anrmq7 341 days ago [-]
Fake followers and fake engagement are widespread on Instagram - much more prolific than advertisers realize. I’ve worked with “influencers” for years. In the past few months brands, advertisers and agencies I speak to are voicing their concern. So, I'm helping build a product (based on analysis of 80 million profiles so far) which will allow brands steer clear of fake influencers, e.g. those women being paid $X/month to walk around NYC and promote handbags/yoga clothes/etc. based on the 50k fake followers they've bought from some website.
FilterSweep 341 days ago [-]
> based on the 50k fake followers they've bought from some website. This, and its not just limited to buying fake followers.

There is an effect with real followers unintentionally following pages by either churn&burn or clickfarms, as well.

Facebook has this occur more widespread. The churn n' burn goes as follows: 1) Build Meme or image sharing page, 2) Accrue 50K+ followers, 3) Sell page to an influencer, influencer updates pictures and profile 4) Followers of old page have no idea the influencer they're following, and now see ads mixed in with the old content.

All and all, the overinflation in that industry is reaching bubble levels, if it hasn't already passed it.

MuffinFlavored 341 days ago [-]
Can't Instagram very easily determine what accounts are fake? I know they already have measures in place to prevent fake account creation.

They could even use the algorithm in this article to flag 90% of all "most likely fake" accounts, and require a series of daily captchas to remain active?

technotarek 341 days ago [-]
I can't help but think of the movie the Big Short. There's a large pool of social media marketers, influencers and other tangentially related businesses, all in some way propped up by this fake activity. They're not unlike all the hands that got involved in the mortgage business and securitizing of sub-par, high risk loans. If only there was a way to "short" the Instagram industry.

p.s. To those debating what constitutes a "fake" account, don't forget that the term was only recently invented:

have_faith 341 days ago [-]
Similarly, I caught some non-technical friends recommending some other friends that they should buy into bitcoin / altcoins as an easy way to make money, reminded me of the same film.
nrhk 341 days ago [-]
Well they haven't been wrong yet, just doubled in the last month.

Regardless these are all small bubbles that will only have localized effects when they collapse. Much more important bubbles to worry about like the derivatives market.

irrational 341 days ago [-]
Long ago when bitcoins first came out and it was simple to generate some on a personal old pentium desktop I considered doing so, but never bothered to since I assumed they wouldn't last. Doh!
ringaroundthetx 341 days ago [-]
Their definition of fake is too limited, and they ackowledge that and then rely on it anyway because there isn't an agreed upon definition of fake.

That is very strange, given that there are already better ways to determine if the accounts "will provide value to advertisers", and that is by raising the bar a bit more to look at the account's recent likes and activity.

just because an account has less than 10 posts or is private doesn't mean that they aren't actively engaged humans. that is the root flaw of this analysis.

ivan_gammel 341 days ago [-]
I think the classification can still be supported by the analysis of the behavior of these accounts and the other statistics on the accounts of "influencers" with high ratio of fake accounts. For example, the latter tend to have significantly higher than average ratio of likes to followers on their recent publications. Also, purchases of followers usually happen in batches, so it is easy to observe the spikes in followers numbers. The behavior of "like" bots also produces some observable patterns.
ringaroundthetx 341 days ago [-]
> Also, purchases of followers usually happen in batches, so it is easy to observe the spikes in followers numbers.

I would disagree with this. It is very easy to have a steady stream of followers on instagram, I've run a few campaigns myself.

When an account is private, and promoted, instagram queues up follower requests to 100 at a time, and you can't see how many are queued but you can predict how many users have been attracted to the account. Bot accounts or otherwise.

(Note: this is part of a larger strategy of getting instagram's algorithm to self-promote your posts in the explore section, or as popular in a hashtag. As you can post something new and your new followers will like your post and then you can switch your account public soon after and it is boosted in activity on many feeds, attracting more people to your account)

The user - who is the influencer - account can manually approve the pending followers, or take their account public. When taking the account public, 100 pending followers are added automatically, and it has to be switched back to private to show the next pending followers, only for them to be added when the account is toggled to public yet again. If the account stays public, the other pending followers will stay pending indefinitely and the user can't see that they are pending.

Therefore, this gives the account the ability to control a steady growth of followers, 100 at a time, indefinitely. If they keep the account private during their actual campaign. The authors might be surprised that this strategy obtains more humans than bots.

The main thing here is that there are many ways to "buy followers". And I think the analysis here assumes way too much about a single way that followers are bought. The authors have probably never created their own influencer account, to know about the infrastructure that really exists.

nsomaru 341 days ago [-]
Could you share more thoughts about the larger strategy of post and hashtag promotion? Interested to hear your experiences, as we've been dabbling in some FB and Instagram marketing.
ringaroundthetx 341 days ago [-]
On instagram, It is easy for an account to come out of nowhere in hashtag promotion.

Just like I was explaining it can basically bake in followers while private. An account can go from 0 - 10,000, or 20,000 pending followers almost overnight, but more practically in a week's time. This won't register on any analytics site.

The account can accept as many of the new followers as it likes (this takes time though) and can make a new post and stack it with hashtags while private. The post will begin getting a lot of likes in a short time span, and then the profile can go public.

For maximum efficiency, this should still be in conjunction with at least one other campaign, still funneling new followers in while the account is public. The post will go to the popular section of every hashtag except the most popular hashtags, as well as trend in the explore section for many users who don't follow you yet.

