I actually took the time to attempt a reasonably objective 'experiment' on Google search suggestions. The current candidates and the presidents of the last 30 years were used as subjects. Several positive, neutral, and negative terms were used. The queries were formed from the subject and the term. To simulate partial terms, each term was attempted in three ways: the first letter, the first three letters, and the entire term.
Here are the results.
It's rather long, but there is no evident bias in the results. Whether it's Hillary Clinton or Donald Trump or a dead president, the results are generally neutral at worst. There are some queries that result in negative suggestions, but that's not common. Hillary Clinton's suggestions include: "hillary clinton lying for 13 minutes", "hillary clinton racist kkk", "hillary clinton hates troops", and "hillary clinton violated law". For comparison, Donald Trump's results are pretty neutral sans for 'racist/racism'; better than Clinton's. In his case, 'crime'/'criminal', 'violent', 'lying/lie', 'hate', and 'fraud' have no negative suggestions.
The source code and data for the experiment is freely available.
Here you go. Requires Python 2.7. Modify the 'terms' files or 'subjects'. Run 'test.py' to perform the test. Keep in mind Google may blacklist your IP if you don't throttle the queries. I have it set to 2 queries per second.
In essence, cherry-picked search suggestions are crap. There is no conspiracy.