I just spent the last few hours reading through the research paper that was created by a facebook research team back in October 2018. It highlights and lays out various detecting measures along with blocking counter measures. I have seen this discussed in the comments, but I really believe it deserves a thread of its own. It does clear up a few things and I will outline them below.
if you haven’t read this, I highly recommend you do:
At the very least, read section 6, which outlines the strategy instagram can use to stop and block automation if need be.
Here are some critical points:
-
Instagram is can easily detect botting via checking outbound actions, ie, likes, comments, follows whether you are using mobile proxies, data center proxies or just your home IP. They have baseline user data which shows how normal users behave and how many normal actions are taken. Anything above that threshold is flagged as potential automation and botting.
-
Instagram recognizes that straight up bans would hurt the service, along with real people and their accounts. Instead of full on blocks and bans, they experimented with thresholds to minimize outbound actions such as follows and likes. And so there is definitely a threshold for number of actions. Also, straight up bans indicated, botting services would implement countermeasures which would make it harder to detect. This means, each time insta blocks a service for botting, or users, the botting agencies would implement new methods for botting and the cycle would continue. Hence, costing instagram a lot more money in R&D in the future. Something they would like to avoid.
-
The strategy that seems to have been suggested to work is implementing a threshold, along with a delayed counter measure. So let’s say you followed 200 people on day 1, instead of temp blocking the follow action, instagram allows the follows to go through and a few days later removes those follows. To the software, it looks like it successfully followed 200 people, in reality it didn’t. This strategy takes much longer to detect, and harder for automation softwares to counter. The research paper indicates this as potential solution, although I’m not sure if its being utilized at the present moment.
Further Implications:
From my understanding, instagram is always testing and improving its detection softwares. As of now, according to the research paper its super easy for them to detect all accounts, irrespective of the proxy type used (mobile, DC, residential etc). Instagram is also intentionally blocking some and not all accounts.
Think of it as a control group vs a test group. If you’re in the control group, your actions do not change, you may perform 400 follows per day and you will not get blocked. However, if you’re in the test group, even 100 follows can get you banned or temp blocked.
The very idea that instagram knows all the accounts that are botting, (lets say 90%), and are not intentionally blocking these accounts leads to one interesting outcome.
The data gathered by MPsocial users would very different. I have seen so many discussions on what proxies are best, what settings are best, something working for one person, that does not work at all for another. This can entirely be due to if you’re account is in the test group or control group.
So the next time someone puts out recommendations for the best settings, or the best proxy, just be aware that if you’re in the test group, I doubt these measures would help.
I’d love to open this up for further discussion on what you guys think.