The Only Broad Match Guide You’ll Ever Need *

I’ve spent 6 weekends researching language models to understand the mechanics of Broad Match so you don’t have to. Use this information at your advantage. Like and share if you find it valuable.

Keep in mind that Google “BERT” is used in the Broad Match. BERT is a language model. BERT helps Google understand the topics and concepts behind sentences, paragraphs and search queries.?

Ask yourself a question: When was the last time you saw an irrelevant answer to your query on Google Search? - This is how precise broad match is.?

Before BERT Broad Match was just matching keywords with the queries.

With BERT Broad Match understands the topics, intents and concepts behind the search queries, this moves broad match from semantic match type to the contextual.

History of Natural Language Processing:

• 1950: Alan Turing published his paper, "Computing Machinery and Intelligence," which introduces the idea of the Turing test, a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
• 1955: Frank Rosenblatt publishes his paper, "The Perceptron: A Model for Brain Functioning," which introduces the perceptron, a simple neural network that can be used to classify binary data.
• 1960s Machine learning research is largely dormant due to a lack of computational power
• 1970s Machine learning research re-emerges, and the perceptron is extended to work with multi-class data.
• 1980s The backpropagation algorithm is developed, allowing training neural networks on more complex data.
• 1990s Machine learning is used to solve real-world problems like fraud detection and spam filtering.
• 2000s The development of large datasets and powerful computers leads to a resurgence of interest in machine learning.
• 2010s Deep learning, a type of machine learning that uses artificial neural networks, is developed and achieves state-of-the-art results on a variety of tasks, such as image recognition and natural language processing.
• 2018 The paper "Attention is All You Need" is published, introducing the Transformer architecture.
• 2018 BERT is created, a large language model that is pre-trained on a massive dataset of text and code.
• 2019 BERT is used to achieve state-of-the-art results on a variety of natural language processing tasks, such as question answering, text classification, summarization, and natural language inference.
• 2020s Google started using BERT for Broad Match in February 2020. This was a significant update to Google's advertising platform, as it meant that advertisers could now use BERT to improve the performance of their Broad Match keywords.

What makes BERT different from other language models?

BERT is a language model pre-trained on a massive dataset of text and code. This means that BERT has learned the general principles of language and can understand the context of sentences and the relationships between words. Language models before BERT couldn’t do that.?

BERT uses different mechanics to achieve this:

• Masked language modeling: BERT is trained on a task called masked language modeling, which involves masking some of the words in a sentence and then predicting the missing words. This helps BERT learn the context of sentences and to identify the relationships between words.
• Attention: Attention is a mechanism that allows BERT to focus on the most important parts of a sentence. It does this by calculating the attention score for each word in the sentence. The attention score for a word is a measure of how important the word is to the meaning of the sentence. BERT then uses these attention scores to focus on the most important words in the sentence.
• Self-attention: Self-attention is a mechanism that allows BERT to understand the relationships between the words in a sentence. It does this by calculating the attention score for each word in the sentence, and then using those attention scores to create a new representation of the sentence. This new representation is more accurate and informative than the original representation.

How are these mechanics used in your google ads account?

Landing page.?

Broad match goes to your landing page to understand the relevance of the keyword to the query. For example if the keyword “rose” is referring to rose - the wine vs rose - the flower based on the landing page creatives and text.

How to use: Use one unique landing page per ad group, make the landing page highly relevant to your product/service and keywords in your ad group.

??

Other keywords in the ad group.

Broad match uses other keywords in your ad group to better understand the intent of the keywords. To do that Broad Match? extracts keywords and turns them into numbers, here is example:

How to use: Ad group should be a vessel for the keywords organized around one theme/subject. The more keywords are organized the better Broad Match understands your intent. Add intent and geography words such as: buy, around me, in, now.

Previous searchers.

Broad match uses search activity and location to better understand intent of the search query

Brand awareness.

Broad match understands that queries like “Brand A shoes” and “Brand B Shoes” are different and don’t match each other. This is done with the help of knowledge graphs?

Capturing all relevant queries with one keyword.

Broad match can capture hundreds of a highly relevant queries with one keyword, thousands with a tightly focused ad group

How to use: Reduce keyword count. Make sure they are highly relevant to each other and organized around one theme. = Your account becomes less complex and easier to manage. Broad match will capture search terms that wouldn’t be eligible to show due to low traffic.?

Smart Bidding.

Broad match uses Smart Bidding to optimize and bid on a search term level - that’s one level deeper than for other match types.

How to use: It is vital to feed correct conversion data. Smart bidding will optimize on any conversion provided, but the further away you are from sale - the less relevant optimisation will be. The primary conversion should be sales.

Best practice summary

Consolidate where possible

• Smart Bidding works best when you have a pooled budget and single bidding strategy across all campaigns with the same goals. You will get the best performance by keeping traffic together as it helps Google AI equalize the marginal ROI across all traffic, leading to the most possible volume at a given ROI level.?
• Smart Bidding works best when it can optimize towards the same marginal cost per conversion across all related traffic.?

Ad Groups

• Ad group should be a vessel for the keywords organized around one theme/subject. The more keywords are organized the better Broad Match understands your intent. Add intent and geography words such as: buy, around me, in, now. Write a highly relevant ad copy using the keywords. You shouldn’t need more than 5-10 keywords for a start. Add remove based on results if necessary.?

Landing pages

• Use one unique landing page per ad group. Make the landing page highly relevant to your product/service and keywords in your ad group.?

Keywords

• Reduce the number of keywords, filter by conversions or conversion rate and you will see that 10 - 15% of keywords are bringing 80-90% of the results - That’s Pareto Rule in action.? Make sure they are highly relevant to each other and organized around one theme. Broad match will capture search terms that wouldn’t be eligible to show due to low traffic.?
• Broad match includes all other match types in it, therefore do not duplicate the same keyword in different match types as it only hurts optimization. Instead use Segment -> Search Terms by Match Type

Smart Bidding

• Use broad match only in pair with smart bidding, because smart bidding will optimize and bid on auctions on the search term level.

Conversions

• It is vital to feed correct conversion data. Smart bidding will optimize on any conversion provided, but the further away you are from sale - the less relevant optimisation will be. The primary conversion should be sales.?

Broad Match implementation best practices

1. Implement all best practices
2. Apply Broad Match Recommendation in Optiscore
3. or Apply one click experiment in Optiscore -> Click three vertical dots and click “Apply as an experiment”
4. or Manually change match type

Wait for 30 conversions and analyze the results by clicking on Segment -> Search Terms By Match Type in your Google Ads UI

Conclusion

Have you ever wondered why something works for one person but doesn't work for another?

It's unlikely that you have a very special business case. Most likely, you're not following best practices. Something is off. For example, your conversion tracking might be wrong, your ad group structure might be ineffective, or your landing page might not be converting.

Now that you are equipped with best practices: I urge you to go and test Broad Match. It just works and brings incremental conversions within your targets.

If you like the post - please share the wisdom.

If you don’t like the post, thank you for reading for so long.

I am creating the same material with a checklist for targeting. Let me know in the comments if you want to get Broad Match and Targeting checklists.

Thanks

Sergey

* This post is my personal opinion on how BM works using logic, my understanding of language models and anecdotal data working with my clients with over 25 million in Google Ads revenue. I am not a software engineer, nor an AI expert.