Advertising to ⅔ Fake Users on Facebook

Short summary:
Facebook has a ton of fake users and they will sell you fake likes for your posts from them.

The full story:

I do a fair bit of advertising on Facebook, sometimes as much as a couple hundred dollars a month. This is mostly me doing “Boosted Posts”, which are easy to set up and don’t require going through the Ads Manager. Based on past experience, I try to set the targeting to “People who like your page” or “People who like your Page and their friends”. I also make sure that the location is “Living In: United States,” as that’s where most of our customers are. I’ve found previously that if you don’t do those two things, your ad will be shown to people who have no apparent interest in what we’re selling and would have no way of buying from us if they did.

Facebook’s “Boost Post” dialog.

Facebook recently started suggesting that I set up an appointment with one of their advertising specialists to figure out how best to use Facebook ads. They have a signup page where you enter your account info and a phone number, and you can book a time to talk to a real person about Facebook ads.

I did that, and here’s what she suggested:

  1. Set up a Facebook tracking pixel on my website. You can then create a custom audience for this in the Ads Manager (as opposed to the similarly named Ad Center) to target those people with ads. This was going to be too complicated to complete during the phone call and requires time to collect the data anyway.
  2. Create a custom audience for people who have visited my business page on Facebook in the past year. This I did during the phone consultation and turns about to be a little over 5000 people.
  3. Create a custom “lookalike” audience for #2, limiting it to around 1% of the total population and restricting it to USA only. This was supposedly an excellent way to get your message to customers who might not have heard of you. We also set this up during the phone call. This is about 2.1 million people, according to Ads Manager.

Ads Manager audience interface

Soon after that, I had a sale I wanted to promote, so I thought this would be an excellent time to put the sale message in front of the lookalike audience. So I boosted the post, setting it to end in 2 days and putting $100 behind it.

Pretty soon, the Likes were flooding in. I had over 1000 after the first day. A great success, so I thought.

Usually after a new person likes a post, I try to invite them to like the page as well, so that they get future updates. I also check out their Facebook profile page to see what they’re interested in so that I can make more content related to that topic in the future.

But I started noticing problems with many of these new likes. They tended to have a few things in common in their profile:

  • No or almost no public posts.
  • The only public posts being an update of the cover photo and profile photo in 2017.
  • The profile photos tended not to be headshots: a photo of a cup of coffee, say, or a landscape.
  • Between 20-100 friends.
  • Friends tended to be in places like India, the Middle East, or Southeast Asia.
  • Their gender didn’t match the name or photo about half the time: a “Martha” referred to as “he” by Facebook, for example.

Here’s a very typical example of this kind of profile:

My apologies to Korolina Rusnak if he turns out to be a real person.

Anyway, I was getting hundreds of these kinds of likes. I can only think this is part of some scheme for selling likes outside of Facebook. So I paused the ad and went back to just promoting to friends of people who already like my page, located in the USA only.

Still, was this money well spent? Sure, I’m getting fake impressions and likes, but maybe the boosted post is getting in front of enough real people to make it worth while. Let’s take a look:

Boosted Post Results

The first line above in the Ads Manager is the Lookalike audience performance. The second line is my typical “Friends of Friends” approach.

Notice that with the Friends of Friends audience, which, though manual inspection, consists of mostly real people, the engagement rate is about 1.7%.

Notice that on the Lookalike audience, I got 1,055 post engagements (likes, clicks) and a reach of 1,524 people. This theoretically means that ⅔ of the people who saw my post decided to engage with it.

I can only conclude ⅔ or more of the Lookalike audience is fake.

Facebook has to know this is going on. They recommended to me this advertising approach. They show my post to random sample of their users, and ⅔ or more of them are not real.

So for now, I’ll skip advertising to the Lookalike audience.

Followup 2019-1-14: The Register covers this blog post, including no on-the-record response from Facebook.


