Vivian, a 25-year-old Los Angeles dweller, has 20,000 Instagram followers. It takes her five hours every day to engage with her audience, post and respond to stories and answer direct messages.
The hustle pays off at least partly in the form of brand partnerships. Companies pay Instagram influencers to post about their products — the more followers you have, the more money you can earn.
What is an Instagram bot?
So, when her friend started a now-shuttered social media management business that promised customers “500+ real followers per month or your money back,” she gave it a try — for a month.
“I hated it. It generated the worst traffic for my page,” she wrote in an Instagram message.
The company’s “around the clock” Instagram growth service cost money, and it didn’t quite work. Vivian’s new followers were largely teenage boys from around the world, she said — not exactly the target demographic for the crop tops and lingerie she helps advertise. Now, she’s back to managing her own Instagram presence.
On the surface, Instagram is a platform for sharing images. Under that swirls a bustling marketplace of brands — both corporate and personal — and the consumers they’re trying to reach. Lording above is The Algorithm, which awards better positioning in the app’s feed and “explore” function to accounts with larger and more engaged audiences.
It’s a game ripe to be played. Folks with time to spare can browse for content related to their own and engage with it, boosting their chances of gaining new followers. Folks with money to spare can pay someone to do that for them — or just pay for robot followers.
Folks with neither can download the open source tool InstaPy, a free, Python-based program that automates Instagram outreach and engagement.
Make It Easy, and Make It Work
InstaPy’s creator, Tim Grossmann, decided to build the program after reading Al Steigart’s book Automate the Boring Stuff With Python for a class at Stuttgart Media University.
Instagram growth companies, he reasoned, underpay offshore teams to do repetitive liking and commenting on behalf of customers. People must view that activity as “boring stuff.”
So he wrote some straightforward script that automated a browser to like Instagram photos with particular hashtags.
“Once I had some really basic automation in place, I realized it’s actually pretty simple to do,” Grossmann said. “From there, I thought about creating a little bit more sophisticated automation because I wanted to give every user on Instagram the same base to get what they want.”
To achieve that, his tool needed to be two things: 1) accessible to non-coders and 2) advanced enough to compete with pay-to-play Instagram management services.
Getting there has been a journey. InstaPy has always been easy to set up, thanks to detailed documentation, a Udemy course and automated installation prompts for Windows, Mac and Linux, Grossmann said. But getting it to do what users wanted was trickier.
At first, users had to edit the application’s command line interface themselves to tell it what to do. Unsurprisingly, adoption was lower than what Grossmann wanted. Then, his collaborator Felix Breuer built a simple graphic user interface so people with little programming knowledge could use the tool more easily.
And as interacting with InstaPy got simpler, its capabilities grew more complex.
“I wanted to give every user on Instagram the same base to get what they want.”
From the beginning, the tool focused on growth through engagement. Early users could focus the bot’s efforts by inputting certain hashtags or geographic tags they wanted their accounts to engage with. For example, a command might tell the instagram bot to find and like 10 posts with the hashtag #beach.
But Instagram engagement gets much more involved, and, over time, so did InstaPy. Now, a user can configure the program so that, every day, it likes 50 posts from the feed, finds five new accounts that have use the hashtag #beach, likes fives posts from those accounts, comments on two posts and watches three of their recent stories.
Eventually, Grossmann and InstaPy’s contributors added yet another layer: Automated image analysis with Clarifai’s API.
Before that AI component was implemented, InstaPy users would write a pre-set list of comments the program would cycle through. So, a picture of a beach tagged #beach and a picture of a dead whale tagged #beach might receive the same comment: “Awesome! :)”
With the computer vision technology, InstaPy can recognize the content of images, so users can command the program to find and comment on pictures of beaches, instead of just inputting hashtags and hoping for the best. Image analysis also helps users avoid content that’s offensive or off brand. Vegan influencers, for instance, can tell the bot to bypass any content featuring animals, in case trolls use a #vegan tag alongside less-than-vegan images of steak dinners or big-game hunting.
While InstaPy initially had a core contributor team of five or six people, Grossmann said, the project has garnered almost 3,000 commits.
The Average Users — and the Nickelodeon Star
Grossmann needed to make his tool sufficiently accessible and advanced — and he did. But measuring InstaPy’s success as an “Instagram equalizer” is difficult.
Since the project is open source, it’s hard to track who’s using it and collect data. According to InstaPy’s GitHub metrics, the repository receives 30,000 views and 500 downloads each week. Based on his communication with users, Grossmann estimated the InstaPy community is about 60 percent aspiring influencers, 15 percent “actual” influencers and 25 percent “very small” companies.
That suggests InstaPy really is doing what Grossmann set out to do — helping everyday content creators compete with people who have more resources to blow on pleasing the algorithm.
But not exclusively. One television actor on a Nickelodeon children’s show approached InstaPy’s maintainers asking for a custom feature — he wanted the bot to target the people who followed his co-stars, in hopes they would follow him as well. Grossmann and team built the feature for the actor, and earned some money doing so.
“I’ve never disclosed any company names or people or account names that use our service, and I think I never will.”
Grossmann declined to share the minor celebrity’s name because doing so could affect the actor’s career, he said.
“I’ve never disclosed any company names or people or account names that use our service, and I think I never will,” he said. “It can hurt their reputations and basically destroy everything we worked on together, even if they didn’t use the automation in the end.”
