Google Labs have released a new tool called Google Correlate.
The tool works in reverse to Insights for Search as the user starts with a trending term to determine keywords which correlate by search interest trends over time or by geographic location (currently US only). The main similarity to Insights is the normalization of the anonymous Google search data resulting in no sense of actual search volumes – just relative scales. Correlate generates terms which closely resemble the trend of a search term or set of data over the period 2003-2011. The tool uses the Pearson Correlation Coefficient in an automated process to determine the trend.
Correlate ‘eggs’ and you get a general uplift around Easter and therefore high correlated keywords include hymns such as ‘He Lives’, ‘buns recipes’ and ‘Origami bunny’.
Why Correlate is not so great
Correlate tends to turn out twice as many non-relevant terms as it does significant ones. We must remember that correlation does not intimate causality largely due to the objective automation of results.
Users of correlate can find any number of spurious correlations, such as ‘wedding day’ and ‘pest control’ (r=0.7872) and ‘Adwords’ and ‘detention centre’ (r=0.9613)
How to Sense Check Correlate Results
Search marketers not only need to assess and filter the list of correlated terms but also:
1. Cross check back, e.g. if term b correlated for term a, re-input term b to see if term a is a high correlation for term a. If not, then this suggests that it is not a good match for term a.
2. Choose terms to correlate carefully as core generics may be too general and thereby correlate with too many diverse categories to draw any significant conclusions. Similarly, niche terms may not correlate usefully either.
A list of correlated terms from the tool is no better for keyword expansion than the Google Keyword Tool or Insights for Search because the highest correlating terms which are relevant will also emerge from these other tools.
Insights Is Better for Search Marketers
Insights has better functionality; being able to select date ranges and different categories – these may well be added into Correlate in the future in order to improve relevance between search term and correlated keywords. Insights is a good tool to assess how a campaign has changed search behaviour, e.g. above the line driving users to search.
When to use Correlate?
- Branding campaign (cf. eggs example above)
- Marketing a seasonal/annual product or for a campaign which highjacks search interest around a singular event
- An additional application is to shift the correlation by a few weeks as an indicator of interest before or after an event for ticket sellers for example.
- Star Use: Upload a data set – this can range from publicly available data (I have used Dow Jones closing prices, Gold prices and precipitation levels), Insights data or your own Adwords data. With the Dow Jones closing prices data set I found high correlations with ‘wellsfargocom’ and ‘how to invest money’ and with a Travel Insurance client impressions data from Adwords I have found strong correlation with ‘check status of passport’. These can be used to track search behaviour and new keyword themes. N.B. You need to log into your Google Account to upload data and also be aware it is saved on the system under your account. Google have stated that they will monitor data for spamming – clients may not want sensitive data uploaded into the tool, although you can delete it.
Impact on the Search World
This release comes during a period when forecasters are increasingly turning to real-time web search data as an indicator in addition to traditional measures (which typically have a time-lag associated to them and are non-reactive to actual events and attitudes). This allows the opportunity to ‘nowcast’ using real-time web search data to analyse a current phenomenon. Just last week, the Bank of England released an article as part of their Quarterly Bulletin describing their continuing use of Google Insights for economic forecasting. By analysing the relative trends for keywords, search marketers have long been able to derive seasonality and singular changes in consumer behaviour – for example the increase in searches for ‘job seekers allowance’ during the recession.
Correlate can make these assertions more concrete by comparing to real world data rather than just being based on search data (as with Insights). This is the really consequential feature – it can be used to determine a relationship between the real world and search. It has its flaws and needs work but this is a step in the right direction for Google data. It is legitimizing search as an indicator of world events and consumer opinion in an age when many commentators are turning to social media for this information as the sheer amount of search data has been too un-mineable. Ultimately, prior to commenting on a newsworthy current affair or trend, most users are likely to search around the topic first on Google, making search the earliest indicator of a trend but also the most unbiased. A comment recorded forever in social media will be carefully thought out; a search term is a blunt & true reflection of the user’s mind.