Using AI in paid media with human-centric guardrails

Artificial intelligence (AI) is something you're probably already familiar with. After all, Google has been using it behind the scenes for paid search optimization and many other purposes for years.

At Imarc, we routinely lean on Google's AI-powered pay-per-click (PPC) and automation features as well as some emerging technology like ChatGPT, but we do so with oversight.

AI PPC advertising tools and other automation options are helpful and very smart, but they still need to be kept in check. The more you use them, the less control you have, so it's still important to build campaigns from scratch and monitor them.

Let's go over how we use AI in paid media to help improve campaign performance as well as things we do to keep campaigns from going off the rails.

Google responsive display ads (RDA)

When our Creative team makes banners for display ads, they create them in common sizes, but some websites require sizes that aren't standard. That's where Google comes in.

As it's just not feasible to have Creative custom-make thousands of banner sizes, RDAs allow us to upload an asset library of copy and images, and then Google combines them in the less common sizes to earn more impressions overall.

That sounds pretty wonderful, so you might wonder why we don't let Google run all the ads using variable copy and images. Well, the reason is that we want to control the majority of combinations.

When using RDAs we:

• Pre-write copy carefully – We consider the headline and description combinations in advance.
• Don't allow landing page scrapes – When Google is given full control to seek out its own content on a landing page, unchecked AI insertions can occur.

Google dynamic keyword insertion (DKI)

Google uses a metric called quality score that evaluates the relevance of your keyword, the ad you submitted to Google, and the content on your landing page.

DKI can be used to insert the query a user puts in the search bar directly into your ad copy or headline. By using the exact keywords a user is searching, it increases your ad relevance which eventually leads to increased click-through rates & improved quality scores.

Additionally, the DKI feature can be useful to scale for larger campaigns as it can be cumbersome to write ad copy for all possible variants.

For example, an e-commerce campaign designed to sell a t-shirt could require one ad with DKI as opposed to manually writing ads for all of your t-shirt's color variants like white, black, blue, and pink, in addition to sizes like small, medium, and large.

When using this option we:

• Reserve use – While we've seen success from using DKI, it's not right for every campaign or tactic. For example, we usually don't use it for competitor campaigns as it frequently generates poor performance & brand confusion.
  
• Perform negative keyword exercises – Performing regular negative keyword exercises allows us to solve for confusion in the ads & exclude search queries we don't want showing in our headlines/ad copy.
  

Google dynamic search ads (DSA)

Many choose DSAs because it means that they don't have to write ad copy. The basic premise is that you upload a landing page and Google creates your headlines and descriptions and automates a search ad for you.

After launch, Google has the flexibility to mix and match the content until it performs well. The downfall is that it's not controlled, on brand, and often doesn't echo your brand voice.

Overall, we don't use this much, but if we do we:

• Make a separate campaign – Making this type of campaign independent will allow you to closely monitor the results.
  
• Exclude brand terms – If you don't pull out your branded terms, the campaign can run amok and you run the risk of duplicating your other campaign efforts.

Google performance max (PMAX)

While PMAX campaigns aren't totally new, Google has started to push them more in the past year. We use it for some clients because it takes DSAs, to another level of automation.

Most commonly we use five headlines, one long headline, five descriptions, one business name, one square logo, one landscape logo, five horizontal images, five square images (20 images are possible), one Youtube video, and one CTA to create the ideal mix.

PMAX also allows us to extend our campaign placements to platforms like YouTube, Gmail, and Google Discover which we usually don't bid on directly.

When we use PMAX, we:

• Divide our dollars – We typically monitor PMAX budgets separately from our normal budgets so that we can control how many dollars are spent.
  

ChatGPT

While we are confident that ChatGPT will never fully replace our content experts, we do let it play a role in paid media.

For example, during your keyword research, if you plug in "hotel in Chicago" it will show 50 other keywords that are similar to that result without needing to enter a keyword planning tool (ie SEMRush or AHRefs).

It can also be used to deliver more persona-driven ad content. It's helpful to ask it to "take this ad and rewrite it for an employer audience with x tone" or to help us "write an ad that's 90 characters from this landing page content but keep the sentiment." Those questions help us speed up the inspiration process and allow us to modify and optimize as opposed to simply generating ideas from scratch.

Finally, ChatGPT is a perfect solution for crafting calls-to-action (CTA). "Learn more," "Book now," etc. are very generic, so it can easily be used to help create new ones.

When we use ChatGPT, we always:

• Manually vet keywords – We always go over AI-generated keywords and pull out what fits best for our clients. We do by using tools like Semrush, AHRefs, and Google Keyword Planner.
  
• Massage the copy – Nothing that comes right out of ChatGPT goes into a client or Imarc brand campaign without modification.

-

Automated paid media is an exciting arena to play in. If you'd like some help from a partner with experience in the trenches, let's talk.