The problem, Grossmann continued, is that when it comes to automation on social media, people paint with too broad a brush. InstaPy isn’t the same as buying automated followers, he said, it just uses automation to optimize what people would do anyway.
Instagram, however, doesn’t make distinctions between good and bad automation, which is why InstaPy’s creators have worked to make the tool fly under Instagram’s radar — at least for now.
Covering Automated Software’s Tracks
Instagram’s terms of service prohibit “artificially collecting likes, followers, or shares” and “posting repetitive comments or content.”
Users who violate those terms can have their accounts disabled or “shadowbanned.” But only if Instagram catches them.
“I think the most interesting part about the whole tool is not the features we implemented, because they’re pretty simple,” Grossmann said. “The more sophisticated stuff we worked on was making sure that Instagram couldn’t instantly detect whether the bot is active in an account.”
InstaPy uses the front-end automation tool Selenium, which opens a browser to send requests to Instagram. The headers on those requests tip off the server that they’re coming from automated software. So, Breuer set up a proxy that acts like a middleman, obfuscating the source of the requests so Instagram is none the wiser.
The headers also show a “user agent,” which indicates what type of browser the requests are coming from. So, the InstaPy team added random user agent tags, like Chrome or Firefox, so the requests would appear to come from standard browsers.
Finally, the team scheduled the automated activity to mimic the habits of real people: There are random delays, to avoid posting a bunch of comments too close together, and no late-night activity.
“Sure, it is below-board in terms of being against Instagram’s terms of service. But you are just emulating what anybody can do manually.”
So, Instagram currently has a hard time detecting InstaPy users. And since the platform doesn’t list the likes and comments a particular user leaves, tools built to analyze for shady behavior on the platform probably can’t either, Social Audit Pro developer and co-owner Louis Nel wrote in an email.
Social Audit Pro is a company that helps brands and influencers detect Instagram bot followers and avoid fraud — brands don’t want to pay influencers with follower counts inflated by bots, and influencers don’t want to get caught with high ratios of bot followers. With their current technology, it would be “impossible” to detect the type of activity InstaPy automates, Nel wrote.
But even if they could, they wouldn’t care to, he added. No customer has ever asked for it, and it’s not the same as buying followers, according to Nel. Automated followers are fake, and they inflate a person’s follower count for purely cosmetic purposes. The followers InstaPy reaches, on the other hand, are real, and they provide real value on the platform.
“Sure, it is below-board in terms of being against Instagram’s terms of service,” he wrote. “But you are just emulating what anybody can do manually. Instead of employing a social media manager who just sits and leaves comments or likes all day long, you could just use automation to achieve the same thing.”
Sounds similar to Grossmann’s point. So if InstaPy just uses automation to optimize what humans would do anyway, does that make it OK to use?
Which Social Media Growth Tools Are ‘OK’?
On one hand, we have Instagram, third-party auditing services and companies condemning automated growth tactics like robot followers. Then, we have Grossmann and Nel explaining that InstaPy is different — it’s an optimizer, and maybe even a democratizer.
But according to Bailey Flanigan, a Ph.D. candidate at Carnegie Mellon University studying theoretical computer science, there’s a fundamental flaw with viewing any social media growth shortcut as a marketplace equalizer: The definition of “marketplace.” Once a community has been monetized, no amount of fancy automation will return it to an egalitarian paradise.
“I think that’s how markets trend, right? When there’s money available, it will become more and more exclusive to those who have more money. I don’t think there’s anything in place within this ecosystem to stop that from happening,” she said.
Flanigan is one of the co-authors of this social media study, which used machine learning and natural language processing to identify “pod activity” on Instagram. In a pod, groups of people agree to like and comment on each other’s content to improve everyone’s performance in the algorithm. Like InstaPy, most pods are free and simple to use. They can boost a participant’s follower account without any money changing hands. But, as Flanigan (and Instagram) put it, these tools promote “inauthenticity” — pods because they’re reciprocal, InstaPy because it’s automated.
Flanigan’s co-author Janith Weerasinghe, a Ph.D candidate at New York University studying machine learning, helped build the model that eventually learned to identify pod-boosted posts with around 90 percent accuracy. He said it’s unlikely that tools like InstaPy could evade detection forever if platforms were dedicated enough to sniffing them out.
“At some point, you would have spent so much energy creating a super smart AI, it might have been easier to create authentic content.”
“I think it’s bound to leave some sort of breadcrumbs that a machine learning algorithm with enough data would be able to pick up,” he said.
But Weerasinghe isn’t sold on the usefulness of a social media AI arms race. He questioned whether those efforts would be worth it — especially on the user side.
“At some point, you would have spent so much energy creating a super smart AI to get past whatever countermeasures social media platforms put up, it might have been easier to create authentic content that generates organic engagement,” he said.
As Grossmann acknowledged, it’s frowned upon to game Instagram’s algorithm. But frowned upon by whom? By a platform looking to improve user retention and time on app? By brands looking to reach customers through influencer marketing? By Instagram users without the tech savvy to download and set up an open-source package?
Clearly, the ethics of Instagram growth are tough to parse. The platform warns against “inauthentic” activity, and automated software could certainly be dubbed inauthentic by some. So could fully manual comment pods. So could literally anything any of us do or say at any time anywhere.
As for Grossmann himself, he only uses Instagram these days for direct messaging with friends.
“My hope is for people to use social media in the way I think the creators of those channels envisioned in the beginning, not to promote themselves to higher levels and try to make other people jealous,” he said. “But I’ve worked too much with people from social media and influencers to believe in that.”