This one was easy to answer, because the Good Magazine article you linked to, links directly to a report (PDF) from Demographics Pro - a company trying to show off their Twitter analytics software.
The report is not peer-reviewed and is self-published, so should be taken with a grain of salt. They include their methodology, but not the raw data.
Trump supporters were qualified as a sample of 10,000 US citizens, each following Trump on Twitter AND tweeting at least one pro-Trump hashtag (see Appendix A2) during a 7-day period prior to the third presidential debate. Clinton supporters were similarly qualified. A selection of 10 white nationalist Twitter accounts (see Appendix A1) were qualified as being affiliated with one or more Southern Poverty Law Center (SPLC) designated hate groups AND tweeting content indicating a white nationalist viewpoint.
From this methodology, we can see that the definitions are a tight subgroup of Trump supporters, so the headline "Study Shows That Over 35 Percent Of Trump Supporters Follow White Supremacists On Twitter" is somewhat misleading.
Some potential sources of error that exaggerate the effect include:
- Not all Trump supporters are on Twitter. An August 2016 article suggests the the number of Americans adults that use Twitter is in the 20-25% range, making the chance that 35% of Trump followers being on Twitter at all unlikely.
- Not all Trump supporters follow Trump. With 24.8 million followers, globally, at the time of writing compared to 63 million votes it is clear that not every voter follows him.
- Not all Trump supporters that follow Trump tweeted about him using one of the given hashtags in a particular 7 day period.
Some potential sources of error that reduce the effect include:
- Only 10 White Nationalist Twitter accounts were identified. If they found more, the numbers on each side would be (at least a little bit) higher.
Other potential sources of error:
They say they restricted the sample to US Citizens, but they provide no information about how that was established.
Some non-Trump supporters, non-supporters of the White Nationalists, non-voters and even non-people may fit the criteria for "Trump Supporter" (and likewise for "Clinton Supporter")
In conclusion, even if the study was accurate, the article's headline about "Trump supporters" or the questions summary of "Trump Twitter followers" are over-generalisations of the studies sample. Those claims are incorrect.