In this blog, we will discuss the most recent advancements in Google AI (artificial intelligence) and its impact on paid search marketing.
Google’s mission statement is “to organize the world’s information and make it universally accessible and useful” – and over time, they’ve done just that. Google has developed algorithms to better understand how people search across the world and how information is organised. In short, Google is trying to make its machines think like any human being, regardless of language. This blog will focus on the two latest advancements that have been announced by Google, exploring how they have impacted the way paid search accounts are built and managed.
The BERT update
Announced in 2018, the Bidirectional Encoder Representations from Transformers (BERT) update brought advancement to the way Google AI matches queries with search results. This model was based on “transformers”, a model that processes words in relation to other words within the sentence, as opposed to individually and in order. BERT models are now able to consider the full context of a word by reviewing the words that come before and after it, making it particularly useful for understanding the intent behind search queries.
This sort of model obviously needs to be trained for each language, with Google initially releasing this update for English and later for many other languages. They’ve implemented several new updates to this model, one of which included BERT’s capacity to understand all the nuances of verb and preposition collocations that have the ability to change the meaning of a word in English (such as “from” or “to”).
BERT’s impact on paid search: Exact Match Variants and improved query matching
The main impact of this update has been the introduction of close match variants to exact match keywords, starting with the inclusion of plurals and misspellings followed by different words and function words. In 2018, Google also introduced variations that share the same meaning as the keyword, including implied words and paraphrases. In short, “exact” doesn’t necessarily mean “exact” anymore.
Whilst these updates were not overly well received by the community, especially for brand keywords where variations may not always be fully relevant, it allowed marketers to showcase their ads for a larger set of queries without having to build out exhaustive keywords lists. Now, Google is able to understand that users searching for “Book flight” or “Book Flights” most likely share the same intent. The same applies if search terms are in a different order. For example, the intent of a user searching for “buy red shoes” and “red shoes buy” is the same.
Another change that has impacted the way we run paid search accounts is Google’s expanded understanding of users’ search queries. If you have been following Google AI updates, you know that Google has been focusing heavily on the system’s ability to understand nuances that are very common in each language. A good example of this is the English word, “to”. In the past, if you were searching for “Train from London to Paris”, the search results may have also included ads for the return journey, although the intent of the query was to find travel options from London to Paris.
With this update, Google AI is better able to match and fully understand queries, thus serving better results to the user. Users can now expect better query matching, which should also lead to less time spent reviewing search queries. In fact, Google has reduced the availability of queries, a change that has been unpopular but also embraced by marketers. This improvement in query matching has shifted the focus towards improving creatives and developing activities of the account.
The MUM update
Announced in May 2021, Multitask Unified Model (MUM) is said to be a thousand times more powerful than BERT! MUM is a new milestone in AI, as it not only understands language, but also generates it. It’s trained across 75 different languages so everyone should benefit from this advancement. Recently, Google demonstrated MUM’s capabilities by surfacing over 800 vaccine names from across the world in just a matter of seconds. This process would usually take weeks to complete.
Additionally, MUM is multimodal, meaning it has the ability to understand information across text and images. The model will also soon be able to understand the copy of a page along with its images, allowing MUM to better understand the context of a particular page and serve better results to the user. In the near future, Google hopes to further improve MUM so that it has the ability to understand audio and video formats as well.
MUM’s impact on paid search: The end of broad match modifiers and “Clever” broad match
In June 2021, the Phrase Match behaviour update began rolling out and was fully available to all languages by the following month. Broad Match Modifiers (BMM) allowed advertisers to increase the reach of keywords without going too broad, allowing for some level of control over the queries that would be triggered. The latest Phrase Match update, however, combines the expanded reach of BMM, while still giving marketers control over queries.
Since July 2021, marketers are no longer able to create new BMM keywords; instead, they will behave like Phrase Match keywords. In short, you will only be able to run three different match types in an account (previously allowed for four), an update that should make keyword management much simpler.
Additionally, Google recently announced another update focused on the increased accuracy of query matching for Broad Match keywords. In the past, Broad Match keywords played a huge role in driving impressions and clicks, but were often responsible for triggering irrelevant queries; thus, marketers were required to carefully weed these out to keep traffic relevant.
The rollout of MUM and its ability to understand images and text is definitely a huge upgrade, as it now has a much better understanding of the context. The multimodal aspect of MUM is key to the refinement of the query matching process of Google AI. In fact, we wouldn’t be surprised if a new “Smart Matching” match type becomes available in the near future. @PPCGreg on Twitter recently noticed Google testing for “Smart Matching”, to which Google responded that this was simply a “bug”.
This new “Smart Matching” match type would completely change the way paid search marketers build their accounts by limiting them to just one match type instead of two or three. In our opinion, this would be a welcome change, as it would allow for a less complex structure, more straightforward keyword planning and more comprehensive ad-copy building. Machine learning and bidding automation would also benefit from this change, as traffic from relevant queries could focus on one keyword instead of two, allowing the system to learn quickly from more data.
What’s coming next?
Advancements in AI will keep bringing users back to Google, especially as search becomes more conversational and machines continue to gain an impressive understanding of user intent, the human brain, and different languages. However, it’s also important to keep in mind that algorithms only see what they are fed, and human input will remain key to running paid search activity. Understanding the objective of a business and translating this into a paid search strategy over time will remain a key part of pay-per-click marketing that the machine is not capable of (yet).
The MUM update has just rolled out last month, and we are yet to see all of its uses and benefits. But if there is one thing we are sure of, it’s that the model should allow marketers to spend more time focusing on crafting compelling copy, developing the account and looking for new opportunities to grow revenue, as opposed to focusing on manual, lower-value tasks.
As a first step, we would recommend marketers review their current approach. Then, they should set a plan for how they could streamline their accounts, by making keyword management simpler and leveraging broader match types to bring incremental volume that may have not been captured previously. Combining automation with expertise, our in-house team at Croud has taken this approach to now see an average 15% increase in conversions, whilst also reducing cost per action by a similar amount.