How We’re Using AI for PPC

There is no more popular buzzword than AI right now. No matter where you turn every business seems to claim they are an “AI” company. And this is no different for PPC agencies. But what does that mean really?

More specifically, this article will explain in some detail exactly how we use AI at Ten Thousand Foot View (TTFV). Before jumping in I want to make a few things clear. First, AI for PPC is nothing new, it’s simply a progression from “automation” which has been used by ad platforms, agencies, and advertisers for many years.

Second, our philosophy at TTFV is to use AI passively. That means we do not allow AI to make unapproved changes at all, excepting where we do not have control over that. I’ll touch on those instances below. Importantly, I see AI as a support tool at this point, not a replacement for a PPC expert. That may change in the future, but we’re still pretty far away in my opinion.

PPC Research

Particularly important at the beginning of an engagement, we’ve started using Chatbots such as ChatGTP and Gemini. The information we can gather supplements research performed with SEMrush, transparency tools, and in-platform auction insights. For the most part, we use the Chatbots to:

• analyze/summarize competitor and client landing pages
• perform SWOT analysis
• generate and organize keyword ideas

Automated Bidding

Over the years automated bidding has improved to the point where it is really only edge cases where manual bidding makes any sense. With the ability to drive far better performance and to free up management time for more important work, it’s a no brainer to run automated bidding across the vast majority of our client campaigns. While we have no way to “approve” automated bidding we do have several controls in place to closely monitor performance so we can quickly troubleshoot if/when things go off the rails.

Ad Copy & Assets

All ad platforms now have the ability to generate ad copy ideas. These ideas can be reviewed and selectively applied (what we do at my agency) or applied automatically. We have chosen not to let PPC platforms takeover control of ad copy because we frequently see the following issues:

• ad copy that’s for the wrong offer, i.e. more suitable for a different campaign or ad group
• ad copy that’s nonsensical
• ad copy that’s irrelevant, e.g. for a service, benefit, or feature the advertiser doesn’t offer
• ad copy that’s not allowed, e.g. including a phone number in a headline which is against the platform’s own rules and will get the ad disapproved

As such, we believe that human intervention is required to ensure only quality ad copy and assets make their way into campaigns. Furthermore, we utilize a specific pinning strategy to ensure that RSA (Responsive Search Ads) copy combinations generate cohesive ads. For more on this you can read our article on RSA optimization.

With respect to assets (formerly called ad extensions) we generally deactivate several automated asset types. In particular we kill sitelinks because they often end up pointing to pages that have a snowball’s chance in hell of converting. For others, such as callouts, we don’t allow those for the same reason we don’t allow auto applied headlines and descriptions.

Despite the unreliability of automatically generated ad copy, Google and others are already jumping ahead to generate completely novel ad copy at auction time. They claim better performance but at what cost? Importantly, you cannot even see a history of what ad copies were generated. We absolutely need better transparency and controls around this type of thing until AI does a better job of writing.

Images & Videos

We have a strong preference to use novel images provided directly by the client, i.e. photographs of their products, staff, happy customers, work samples, offices, etc. These are the most unique and contextually accurate photos and will usually perform better than anything else. It also makes aligning creative with landing pages super easy.

When not possible the next best option is usually paid stock images. Of course this isn’t as on point as client provided photos, but they can often approximate the context fairly closely.

Next up will be AI generated images. We have experimented a bunch and continue to do so with Google’s built-in image generator. There is nothing wrong with the quality of the images but there are two major limitations. This generator will not generate humans. And as it often makes sense to include a human subject this is a huge gap. Secondly, you cannot include brands (even your own) in prompts. Thus if you sell, say Jeep accessories, you can only generate a generic SUV which isn’t a Jeep and won’t look much like a Jeep.

Thus, when we need to generate AI images we’ll usually turn to Leonardo.ai, (used for the image in this article) an extremely powerful and flexible tool for creating virtually anything we want. Despite how advanced this and other generators are, there are still often flaws… e.g. human hands are still a huge problem so you really need to look carefully at an image before slapping it on a campaign.