Given the way that campaigns work, using stories and tags in photos, these should mostly be human followers.

altern8tif 341 days ago [-]
Wouldn't it be obvious if you constantly switch your account back and forth between public and private?
ringaroundthetx 340 days ago [-]
No, nobody notices.

Your existing followers cannot tell if your account is public or private, because they see your profile.

Your prospective followers encounter either a closed down profile, which is still being promoted for some reason making them curious.

Or they are presented with an open profile, which is still being promoted for some reason and worth following.

propogandist 341 days ago [-]
It is extremely easy to buy [fake] accounts that will mimic real user activity. These were available on certain forums 3-4 years ago, before people started "catching on".

Having tinkered in the 'dark economy' meant to defraud advertisers before being employed by the advertisers, I trust nothing. The 'dark economy' will be always be miles ahead

jaclaz 341 days ago [-]
>Secondly we are suprised by the high number of fake followers in the business. Almost a third of 7 million followers of Swiss influencers appear to be fake. About 30 of the analyzed influencers appear to have more than 50% fake followers. Among those are quite some who advertise products of smaller and larger Swiss companies on a regular basis.

So, in a country (with all due respect) hardly "at the center of the universe" for (say) design, fashion, and similar with approximately 8.5 million people, 115 "influencers" have 8.5 million followers.

Even taking into consideration foreigners (after all it is the internet and it is "global"), is it a surprise that a large part of them are "fake"?

fwn 341 days ago [-]
The US has a population of roughly 323 million. @katyperry, @justinbieber, @BarackObama, and @taylorswift13 have 387 million followers. What does that tell us?

Nothing. As you say, the internet is global.

jaclaz 340 days ago [-]
Well, I find hard to believe (remember that this is limited to Instagram and to "market influencers", none of the ones mentioned are rock stars, presidents of the US, etc. that I believe have a more transversal audience) that the total amount of followers is so large.

The intended audience seems to me pretty much "narrow" at least from the few examples given on the article under "Illustrative examples of Swiss Instagram influencers".

These are seemingly all fashion related, targeted to an audience (my personal estimate) of women in the 16-40 years old range, maybe even a more restricted age range.

And Switzerland is not - I believe - considered an international (worldwide) reference for fashion.

Of course my reasoning is based not on any data, so it may be well off, but I would have expected an even higher number of "fake" followers, not a smaller one than what the researchers found.

beepboopbeep 341 days ago [-]
It's amazing to me that more of this hasn't taken place. With valuations being driven by total # of users, where is the due diligence to verify these claims?
eanzenberg 341 days ago [-]
Valuation is more-so driven by ARPU so all this does is shift ARPU up.

Also, there are network effects. It's unknown whether Instagram would be a better service (to investors) if all (half, a quarter, etc.) fake accounts were removed.

sharkmerry 341 days ago [-]
If they know numbers are inflated falsely, arent they duping advertisers than? If using CPM and 50% of impressions are fake, that halves their revenue. Facebook, Twitter, etc, all the same
eanzenberg 341 days ago [-]
I can't speak for instagram facebook etc. but the industry is continuing to move away from cost per impression (which can be duped by fake accounts) to cost per "action", where action happens further down the funnel (registration, 28-day retention, in-app-purchase, etc.)
eric_h 341 days ago [-]
Yeah - we worked with a company that bought some traffic on a cost per impression basis relatively recently, and all of a sudden we had a significant burst in errors being reported.

The user agent for the devices experiencing the errors were all an old version Firefox.

Turns out that version of Firefox was the default for some version of Selenium, and all of that traffic was coming from bots running in data centers.

Moral of the story, don't buy cost per impression from any company with anything but the best reputation, because that market is saturated with bots.

sharkmerry 341 days ago [-]
How is that determined though? If I google for a brand name and the first link is an Ad by the company to their website and the first organic link is their website. Does google get credit for that lead if I click the ad vs the organic link? I was going to the brands website either way
ivan_gammel 341 days ago [-]
I'd guess that it would be traffic-expensive for bots to produce impressions, unless its a distributed botnet or it is a question of survival for bot accounts.
pinky1417 341 days ago [-]
Assuming fake users don't add to revenue (by not clicking on ads, for example), shouldn't fake users depress ARPU? The numerator remains constant while the denominator increases.
eanzenberg 341 days ago [-]
Let's say they are making $100M revenue on 100M users, so ARPU $1 / user. But half of those users are fake. Their actual ARPU is $2 / user but to the outside world it doesn't matter, their $100M revenue hasn't changed.
bennesvig 341 days ago [-]
It's easy to tell on an individual basis if an Instagram account has something screwy going on.

- Check comments to followers ratio. - Check comment quality (are people tagging friends or leaving generic/spam "great pic" comments?). - Follow the account and turn on post notifications. By doing this you'll see how many spammy ads they post then delete an hour later. Many of these posts will have fake social proof on them.

neves 341 days ago [-]
The first fake account is impressive because it links to a real Facebook account. Don't they have any kind of verification?
wnstnsmth 341 days ago [-]
They don't say that "fake" followers have to be bots - they could also be humans that "sell" their profile to platforms so they can re-sell it again for following. If you buy "quality" followers on, for example, you get exactly that. Accounts that appear to be real people - and are probably - but who follow several thousand of other profiles. And that is just weird.
KekDemaga 341 days ago [-]
Alternative idea for a service: Identify other “influencers” in your market niche and buy them thousands of obviously fake followers, when in discussions with advertisers point this out: "X has more followers, that is true, but look at the quality of them most of these are obvious bots. That is why my rates are 20% higher."