According to a recent North Carolina State University study, Facebook doesn’t account for context when assigning a user an interest that is used for ad targeting.
The study (“Analyzing the Impact and Accuracy of Facebook Activity on Facebook’s Ad-Interest Inference Process”), suggests that Facebook doesn’t consider positive, negative, or neutral reactions to a topic. They used green cheese as an example.
Let’s assume you shared a post about how you dislike green cheese. The Facebook algorithm, based on the mention, would infer an interest in green cheese. The algorithm would completely ignore that you dislike it. You could then see ads promoting green cheese.
While the example may seem silly, it makes an important point. Replace “green cheese” with the interest you are targeting. In most cases, you assume that the users you target have a positive interest in that topic. You may be wasting your money if you pay to reach those who actually dislike it.
The researchers conducted a couple of studies to come to the conclusion that this was happening.
Study #1
In the first study, the researchers created 14 new Facebook user accounts. Researchers controlled the demographics and behavior of each account. They then tracked the interests that Facebook generated.
The researchers could see these interests within each account’s Ad Preferences. You can, too. Read this post about your Facebook Ad Preferences.
Researchers found that actions as simple as scrolling through a page led to Facebook inferring an interest. According to the study, 33% of the inferred interests were inaccurate or irrelevant.
Of course, determining whether something is “inexact” or “irrelevant” is a sliding scale. At what point is an inferred interest accurate and at what point is it inexact? When is it relevant and when does it become irrelevant?
Regardless, we may be splitting hairs when one of three inferred interests is problematic. That’s a lot!
Study #2
The researchers performed another study. This time, they recruited 146 people from different parts of the world. Each participant downloaded a browser extension that allowed researchers to collect data from Facebook about their interests.
The participants were then asked questions about the accuracy of the interests Facebook had inferred. In this case, researchers found that 29% of the interests Facebook listed were not actually of interest to these people.
In other words, these results were consistent with the findings of the first study.
Check Your Ad Preferences
If you’re skeptical, you can check the ways that advertisers can target you based on interests, too. It’s actually really hard to find now, and I’ll dig deeper in a separate post. But, from the desktop app, go to…
Settings & Privacy > Settings > Ads > Ad Settings > Categories used to reach you > Interest categories.
Like I said, it’s buried! That’s six clicks just to get there.
I’ve found that some of the interests that can be used to target me are creepily accurate. Some, though, are the complete opposite. I have no idea why they’d be associated with me.
Some examples:
- RVs
- CamperVan
- Motorhome
- Golf Stroke Mechanics
- Workers’ Compensation
- Canvas Print
- Baking
- Diaper
- Preschool
I have no idea why I’d have inferred interests related to RVs and camping. We almost never go camping, and we’ve never used an RV, campervan, or motorhome.
I don’t golf. Not one bit.
I’m self-employed and workers’ compensation is not something I’m concerned about.
I don’t bake.
I have kids, but they’re all 14 and older. Diapers and preschool haven’t been in my world in a long, long time. Maybe interests aren’t updated and they can be outdated? That’s not good either.
There are literally hundreds (if not thousands) of these, so I stopped looking! And what’s crazy is that I actually cleaned my interests up years ago, so mine are likely more accurate and updated than most people’s may be.
My Experiences
These results could explain some of what I’ve seen with targeting interests.
For example, I recently promoted a highly technical blog post related to Facebook advertising. I targeted interests that included some of the following:
- Advertising agency
- Advertising campaign
- Digital marketing
- Online advertising
- Social media marketing
- Social media strategist
The vast majority of the comments I received on that ad were from people who were quite obviously not advertisers. Instead, they wanted to voice their distaste for ads and how they wished they didn’t exist.
So, how is it that I may have reached these people?
If this study is accurate, we can guess that some of these interests may have been inferred for these people if they previously commented on posts or ads about how they dislike Facebook ads.
This is obviously a guess and a small sample example, but I’d be curious if you’ve had similar experiences that could be explained by this study’s results.
It Matters
Now, we don’t know how many of these interests are actually used for targeting. And, of course, targeting interests were cleaned up after these studies were conducted.
But, this certainly raises questions about how these interest lists are generated. Poor targeting impacts the user experience and it certainly impacts the advertiser’s results.
You can make an argument that in some cases, the context doesn’t matter about a person’s relationship with an interest. But, I’d suggest that advertisers would hope that, in most cases, the interest is actually due to a positive relationship.
Your Turn
What do you think? Is this an indictment on interest targeting?
Let me know in the comments below!