The 'Learning' section allows users to get some insights into the ranking of search results for any search term or URL. You can enter a query to see which results performed best for that query, or you can enter a record ID (typically a URL) to see what queries are driving traffic to that record.
When optimizing content, it's common for people to strongly believe a specific result should rank highest for a particular search query. However, this is a biased view and often does not reflect views of your customers. Hence, it is a useful spot-check before you decide to try to optimize the ranking algorithm.
You can find relevant query-result pairs by searching for either a query or a record ID (typically a URL). You can also enter both to find one specific query-result pair only. This is useful if the combination you're looking for is not appearing in the top results list for a given query or visa versa.
When you run a query in this view, we show the results, query, the count of times the query-result pair appeared, and its performance based on a confidence interval calculation. 'Performance' represents the relative success of the query-result pair in generating user interactions (typically clicking on a result).
You can also download the data in CSV format by clicking the button on the top right-hand side of the page.
Consider the following example in which a user wants to understand click-behavior for a search query. They enter the search term "water restrictions".
The first URL for the search term "water restrictions" received the most number of clicks. Hence, its performance is "Good", with a 73% confidence. The second result received far fewer clicks than the first result, and hence it only has a 37% confidence.
Another way to look at this is to check the URL, and see which other queries users searched for that brought them to this page.
Here we can see that the queries "water restrictions" and "irrigation restrictions" have a higher confidence percentage than the other results. This means that the clicks for this URL were the highest from the two terms "water restrictions" and "irrigation restrictions".
However, note that the amount of data (number of appearances and clicks) in the example is quite small. As the data volume increases, the confidence value for the query-result set will be much more accurate. Secondly, the number of appearances in this report is the number of times the result appeared in a set where there was an interaction. So this number can be much lower than the actual number. This is because the results are competing against each other. Where there is no interaction they are all equal losers. That is valuable in other ways, but not so much for this report.