Unfortunately, Google has attempted to make them ubiquitous.

Google Ads has been rapidly expanding their use of auto-applied recommendations recently, to the point where it briefly became my least favorite thing until I turned almost all auto-apply recommendations off for all the Google Ads accounts which we manage.

Google Ads has a long history of thinking it’s smarter than you and failing. Left unchecked, its “optimization” strategies have the potential to drain your advertising budgets and destroy your advertising ROI. Many users of Google Ads’ product ads should be familiar with this. Product ads don’t allow you to set targeting, and instead Google chooses the targeting based on the content on the product page. That, by itself, is fine. The problem is when Google tries to maximize its ROI and looks to expand the targeting contextually. To give a practical example of this, we were managing an account advertising rotary evaporators. Rotary evaporators are very commonly used in the cannabis industry, so sometimes people would search for rotary evaporator related terms along with cannabis terms. Google “learned” that cannabis-related terms were relevant to rotary evaporators: a downward spiral which eventually led to Google showing this account’s product ads for searches such as “expensive bongs.” Most people looking for expensive bongs probably saw a rotary evaporator, didn’t know what it was but did see it was expensive, and clicked on it out of curiosity. Google took that cue as rotary evaporators being relevant for searches for “expensive bongs” and then continued to expand outwards from there. The end result was us having to continuously play negative keyword whack-a-mole to try to exclude all the increasingly irrelevant terms that Google thought were relevant to rotary evaporators because the ads were still getting clicks. Over time, this devolved into Google expanding the rotary evaporator product ads to searches for – and this is not a joke – “crack pipes”.

The moral of that story, which is not about auto-applied recommendations, is that Google does not understand complex products and services such as those in the life sciences. It likewise does not understand the complexities and nuances of individual life science businesses. It paints in broad strokes, because broad strokes are easier to code, the managers don’t care because their changes make Google money, and considering Google has something of a monopoly it has very little incentive to improve its services because almost no one is going to pull their advertising dollars from the company which has about 90% of search volume excluding China. Having had some time to see the changes which Google’s auto-apply recommendations make, you can see the implicit assumptions which got built in. Google either thinks you are selling something like pizza or legal services and largely have no clue what you’re doing, or that you have a highly developed marketing program with holistic, integrated analytics.

As an example of the damage that Google’s auto-applied recommendations can do, take a CRO we are working with. Like many CROs, they offer services across a number of different indications. They have different ad groups for different indications. After Google had auto-applied some recommendations, some of which were bidding-related, we ended up with ad groups which had over 100x difference in cost per click. In an ad group with highly specific and targeted keywords, there is no reasonable argument for how Google could possibly optimize in a way which, in the process of optimizing for conversions, it decided one ad group should have a CPC more than 100x that of another. The optimizations did not lead to more conversions, either.

Google’s “AI” ad account optimizer further decided to optimize a display ad campaign for the same client by changing bidding from manual CPC to optimizing for conversions. The campaign went from getting about 1800 clicks / week at a cost of about $30, to getting 96 clicks per week at a cost of $46. CPC went from $0.02 to $0.48! No wonder they wanted to change the bidding; they showed the ads 70x less (CTR was not materially different before / after Google’s auto-applied recommendations) and charged 24x more. Note that the targeting did not change. What Google was optimizing for was their own revenue per impression! It’s the same thing they’re doing when they decide to show rotary evaporator product ads on searches for crack pipes.

Furthermore, Google’s optimizations to the ads themselves amount to horribly generic guesswork. A common optimization is to simply include the name of the ad group or terms from pieces of the destination URL in ad copy. GPT-3 would be horrified at the illiteracy of Google Ads’ optimization “AI”.

A Select Few Auto-Apply Recommendations Are Worth Leaving On

Google has a total of 23 recommendation types. Of those, I always leave on:

• Use optimized ad rotation. There is very little opportunity for this to cause harm, and it addresses a point difficult to determine on your own: what ads will work best at what time. Just let Google figure this out. There isn’t any potential for misaligned incentives here.
• Expand your reach with Google search partners. I always have this on anyway. It’s just more traffic. Unless you’re particularly concerned about the quality of traffic from sites which aren’t google.com, there’s no reason to turn this off.
• Upgrade your conversion tracking. This allows for more nuanced conversion attribution, and is generally a good idea.

A whole 3/24. Some others are situationally useful, however:

• Add responsive search ads can be useful if you’re having problems with quality score and your ad relevance is stated as being “below average”. This will, generally, allow Google to generate new ad copy that it thinks is relevant. Be warned, Google is very bad at generating ad copy. It will frequently keyword spam without regard to context, but at least you’ll see what it wants to you to do to generate more “relevant” ads. Note that I suggest this over “improve your responsive search ads” such that Google doesn’t destroy the existing ad copy which you may have spent time and effort creating.
• Remove redundant keywords / remove non-serving keywords. Google says that these options will make your account easier to manage, and that is generally true. I usually have these off because if I have a redundant keyword it is usually for a good reason and non-serving keywords may become serving keywords occasionally if volume improves for a period of time, but if your goal is simplicity over deeper data and capturing every possible impression, then leave these on.

That’s all. I would recommend leaving the other 18 off at all times. Unless you are truly desperate and at a complete loss for ways to grow your traffic, you should never allow Google to expand your targeting. That lesson has been repeatedly learned with Product Ads over the past decade plus. Furthermore, do not let Google change your bidding. Your bidding methodology is likely a very intentional decision based on the nature of your sales cycle and your marketing and analytics infrastructure. This is not a situation where best practices are broadly applicable, but best practices are exactly what Google will try to enforce.

If you really don’t want to be bothered at all, just turn them all off. You won’t be missing much, and you’re probably saving yourself some headaches down the line. From our experience thus far, it seems that the ability of Google Ads’ optimization AI to help optimize Google Ads campaigns for life sciences companies is far lesser than its ability to create mayhem.