AI SEO Weekly

Keyword Research Using AI

By @ValeIzzo86 & @CyberAndy

What is keyword research?

Keyword research is a practice SEO professionals use to find and research search terms that users enter into search engines when looking for products, services, or general information.

What is keyword analysis in artificial intelligence?

AI is used to differentiate the intent of a query in “informational” or “transactional”.  Here is an example implemented on Google Sheet using WordLift classification API.

This allows us to quickly understand if and when content should be split or consolidated.  Creating topic clusters around keywords is in general a good practice to organize content more effectively and improve rankings.

Keyword Insights

- Keyword Clustering - Keyword Intent/Context - Keyword Ranking

What is Keyword Insights?

Why to use AI in keyword research

To find clusters of related queries

To highlight opportunities based on the content that we already have

To discover new content areas 

How AI is powering real-time SEO

When we talk about query analysis using machine learning we can use different techniques to cluster queries, let’s review them together.

Supervised text classification 

When we know the correct output class for each text in a sample dataset.

Unsupervised text classification

Where the algorithm observes samples from classes that were not observed during training, and needs to predict the class they belong to, we use it for instance for intent classification. 

Named entity recognition 

We can extract entities from queries as we do on longtail.wordlift.io

K-Means clustering

It averages the data by identifying a centroid for each group and grouping all records in a limited number of clusters. A centroid is the imaginary center of each cluster. 

Word embeddings 

Embeddings encode the meaning of words using, real-valued vectors. So that words that are closer in the vector space are expected to be similar in meaning.  Once meanings are turned into math, we can use K-means to group them. 

Microsoft ORCAS

It is a click-based dataset with 18 million clicked query-document pairs for analyzing search

How to create a tool that generates queries

We generated a Projector starting from Microsoft ORCAS

We chose automotive related queries of Porsche:

Porsche 911 Porsche Porsche 986 Porsche Boxter Porsche Panamera Porsche Cayman Porsche Taycan

Let’s review the Porsche “buying intent”

Colab to create embeddings from a set of different queries