On April 16, 2024, Google announced the release of a new generative AI feature for Demand Gen campaigns in Google Ads.
The feature allows advertisers to generate original AI images using text prompts, similar to DALL-E and Midjourney but more specialized for ad images.
According to Google, this feature is part of a broader effort to improve ad relevancy and engagement across its platforms, including YouTube, Discover Ads, and Gmail.
So what does this new feature mean, what are Google’s generative AI images strengths and weaknesses, and how can advertisers take advantage.
All this and more, read on!
How Google Ads Generative AI Image Feature Works
The feature is now broadly available to English-speaking advertisers and will be rolled out in more languages soon.
- Find the feature by going into the Ads section of an Ad group within any existing or new Demand Gen Campaign.
- Click on “Generate Images” in the Media section
- Input prompts to create different images
(Optional) Click the ellipses of generated images and select “Generate more like this” to build more variations
Psst… want more? Check out all the technical specifics and resource links in the Google Ads feature help page here.
Strengths: What the AI Image Tool Does Well
Having the ability to create new ad creative on the fly within the Google Ads interface is a nice perk. Here’s what I like about it:
- Fast – compared to other AI image generator tools, Google’s spits out images in a fraction of the time, usually just a few seconds.
- Perfectly Sized – images are sized exactly for any or all of Google’s placements across YouTube, Discovery, and Gmail.
- Convenient – the tool is right in the interface, saving the need to download, reformat, and upload external image files.
- Safe – images generated are intentionally generic, they won’t offend anyone and won’t get you into any copyright issues. You also won’t run into some of the other AI image challenges such as demon-faced people, provocative poses, or just strange and cringe output.
- Rapidly Iterate – this is a sub-strength of everything above. Because I can create ad images so quickly, I can make ad-group specific creative, most test variations, and brainstorm more ad concepts than I could without it.
Read on (or jump down) to see some of the ways I’m putting this feature to use.
Weaknesses: Where the AI Image Tool Lacks
Users already using ChatGPT-4’s DALL-E or MidJourney for AI image creation, will likely find Google’s capabilities and output lacking. It feels a year behind in terms of the image quality and the ability for the user to detail their prompt so that they get exactly what they want.
In my early testing, the biggest challenge is how restrictive the tool is in what it will not generate. About 2 out of 3 prompts I cooked up wouldn’t provide output, where other AI image tools would have spit them right out. Google is likely being overly cautious following the AI image backlash received from it’s Gemini image generator.
So what’s limited? Here is exactly what Google’s product guide states:
Content restrictions
Generative AI tools in Google Ads are designed to automatically limit the creation of certain content. This includes:
-
Faces, children, or specific individuals
-
Branded items and logos
-
Responses to matters of opinion or advice
-
Images that may violate Ads Policy or our Generative AI use policy
Not being able to have people’s faces is pretty prohibitive. Furthermore, we found that 60-70% of the prompts we use in other AI image tools, get declined or have “no results” in Google’s. We really have to use more generic terms in our prompts. As an example “a picture of trees in the style of Ansel Adams” would definitely get rejected.
Compare Google Ads Images to MidJourney images
As an example, I mocked up some images to use for a plumbing company ad.
Midjourney
The Midjourney output is much more detailed, eye-catching, and brand-elevating in my opinion. I can also customize Midjourney’s output much more, where Google’s “more like this” feature doesn’t provide a lot of variation…
It’s all hands holding wrenches, where Midjourney provides more variety and creative directions.
Here are some other Google outputs to give you a better idea of the limited style you can expect:
In most cases, Google Ads’ free stock photos library, which is also directly in the interface, provided better imagery. Consider these options compared to the massage business image above — with the bonus thumb 🙂
No Brand Building or Personality
If you’re trying to build brand recognition, highlight unique differentiation in your brand or service, or have a personality, then Google’s AI images will make it difficult.
If your product or service is fairly undifferentiated then that’s not a problem.
How to Use Google Ads’ New Generative AI Image Generator Effectively
Just like photo images, Google AI images are generic by nature. They’re meant to be broadly used and not specific to your business, and sometimes, that’s okay.
I get the most out of this feature when I focus on a few use cases where it really shines.
When Branding Isn’t Necessary
If I don’t have a brand style and image to maintain, then I feel totally fine using more generic AI or stock images , then you shouldn’t be using images. If I were running Google Ads for Nike, Apple, or Starbucks I wouldn’t touch the stuff.
If you’re a local pet sitter, auto repair shop, lead aggregation website, or small real estate agent. You can do a lot with this feature.
Rapid Iterations
Since I can create images rapidly and easily, I’m more inclined to create images that closely align with different ad groups.
Here are some examples.
A plumbing company could create images for:
- Kitchen remodels
- Hot water heater replacement
- Bathroom remodels
A real estate company could serve different images for:
- Condos
- Luxury properties
- Starter homes
An insurance company could build ad images for:
- Home insurance
- Boat insurance
- Auto insurance
Instead of just reusing the same image, we can dial up the relevance with unique images.
Sure, I could also grab these from the stock photos tab, but the AI will size them perfectly, they’ll be original images, and I can set the image style to match my brand, such as color tones, photo style, etc. I can also create more unique variations for more relevance. For example, If I’m a realtor, I can create “Atlanta, GA luxury home exterior with lush green lawn, iso 200, professional photography” images, not easily found in stock photos.
Modify Your Prompts
I’m still figuring out the prompt best practices and always encounter restrictions when prompting, the same way I have for other AI image tools.
Here is what I’ve figured out so far:
- Use shorter, less specific prompts. Google is for quick and dirty images – keep that in mind.
- Avoid anything related to people, brands, or artists. Even a prompt like “real estate agent” or “office worker” will get rejected.
- News Story Style Images. Shoot to generate images that represent the service or ideal customer outcome.
Think Native Advertising
Native ads don’t look like blatant ads, they look more like news stories. If you lean into that, then AI images can work. There is no brand logo, there is no button, and there is no text — sounds like lot more like a news story feature image than an ad, right?
So, to play up the native ad angle, you need to plan your headline and description text accordingly.
Typical Ad Example: “Boost Your Business with Top-Tier SEO Services!”
Native Ad Style Example: “Tech Startup Sees an +87% Leads Using this AI-Driven SEO Model”
The native ad example leans more into an editorial style, blending seamlessly into the surrounding content, which makes it feel less like a conventional ad and more like a valuable piece of content. This approach is less intrusive and can lead to higher engagement rates from readers who are interested in the subject matter rather than feeling like they’re being directly sold to.
