We’re Daypart AI. You’re an intrepid internet and AI power player. This is your new favorite newsletter.

Any questions? Save ‘em for Claude and let’s get to it.

When it comes to machine learning driven personalization of digital ads, few companies can claim more expertise, or to have had a more significant impact on the space, than Google.

That being said, there is inevitably always the bridge that is too far, even for the most intrepid of pioneers.

Looking at the state of image and video generation in Google Ads today, as a product and as a service with terms every user accepts in exchange for access to those sweet, sweet click-based conversions, we might very well be on the other side of said bridge.

Google’s product and terms have changed and there may be reasons for concern over their AI’s impact on your ads.

To be ready requires understanding- you must know thy frenemy.

Google Ads is one of the most enduring and widely used large-scale digital ads programs, used by everyone from the world’s largest enterprises to the S-est of SMBs.

As a result, nothing has a Culture of Running Ads like the product still most affectionately known to many of its longest-running users as “AdWords.”

One of the core pillars of this culture, likely since the moment anyone ever placed a single iota of trust in Google, has been “never trust Google.”

This discourse lives within the now very familiar cycle we call “enshittifcation,” a term that is actually much younger than this dance of trust-but-verify-and-complain-anyway that’s been done by Google Ads practitioners for a couple decades.

Google Ads attracted advertisers in large part because in their eyes, it just worked, offering measurable, performance-based results that justified continued and growing spend.

Once this customer base had been secured, placed on the table before Google were a slew of ways to optimize both their revenue, and their customer experience.

Any given choice moves both sliders up or down, and any wonderful boon could SECRETLY be a sickeningly sweet poison pill hiding only revenue gains for Google and no real benefits for the customer.

A trusting Google Ads manager might look at platform features and have a thought like: Google will automatically adjust my daily spend based on high value click volume? Wow, sounds like a great way to no longer need to micromanage my campaign spend!

“WRONG! SOUNDS LIKE A GREAT WAY TO LET GOOGLE DECIDE HOW MUCH OF YOUR MONEY IT GETS! SPOILER ALERT: IT WILL GET EVERY PENNY!”

So screams the chorus of midmarket performance agency managers and basement bid jockeys generating warm leads for cosmetic dentists every time there’s a new feature release.

How right or wrong these people are seldom matters in the end, because Google has sole control over how you buy their search ads. If you don’t like a new feature, that may be too bad, because yesterday’s closed beta and today’s prominently featured part of the UI become tomorrow’s mandatory participation, or if not a mandate, an option with an opt out path that is difficult for many users to locate.

For this reason, one must be pragmatic about adopting new Google Ads features. Once the league has a three point line and is calling fouls out there, it’s better to draft Steph Curry than Ben Simmons, even if you’re going to get razzed by your peers for picking a little guy who played at Davidson.

To understand why we may be entering a period of rare exception, in which brake slamming could be prudent, you need to understand the two places Google’s ML really made its bones and how what’s coming is not really anything like them.

The first wave of Google Ads automation was squarely focused on helping advertisers manage their spend.

This era was mostly about introducing portfolio systems that allowed advertisers to achieve maximum marginal gains across a wide array of line items, while allowing them to set a top level spend cap for an entire initiative.

Elements of bid optimization did give Google its angle to start wresting control away from third parties, which back them meant the search management platforms that were the first to bring automated bidding to bear.

Initially, bid automation was almost entirely about keywords, and the best one could do was to rapidly respond to real performance changes.

Later, third party search management platforms would become a bit more advanced, inferring relationships between keywords that allowed a system to intelligently change the bid on lower volume terms based on a larger data set from high volume terms that had a mathematical relationship.

The move away from external parties managing automated optimization started when in 2013, Google popularized the concept of bid modifiers based on things like time of day, geography, demographics, and other signals, paired with enabling data points from their suite of products with a billion-plus users.

This introduced the notion that a user’s search query for “size 12 red shoes” was only half of the story that told their value to an advertiser-the other half was where they lived, how old they were, their gender, and other basic demographics.

What this paved the way for was that there was an even more tantalizing tale being woven behind the search query in their browsing and total search history, plus a whole constellation of other signals that came from Google’s many massive consumer internet services and their ad network with tags on a vast swath of the open web.

This is where Google also got a lot of early practice with their favorite bludgeoning stick to use against advertisers and ad tech: user privacy!

That treasure trove of user data is simply too sensitive to trust to advertisers was a position staked out early on and vigorously defended at every turn.

Well then, if that’s the case, it looks like the Google’s own ads bidder is clearly the buying tool positioned to deliver the best results!

Crazy how that happened Sorry, they just really care about user privacy!

A major reason all of this went over well with advertisers so quickly as it did because it was about efficiency.

Reshaping a user acquisition curve certainly theoretically means a little bit more scale, but this was mostly about getting CPA down for a given budget level. That’s almost always a very easy pill for advertisers to swallow, as it delivers value without requiring them to invest more.

What Google Ads mostly sold for the first half of its existence were text ads, which is the best place to start dynamically optimizing and generating creative. Sam Altman et al seem to have agreed, we call them Large Language Models, not Big Picture Models.

ML driven text creative also offered advertisers interesting new tools with which to work efficiently and improve campaign performance.