Walking Fighting Robot

My current fighting robot plan is to keep and maintain my 12 pound robot Flipper, but forgo any extensive upgrades (like going brushless). Here’s Flipper winning at MomoCon earlier this year:

What I’m really after is to build a walking robot for robot combat competitions. I’ve been wanting to do this for quite a while, and it looks like the technology has progressed to the point where that’s now possible to do as an amateur. If Boston Dynamics can build a really awesome walking robot for $100 million, I figure I should be able to build one that’s 1/4 as good for a couple thousand by piggybacking on all the existing R&D. This will likely be a 4 legged robot, because that seems a lot easier than 2 legs.

The plan is to start with a software simulation of the robot to work out the control and gait using ROS/Gazebo or something similar. Then step up to small physical bot using Dynamixel servos or one of the Chinese knockoffs. That size robot might be able to compete as a beetleweight or mantisweight. Depending on how that goes, I’d then want to step up to a sportsman-size or maybe something even bigger.

The amateur walking robot world has stuff like openDog going on, as well as this robot dog on hackaday.

I expect that this will all take quite a while. I’m still learning ROS and Gazebo, but not finding much time for that.

Alternative to Google Custom Search

Google Custom Search is really, finally coming to an end in just a few days. If you’re looking for an alternative, may I recommend AddSearch, which seems to do do everything GCS did, with a few nice bells and whistles added on. It starts at $30 a month, with annual discounts, so it’s cheaper than many of the alternatives, but still affordable for a smaller site.

Double Skeumorph Lightbulb


I noticed that these flame tip light bulbs are a DOUBLE skeumorph.

Obviously the shape of the glass outer container is supposed to bring to mind fire from a candle or gas light.

But this is also an LED bulb whose light elements are arranged to bring to mind an old fashioned incandescent bulb.

Services I’d Like to See: Stripe for Purchase Orders

Business-to-business purchases are a huge part of the economy, but the way they’re typically implemented just doesn’t line up with how e-commerce is conducted.

The traditional purchasing process goes like this:

  1. The operations or engineering department of a business needs some widgets. So they tell the purchasing department that they need some widgets with a certain specification.
  2. The purchasing department sends out quote requests to one or more widget suppliers.
  3. The sales department of each supplier replies with a quote.
  4. The purchasing department picks a winning quote and replies with a purchase order (or PO) to the winning supplier. This is a legally binding document that specifies what they’re buying and at what price. These typically have an associated PO number that is unique within the buyer’s organization.
  5. The supplier ships the widgets. The supplier’s accounts payable department sends an invoice to the buyer.
  6. Some time later (as negotiated), the buyer’s accounts payable department pays the invoice.

Compare this to a typical e-commerce purchase:

  1. The customer finds the widget online.
  2. The customer adds the item to the shopping cart.
  3. The customer checks out and pays using a credit card or PayPal.

There’s a lot fewer steps here. It turns out, though that the traditional purchasing process has certain advantages. By having all purchases go through a purchasing department, it’s much easier to keep track of expenses. And by negotiating payment terms, the customer can take their time paying: a great help to a business’s cash flow.

Also, it turns out, many large purchasing departments simply don’t have access to a credit card or PayPal to pay for stuff. They can pay via PO and that’s it.

As a small business owner, I like the purchasing power of companies that have a traditional purchasing process, but don’t like dealing with waiting for them to pay, sending reminders when payments are due, and handling physical checks.

So what can be done to smooth out this process? How about a payment processor that deals strictly in purchase orders! The first half of the purchasing process would be the faster e-commerce model: the customer finds the widgets on the vendor website and adds them to the shopping cart. But when it comes time to pay, they can simply use their PO number.

Sending reminder emails, collecting the money, assessing late fees, and so on would all be handled by a third party. This would be the “Stripe for Purchase Orders“. Let’s call them SFPO for short.

SFPO could offer an API, much like Stripe, PayPal, etc. that would ensure that the customer is approved for purchase orders. The API would do typical payment processor stuff, like providing a form, registering a sale, and so on.

SFPO could also provide other valuable associated services, like running credit checks on companies who want to pay via PO.

You could cobble all of this together yourself, but if you’re a small shop, why would you want to spend time chasing after customers to pay, when you could be building your business?

This is potentially a HUGE business as it touches many of the high dollar business to business transactions.  If you build this, please let me know.