Up to 95 million Instagram users may be bots, according to a new study conducted by research firm Ghost Data for The Information.

In its most recent study, Ghost Data estimates 9.5% of Instagram users are bots. In 2015, using the same methodology, the firm found about 7.9% of Instagram users were bots. That analysis came after Instagram announced it had conducted a purge of millions of the fake accounts. At the time, Instagram had 300 million users. -The Information

Ghost Data reached their conclusion after buying 20,000 bots to analyze their behavior. The firm then matched said behavior with around 1 million accounts, and found that they were very active among the followers of popular Instagram accounts. When they did post content, they were usually photos of models that were found on other websites.

And the bots are evolving...

Andrea Stroppa, founder of Ghost Data, a five-person firm based in Rome, said bot makers have gotten smarter in recent years as social networks have started to crack down on them. While bots used to follow about 40 accounts for every one that followed them, they have reduced that ratio to about 5-to-1. -The Information

“For a long time, social networks didn’t do enough to find bots. I think they are doing better now, but I think they can do much more,” said Stroppa.

The Information also reports that other researchers who scan for Instagram bots have reached similar conclusions to Ghost Data.

Mike Schmidt, co-founder of Dovetail, a venture-backed startup that does research on bots for marketers, said he analyzed a sampling of the fake accounts found by Ghost Data. Mr. Schmidt said Dovetail’s method, which uses machine learning, wouldn’t label some of the accounts as bots. Still, Mr. Schmidt said that overall, his company sees roughly the same percentage of bots on Instagram that Ghost Data estimates. -The Information

The rise of the bot armies have defied efforts by social media platforms to combat the phenomenon. Facebook, which owns Instagram, says it has 10,000 people working on safety and security, which includes scanning for bots and malicious behavior. The company plans to double their staff of flaggers over the next year, while also using matching-learning techniques to identify fake accounts.

“We take spam, inauthentic and other abusive behavior very seriously,” a Facebook spokeswoman said in a statement. “Our internal estimates show that spam accounts make up a small fraction of Instagram’s monthly active user base.” The spokeswoman didn’t say whether Facebook’s internal estimates differed from Ghost Data’s.

Researchers who study online bot activity say that while the fake news and propaganda surrounding the 2016 election focused mostly on news articles, images and videos on Instagram may play a larger role going forward, simply because messages contained in images (including memes) are difficult to track and identify.

Bots also are “very good at amplifying,” said Giovanni Ciampaglia, researcher at the Network Science Institute at Indiana University. He said that fake news and political propaganda is often created by real humans who control social media accounts. But those posts get an initial boost by simple bots like the ones detected by Ghost Data. Millions of those bots can be controlled by one person and told to retweet or “like” a piece of propaganda or fake news. -The Information

The bots shown to The Information by Ghost Data appeared to be barely active - and used primarily to follow people and help amplify various postings by clicking on them. The bots are cheap to create since the process requires nearly no human involvement, while a tiny fraction of server time is required to run the software which controls the digital menace. Researchers at Dovetail refer to the unsophisticated bots as "Mules" - which are distinct from other bots which post divisive and misleading content, or fake news. The mules do, however, amplify such content to a larger audience.

While Instagram has gotten little attention over the growing bot problem, Facebook's bot issues are well known and hard to quantify.

It isn’t known how many bots reside on Facebook. Researchers said it was more difficult to track that information because Facebook offers privacy protections that prevent outside researchers from analyzing accounts. Instagram doesn’t have the same privacy settings. In a securities filing, Facebook estimated the number of misclassified or “undesirable” accounts on the flagship social network had increased in the third quarter of 2017 to 2%-3%, up from 1%. -The Information

Meanwhile, Twitter announced on Thursday that it had conducted a massive bot purge that left celebrities and other people with large accounts significantly diminished. Former President Obama lost over 2 million followers, around 2% of his 104 million. President Trump shed over 200,000 of its 53 million followers - around 0.37%.

Researchers have suggested that up to 17% of Twitter users may be bots - however the company says that number is closer to 5%.

Instagram, meanwhile, is becoming increasingly important to Facebook's business. Acquired for $1 billion in 2013 - the photo sharing social media platform recently hit a billion users - leading some Wall Street analysts to value it at $100 billion by itself.

Meanwhile, most of the bots found by Ghost Data are still on the platform, according to the consulting firm.