A warning to those who want to better understand human behavior through social media: a new study says that most of the stuff posted online is relatively meaningless.

Two scientists published a study in the journal Science claiming that using data from social media networks like Twitter and Facebook get biased results, according to Canada Journal.

While the amount of data submitted through the networks may seem like a gold mine for advertising hoping to reach their target audiences, studies claiming to be able to predict summer blockbusters and market fluctuations are turning out to be woefully wrong much of the time.

Computer scientists at McGill University in Montreal and Carnegie Mellon University in Pittsburgh found that there were “serious pitfalls” to working with large social media data sets, which is a problem as a lot of academic research in recent years depends on data taken from social media. They often use the data to justify decisions and investments that can have worldwide impacts.

But the reality is that social media platforms attract different types of users — for example, Pinterest primarily appeals to females between the ages of 25 and 34, but researchers rarely account for this fact.

Also, many social media providers filter their data streams, something researchers aren’t privy to much of the time. As a result, the data the researchers are getting may not be an accurate representation of the overall data set.

Another factor is how social media platforms are design, which affect how users behave. On Facebook, the lack of a “dislike” button makes it hard to spot negative responses to certain content.

Bots often run rampant on social media networks, spamming the platforms with content. Oftentimes researchers mistakenly include this content in their research, being unable to distinguish them from normal users.

The fact of the matter is that research that attempts to explain things based on social media usually get it wrong, according to the study. For example, efforts to tag Twitter users with a political affiliation claim 90 percent accuracy but are usually right only 65 percent of the time.

The solution, the scientists say, is for researchers to be more careful in their data collection. They simple need to be aware of what they’re analyzing when they take a look at social media data and take extra measures to ensure what they’re looking at is unbiased and accurate rather than simply cast a wide net. Such research shouldn’t be dismissed, it simply needs to be refined, the study states.