Machine learning and the loss of control

Google has taken away control from marketers over the last 3 years and replaced it with machine learning automation.

Machine learning and the loss of control

The boiling frog phenomenon describes the concept of a frog jumping out if thrown in a boiling pot of water but sitting still if the water temperature gradually increases from lukewarm to boiling.

The story has been refuted. It’s not true, but the metaphor is still very much helpful.

In the last 5 years, Google started to use more machine learning in different areas of its business. In the last 3 years, Growth experts lost more control as Google automates processes in organic and paid marketing.

Google takes more control of page titles

The most recent use case is Google’s automated title changes in organic search. In late August, Google rolled out a change that customizes titles even more than before.

In 2013, Google started to change page titles based on the query searchers used. The new model uses anchor text from links, headings, or onpage text to pick a “better” title that is supposed to work better for documents overall.

Last week, we introduced a new system of generating titles for web pages. Before this, titles might change based on the query issued. This generally will no longer happen with our new system. This is because we think our new system is producing titles that work better for documents overall, to describe what they are about, regardless of the particular query.

(source: https://developers.google.com/search/blog/2021/08/update-to-generating-page-titles)

Even further, Google might pick a different title for desktop vs mobile results, making it much harder for SEOs to know which title variant appeared when and where.

When I asked whether Google will determine the relevance of a page based on the rewritten title, John Mueller clarified that that’s not the case. That would’ve meant even more lost control. At the same time, we can no longer judge the title optimization based on what we see in the SERPs.

It’s also still fuzzy how big the impact is. We know roughly how many results are impacted but not how severe the impact is. SEOclarity shared that 23% of titles changed (source: https://www.seoclarity.net/blog/page-title-rewrite-analysis), Moz says 58% (source: https://moz.com/blog/ways-google-rewrites-title-tags). But the question is how many titles were completely rewritten vs truncated or expanded?

Google takes more control of keyword matching for ads

In September 2018, Google started to include close variants with exact match ads (source). According to, conversions decreased by roughly -28% (source).

In July 2019, Google started matching match words with the same meaning with broad and phrase match (source).

In September 2020, Google started showing fewer search terms in Google Ads (source), leading to another -28% decrease in shown search terms, according to Seer (source).

In February 2021, Google started to include broad match with phrase match (source).

Each of those changes slowly takes away control from advertisers. Step by step.

Optimizing for the ecosystem vs. suppliers

The trend is clear, but the impact is not. We don’t have enough large-scale analyses showing us how impactful the title and keyword-matching changes are, even though early data hints at significant impact.

Another question is “why now?” Google has been under anti-trust scrutiny for years. Taking control away from advertisers and suppliers won’t help.

How helpful are these changes for Google? The only explanation for Google launching such a risky change is that poor titles were a bottleneck for engagement growth. You could assume that a significant number of sites have helpful content but poor titles, which must have prevented users from clicking through and getting “their needs met.” Remember, Google is all about showing ads and helping users so they come back for more.

Pretty much every keyword with search demand or buying intent has at least one optimized site in the mix. It seems that wasn’t enough, though, or Google wants to pay more attention to topics without high demand. One problem, after all, is that SEOs tend to only go after keywords that have search volume as reported by Google keyword planner, which is often a fallacy.

I also see a potential conflict of interest: which sites does Google help and why? And where is the line? Things get spicy quickly if Google were to change titles in sensitive topics (think: health) that all of a sudden reflect incorrect information.

On the other end, control might be an interesting competitive advantage in the future. The company that can monitor its title and ad spend the most accurately might have an edge over smaller/slower competitors with fewer means.