything you can to control the output.

AI in Meta Ads

Speaking of election cycles, this is a really tough time to ask Facebook advertisers to give up any control, since CPMs are skyrocketing with the glut of political content jacking up feed costs.

Facebook, of course, is leaning heavily into AI/machine learning with Advantage+ campaigns and AI tools for creative production. Facebook’s audience targeting has improved quite a bit since iOS 14 took a chunk of data away from advertisers, but constant privacy changes have hampered our ability to refine targeting – for instance, today, we can’t effectively exclude people interested in politics from our campaigns, which means those CPMs will stay high.

All of that said, with Facebook campaigns – and to some extent with Google – we do recommend heavily leaning into AI in audience targeting. Use your discretion, though; if you have a super-niche product that is really only relevant to a small subset of users (e.g. parts for heavy machinery), keep a very close eye on automatic targeting and look for ways to refine as needed.

AI in Creative

Before I get into types of content and platform AI features, my general opinion is that quality is highly variable (some of it’s good, some of it’s very poor), and I would never give AI carte blanche to produce your content. It’s a great place to start, and if you’re just starting out or launching a new product and trying to come up with iterations, it can give you fresh blood.

Platform-wise, Google has recently rolled out generative AI ad copy. From what I’ve seen so far, you have to put in a lot of info and keep refining your prompts to get anything usable, which kind of defeats the purpose of AI. It’s a fun toy at this point but nothing to incorporate into your strategy.

Several SEO clients have asked us why they should pay us to write content when they can just put it into ChatGPT and see how the Google algo responds, but ChatGPT just doesn’t have the nuance to produce copy that understands user intent. Even if the content looks good on the surface, once you get into it and think about it and listen to it, it often falls short. Most importantly, it’s not advancing the conversation; generative AI only pulls from what already exists. At a time when topic authority and first-party, expert perspectives are key factors in Google’s ranking algorithm, GPT content on its own is not something to rely on.

Facebook has some fairly flashy creative AI tools, but there’s a lot of room for improvement. We’ve had a few clients test Facebook’s creative AI tools after nudging from Facebook reps, but it ends up not following the brand guidelines.

One very promising (and time-saving) GenAI use case that also needs refinement is text overlays on video creative. If it’s accurate, it can be a big help for advertising capitalization on a trending moment, but if it’s not, it can result in negative sentiments directed toward the brand. In short, it’s another “use with caution” scenario – which you may have noticed follows a pattern.

Final Thoughts

Both humans and AI absolutely do have a role to play in marketing campaigns – and one of the fastest-emerging differentiators humans can flex right now is figuring out responsible, effective use cases of AI that serve the brand (or the client). Putting your head in the sand and pretending AI doesn’t exist won’t get you there, but neither will mass-adopting every new feature without serious QA.

Our approach at DMi is to test, learn, stay current, and stay open to new ideas – and to make sure we’re relying on our team to provide both common sense and human creativity whenever there proves to be no substitute.