What exactly can Google’s Safe Search teach us about becoming better digital marketers?

It’s proven to be very helpful because it perfectly illustrates how to use algorithms to automate search, and the same principles Google uses can help PPC professionals improve their game.

When Google automates almost every aspect of search marketing, you don’t have to sit on the sidelines and hope for the best, you can deploy “automation tiers” to stay in control.

I believe the concept of automation layering is critical to the success of PPC in an increasingly automated world.

Secure Search: Algorithms + Humans

I’d like to begin with an anecdote about my involvement with Safe Search, back in the early 21st century, shortly after I joined Google.

If you’ve heard me speak at a PPC conference, you already know why Google hired me: they needed a native Dutch speaker to work on AdWords.

Since I grew up in Belgium, speak Flemish (the Belgian version of Dutch) and I live in Los Altos, just a few miles from Google’s new campus in Palo Alto, I was interviewed for the position.

Since the other native Dutch speaker they interviewed insisted on wanting a private office and an administrator, they quickly decided that I might be a better fit.

I started working at Google, first translating AdWords into Dutch, then reviewing all Dutch-language ads and providing email support for customers in Belgium and the Netherlands.

One day, while I was reviewing the Dutch ads in the “spam box,” a search engineer came into my cubicle with a stack of printouts and put them on my desk. Upon reading the list, I was shocked by some of the words on it.

The engineer told me that it was a sorted list of pornographic searches. Their algorithm had sorted these searches from highest to lowest level of pornography, now they just needed to figure out where to set the threshold for safe searches.

They wanted me to read the list carefully and draw a line to indicate that the search query ranged from definitely looking for pornography to something that could be construed as pornography but could also include other content.

I drew a line with a red marker, and that’s how safe search thresholds are set in the Netherlands: machine algorithms and Fred with his red marker.

I learned from this experience that algorithms are good at calculating similarities between things.

By assigning a numerical value to each query, they can rank them on different dimensions, such as how pornographic they are.

But manual review of the ranked list is required to determine a reasonable threshold that meets the business criteria for a secure search function.

Thresholds in PPC

Then I realized that Google uses thresholds in hundreds of places.

When an ad has a high enough quality score, there is a threshold that qualifies it to appear above the natural search results.
There’s a threshold for when an ad experiment ends.
A threshold exists for when queries are similar enough to keywords to show ads.
And so on.
If Google uses so many thresholds, we advertisers should also think about how to recreate these systems so we can set our own.

Google uses thresholds to determine if the meaning of a query is close enough to be considered a “close variant”. Advertisers can use their own automation to set their own lower thresholds to better control how long their ads are displayed.

Automated hierarchical control of close variant keywords

The same algorithmic ranking happens every time you bid on an ad on Google. It happens in a variety of forms, but one of the easier to understand forms is keyword matching.

For example, consider Google’s recent expansion of “close variants,” where exact match keywords no longer match exactly, but can also include searches with the same meaning or intent.

When a user searches for a keyword that is not identical to an advertiser’s keyword, Google’s algorithm calculates a score that indicates the likelihood that the search result “means the same thing,” or how close the search result is to the keyword.

The machine returns a score, and if that score is above a certain threshold, the ad is eligible to be shown.

Overall, these automated features mean that advertisers can do more with less. Instead of figuring out all the possible variations of each keyword, they can let Google’s state-of-the-art machine learning system solve the problem.

I like this automation, perhaps because when we looked at our own data to find similar variants, the vast majority were misspellings of our company name, Optmyzr.

I think we’re probably worse than Britney Spears in terms of the number of times we misspell our name (if you’re curious, Britney’s name is misspelled about 500 times).

But people are always skeptical about using these kinds of black box systems because it’s impossible to predict what they’ll do in every possible situation. This worries advertisers who are used to having tight control over search ad targeting.

Automated Layering Sets Your Own Thresholds

This is where automated layering comes into play.

There’s a false assumption that when an advertiser enables Google’s automation features, only Google has control. In this case, the automation feature is a close variant version of a keyword, with no off switch.

But the reality is different: advertisers have some degree of control.

For example, they can add negative keywords when they find that Google is showing ads related to close variants they don’t like.

The problem is that automating the process of manual monitoring is tedious and time-consuming. It may not be worth it to get better results.

But what if advertisers could automate their own processes and combine them with Google’s automation capabilities?

This is automation layering.

In a close variation of the example, here’s one way to use automation layering. You can use a rules engine (Optmyzr provides one) or a script to identify similar variant search terms and apply an algorithm to rank those search terms.

In another post, I explained that advertisers can use the Levenshtein Distance Score to calculate the difference between a search term and a matching keyword.

An edit distance of 2 or 3 is usually a spelling error. Any edit distance longer than that is usually more than just a spelling error and may be worth double checking:

Or as a new keyword idea.
Or you can exclude something by adding a negative keyword.
Now that advertisers have their own numerical score, they can draw their own line and set a threshold they think makes sense for their business.

Google uses its thresholds to show ads more frequently, and advertisers use their thresholds to show ads less frequently.

When the two automated thresholds are combined, advertisers can benefit from showing ads without keywords. They can also benefit from tighter control over the relevance between queries and keywords.

Conclusion.

The purpose of my work on safe search and the example on proximity variants was not only to explain that advertisers can control proximity variants, but also to illustrate some ideas that are critical to the future success of PPC.

Most advertisers agree that humans + machines are better than machines alone, but humans don’t always have the bandwidth to manually monitor all the new automated features that engines keep introducing.

So it makes sense that humans might want to find ways to create their own automation systems to override the engine’s automation systems.

Ultimately, we just want to have a say in how aggressive the level of automation should be, and that’s where the threshold comes in.

Only by layering our own automation can we control Google’s undisclosed thresholds.

And since most advertisers are not tech savvy enough to code their own automation, I firmly believe that pre-built scripts, rules engines, custom monitoring and alerting systems, etc. will be key tools for successful PPC managers in 2024 and beyond.