Dorothea Wiesmann Rothuizen and Wojtek Skut are engineers at Google, where they work on continually improving AI and large language models. They are a part of the team implementing new capabilities into Google Ads features, such as broad match.
There’s been a wave of interest in artificial intelligence (AI) in recent months, largely due to immense improvements in the underlying technology. Large language models, for example, now enable a deeper understanding of language, which has led to conversations with AI that are much more natural and intuitive.
This same technology has supercharged broad match, a Google Ads setting that helps brands reach more people with their ads and reduces manual keyword work.
Broad match has existed since the dawn of Google Ads, but it wasn’t perfect. And it started losing momentum about 15 years ago. Back then it could have matched a search query such as “treating a pet at home” to an irrelevant keyword like “treats for pets”. At the same time, “treating a pet at home” would not have matched to the related “without a vet”, as the two phrases don’t contain any of the same words. But thanks to new advances in AI, and continuous updates, broad match can now better interpret nuance and context.
Today, broad match understands people’s search queries on a deeper level and can identify intent. And this is making it one of the most effective solutions for search advertising.
How new AI advances are broadening marketing horizons
When broad match launched, our engineering team at Google would manually write out the synonyms for a keyword and cast a really wide net. For example, an ad keyword including “cheap” could match with searches for “inexpensive” and “shoestring”. While “shoestring” can be a different way of saying “budget”, a user could alternatively use this word to search for shoe laces.
Thanks to new machine learning capabilities, large language models now better understand a user’s intent. We train it with billions of pieces of text so it can learn all the different variations and meanings of a word or phrase and what sequences make sense. Only once that fundamental foundation is in place will we assign it a specific task, such as matching ads to search queries.
While the language technology can understand the potential demand that’s out there, it’s most effective when combined with Smart Bidding. This automated bidding strategy factors in a wide range of signals and surrounding data points in a privacy-safe way — such as search history, interests, and past purchases — to build a predictive model that helps advertisers identify those audiences most likely to convert.
Today, broad match understands people’s search queries on a deeper level and can identify intent.
Broad match in 2023: What’s new for marketers
Here are some of the biggest, recent improvements our engineering team has made to broad match:
Order matters, and broad match knows this too
In the early days of broad match, the technology would connect the dots between keywords in a search query and an ad, but it wouldn’t necessarily take the order of the words in the query into account.
If someone was looking for a flight from Beirut to Abu Dhabi, they could have been shown an ad for a flight from Abu Dhabi to Beirut. To prevent this from happening, search specialists within Google would’ve recommended a different keyword strategy to those advertisers in the past. But today, broad match knows that going from A to B isn’t the same as the other way around. And this same improvement has also been applied to other areas where a search query only works one way.
A user might be looking for a place to “exchange dirhams to egyptian pounds” or they’re searching for a “usb to usbc cable”. Seeing an ad in which these words are swapped around would not be relevant to them — but broad match now understands their intent.
Routing traffic to the right keywords in Google Ads
A question we’ve been asked in the past is: “What happens when two ads for the same advertiser match to the same search query”? Previously, broad match would look at an advertiser’s Ad Rank to help determine which ad to show, and this may not always have been the best choice.
To improve relevance — for the advertiser and the user — it now takes supporting information even more into account, such as other keywords in the ad group and the ad’s landing page. And thanks to the increased capabilities of the large language model powering broad match, it’s also able to better understand the meaning of the search query and all of its variations to improve relevance there too.
Another area that has seen a lot of development is connecting the dots on multilingual searches. Broad match now recognises that some people might switch between languages when looking for something online — and it uses that information to serve the right ads.
This is often the case with expats, such as Spanish natives living in the United Arab Emirates, or British people residing in Saudi Arabia. They may be searching in their native language, but are looking for local-language results for the country they live in.
We’re not saying that you only need to run your campaign in one language and that you’ll get traffic for all, as the success of your campaigns relies on your own language settings and localised creatives. But in countries with a lot of bilingual expats it’s good to know that broad match can match relevant, local traffic to an ad, even if the search is made in another language.
AI-powered language upgrades have made broad match a much more effective tool for search advertising, but it’ll continue to need a smart marketer in the driver’s seat.
The role of broad match in search advertising today
Broad match may not be quite as old as “Back to the Future”, but it’s been around for a long time. And just like in the iconic 1985 movie, it shows us that sometimes you have to look at something from the past to realise its value in the present.
That said, while AI-powered language upgrades have made broad match a much more effective tool for search advertising, it’ll continue to need a smart marketer in the driver’s seat. These savvy marketers will check performance, select the right bidding strategy, and monitor the new queries that broad match identifies to help bring in new customers. After all, you might discover new insights about your audience and their behaviours throughout the campaign.
AI learns from you and you learn from AI. Broad match’s success depends on the team that uses it, just like the time-travelling car from “Back to the Future” needed the input and creativity from its creator, Doc Brown, to drive into the past — and the future.
Tune into the Google Marketing Platform podcast episode "Talking AI: How a deeper language understanding is transforming broad match" to hear more about the ways large language models are supercharging broad match. You can find it on the GMP Academy website and all major podcast platforms.