(Update: It seems you can use this to detect electoral fraud)

The method that works is:

1) Last names - not full names - (or whatever is a simplest shortest way to identify the candidate . E.G: Beto is better than O'Rourke - Mccain is better than John or John Mccain)
2) Date range starts from beginning of campaign to 1 week after the election.
3) Always Filter by geographical area where election takes place (Country, state, county etc..)

You do that, 9 out 10 times - the results match reality.

Note: the predictive system starts to breaks down when there are less than 500,000 voters, or when the election is too tight (candidates within 3% of each other.)





2004 election

[link to trends.google.com (secure)]

2008 election:

[link to trends.google.com (secure)]

2012 election:

[link to trends.google.com (secure)]

2016 election:

[link to trends.google.com (secure)]

2020 election: (past 90 days)

[link to trends.google.com (secure)]