In the early days what these systems touted was the ability to automatically show different ad permutations to audiences with different preferences, plus modals to properly A/B test ad elements within this framework.

However, a means for Google to yet again wrest control away from third parties quickly became clear.

What if the big lever isn’t showing slightly differently worded ads about red shoes to everyone searching for “red shoes,” but showing value based messaging to someone inferred to be in a lower income bracket and quality based messaging to someone inferred to be in a higher income bracket?

The data needed for this optimization is sensitive, and requires, you guessed it-privacy! Only Google can handle this precious data, and anyone who thinks otherwise must really not respect consumers and their pristine data security!

Thus, we entered the “put many disparate types of messaging elements into your Google Ads and let us deal the cards” era, which has essentially been the platform-mandated approach for several years, under the banner of Expanded Text Ads.

Again, this mostly ended up being about efficiency and operational drag reduction for most advertisers-this was not a massive “scale unlock” in terms of total ad budget mix in most cases.

Lurking in the background through all of this was a problem that weighed heavily on Google: they had one inventory pool people cared about.

The vast display ecosystem was well tapped into by the beginning of the 2010’s, with text ads, and later simple image ads, but scaled for very few advertisers relative to search, and what was then known as GDN was largely scoffed at by the display advertising industry.

YouTube rapidly became on of the world’s largest attention sinks and the capacity to buy it sat right there in Adwords for a long time, but in terms of the sheer number of advertisers engaging with it, it was still small relative to search ads.

Google did have one major inventory source expansion success in Google Shopping, and a key distinction this product had versus previous search engine results pages for products is that it presented, and came to heavily rely on, images of the products in question.

In this Google Shopping win, a good but probably troublesome idea germinated out of a real truth: a significant number of smaller advertisers lacked high quality images of their entire product lines in the early days of e-commerce.

After years of programs to help advertisers improve their product images and get better results, it became clear to Google that this lack of available product images, and variety of lifestyle images to go with them, had been a significant factor discouraging advertisers from more participation (and spend) in Google Shopping ads.

When generative AI image and video generation exploded onto the scene in the early 2020’s, they had the same idea that people at Meta and in independent ad tech had: is this not the path to provide millions of small advertisers with the one thing that’s holding them back from spending vast sums of money on not just Shopping ads, but the whole display and video ecosystems?

There it is, folks: a Google Ads paradigm change towards less putzing around in the realm of efficiency, and more towards unlocking significant added revenue scale!

Without even attacking the premise that search and shopping advertisers can successfully invest heavily in display and video for the same kind of results they’ve come to expect, there’s a big problem here.

When you see a video of a man who vaguely looks like a doctor selling you mushroom coffee and he has three arms, you don’t return the product and post angrily to your IG followers about it because you didn’t sprout any extra appendages after drinking it.

When you buy an expensive piece of lingerie where the bra appears to have a delicate front clasp, and it’s a multi-hook climbing harness grade affair with all the bolts on the back, you absolutely return it and probably never buy from that company again, at least online.

This is the crux of a major issue many e-commerce advertisers are having with generative AI: it has documented accuracy limitations when rendering advertiser products, including image hallucinations and incorrect text generation, that means advertisers need to manually review 100% of its output before publishing any of it.

Customers won’t trust a merchant that misrepresent what they’re selling, and merchants won’t trust something that puts them in a position where they might violate customer trust simply by not catching erroneous output from a tool that’s supposed to make their business better.

Further afield from that, there are ad quality concerns for any advertiser that cares about more than one click-and-sale today.

Brands are real and branding is important. If you’re a luxury automaker and you’ve suddenly got stuff on YouTube that looks like a collaboration between your realtor’s Canva account and an Amazon knockoff brand called WONDEEFOOO, that’s a huge issue.

Of course, all of this hot nasty techno-rash is fine so long as we have the soothing balm of advertiser control to apply liberally to our body of work.

BY GOD, IS THAT-IS THAT THE GOOGLE TERMS OF SERVICE TEAM’S MUSIC?

There it is, straight from Google’s own AI overview, which has already NAILED CONTEXT given how it’s taking the tone of a told-you-so absolutist delivering you this news GLEEFULLY.

Remember when Bard stunk and lots of y’all thought Alphabet was cooked? Lol. Lmao.

When you set something up, Google defaults to enabling its generative AI “enhancements.”

If you don’t stop it, that’s your fault, and if you get in trouble, Google told you that could happen.

The short answer is “roll back to only the automation that has served us for years, and not entirely off of Google Ads.”

The extent to which this is possible is in the hands of Google themselves, of course.

That being said, it’s not an seen as an optional channel by millions of advertisers, and so it will be imperative to lots of people to solve this problem.

The best thing you can do is arm yourself with the knowledge and budget for diligence that will keep your ROAS high and your sweaty areas pleasantly dry-or put someone on your team who can.

You need to understand your Google Ads creative components at the most granular level, understanding that everyone one of them is essentially an atomic claim, and ensure your marketing is legally ironclad accordingly.

Automated, large scale third party review is also going to come back in vogue, even though we’re all going to roll our eyes and rightfully decree “AI solutions to AI problems” is what’s afoot here.

Either way, the most diligent organizations with the best handle on what their ads look like and what said ads purportedly do for them are going to come out on top-just as they have been up to this point.