There are some other issues with using AI. Platforms are starting to require us to declare when we use AI generated images. This then gets added as a disclaimer to the ads. There is some negative sentiment towards brands using AI-generated images, often because it’s perceived as taking away jobs from photographers, designers, and models. It’s not untrue of course, so we generally only use AI when we need to.

The last option, if you don’t have access to any others, is to use free stock images. While there are tons of great photos available, it’s hard to find anything with the specific context required. Furthermore, these images are used so often that is can make an advertiser look amateurish or cheap.

As for videos, there aren’t many good AI options right now. Sure you can acquire a generic loops or make something AI that looks the part… but we’re not quite there yet. We generally cross our fingers that we don’t have to use Google generated videos. While I’m not a fan of explainer videos, those can be a lot better than anything Google Ads cranks out automatically or even using their built-in video creation tools. This area is still quite a challenge for small budget PPC advertisers. Perhaps AI will truly come to the rescue in 2025.

Campaign Insights & Recommendations Engine

All platforms have launched different versions of various Insights tools over the years. The current state of the art from Google is their Insights tool which can be used at the account or campaign level. It’s far superior to older generations but still contains a lot of bloat, i.e. useless information. In the meantime, their Recommendations Engine only gets about 10% right. Most recommendations just don’t make any sense at all.

Microsoft has a much lighter offering with far less information and more recently launched their new “Summarize” tool. By default the tool generates a summary of campaign performance for the dates and KPIs currently displayed. It does exactly what it says it does, here’s an example.

Now it would be really nice if the tool offered comparative data. For example, when I’m looking at January compared to December it should summarize the change in performance. It can’t do that even if you ask specifically. Microsoft’s Recommendation Engine, in the meantime, is also wrong most of the time. Interestingly, it seems to generate many more keyword recommendations than Google does. But it frequently recycles the same bonehead ideas like constantly asking for us to import campaigns from Meta Ads.

Meta Ads offers recommendations directly in campaign, ad group, and ad views. These mostly revolve around applying creative ideas as well as turning on various Advantage+ features such as for audience targeting. And as with other platforms, the creative is hit or miss and usually full of errors.

Importantly, all of the platforms offer “auto-apply” settings to actively implement any/all recommendations they come up with. This is super dangerous and something we’ll never activate at my agency. All recommendations get a look during our regular optimization routine. And these tools are helpful. But it’s probably more important at this point to know which ones to dismiss more than which ones to apply.

What’s on the Horizon?

Here are some things my agency will be adopting AI for PPC in the near future:

We’ve been working on an integration with our reporting tool that will automatically generate a summary of performance along with a comparison to the previous period’s performance. This is something we have to write-up manually in every report and it’s not very valuable work since we’re effectively regurgitating numbers. But so far getting something to work easily and consistently has proven elusive. We’re hopeful that our reporting tool provider will bake this feature in at some point. If not we’ll keep experimenting with different options.

We’ve developed our own script that allows us to quickly check performance each morning for all of our accounts. This has been in place for quite a few years and works like a dream. As a next step we plan to put AI in place that will detect and highlight performance discrepancies. This will not replace manual review rather it will be used a second set of eyes.

We’ve adopted Mike Rhodes’ P-Max script and use it extensively to optimize campaigns. This tool offers AI analysis, which I’d say in in “Alpha” at this point. We haven’t started testing that component yet but we do plan to evaluate it this year.

We might see Google and other ad platforms significantly improve automated creatives this year. If so, we will, selectively, start experimenting more with these automations. However, we are not going to allow any system to run creatives without full transparency, i.e. the ability to view active presets and/or ad copy history.

Summary

AI, indeed, will revolutionize the PPC industry and has already made inroads. But we’re very far away from pushing the “Easy” button to create and optimize PPC campaigns. If anything, we’re currently at a stage where PPC management is more complicated and convoluted as we wrestle with black box solutions and how to best manage those.

For agencies and advertisers to be successful they’ll need to carefully navigate what’s on offer from “AI” and know when to pull the trigger. More specifically, AI bidding, insights, and user-moderated content are no-brainers. But we’re not there yet with automatically applied recommendations or letting campaigns run completely on auto-pilot.

At TTFV, we’ll continue to adopt a “passive” AI approach wherein we’ll leverage the tools but remain gatekeepers for now.