Date published
  • Damien Garaud

Time Series Prediction for SEO


Many events can affect the performance of a website: a deployment, a new internal linking of your website, a Google ranking algorithm update, etc. In SEO, it's extremely useful to be able to pinpoint when the effect of an event appears and to estimate the impact of these changes on your main KPIs: impressions, positions, SEO visits, CTR, etc.

These KPIs are time series: each value is associated with a time or date. For the impressions and clicks, your Google Search Console provides a value per query and per URL for each day.

This historic data can be a gold mine for your Business Intelligence teams to (1) analyze the past and decide what you have to do now, (2) understand the impact of your decisions, e.g. get insights based on new A/B testing and (3) predict future trends.

In this notebook, we're using Kats, Prophet and TensorFlow to analyze time series from GSC, to forecast future trends and to measure the impact of an event thanks to Causal Impact.

  • Detect some changepoints within the range of influence of a known event
  • Forecast clicks starting from a specific date
  • Measure the difference between forecasted and observed data
  • Connect Google Search Console and retrieve historic data
  • Save and/or load your data (CSV format) in Google Drive
  • Retrieve (1) impressions and (2) clicks
  • Process the data
  • Explore your data and plot impressions and clicks
  • Detect changepoints
  • Generate a prediction starting from the Change Point
  • Measure the impact using Causal Impact with TensorFlow