In many ways, personalization is a good thing. Internet users are so inundated with content, that filtering through the noise and presenting only information of interest can reduce user effort and enrich online experiences.
Many users know that every interaction online is tracked and analyzed. All of this data is tagged and segmented to create individualized customer profiles, which drive the delivery of personalized content (stories, products, ads, and information) to us online. But when does personalization become a problem?
In our research, we observed some of the downsides of personalization on the web. In particular, one of the problems with a personalized experience is that users are placed into a niche and start experiencing only information that goes into that niche. However, individuals are often multifaceted and change over time. A system that caters to a single user facet risks becoming boring or even annoying and can miss opportunities.
Precision and Recall in Personalization
The problem with overpersonalization is similar to balancing precision and recall in information retrieval (IR).
In information retrieval:
- Precision is the percentage of retrieved search results that are relevant.
- Recall is the percentage of all relevant results that the search system actually retrieves.
When applied to search, these metrics define the relevance of search results shown to users in relation to all possible relevant results in order to assess and optimize the search tool. For example, one of Google’s original competitive advantages was that it recognized that precision was more important than recall in the context of searching an almost infinite data source like the web — in other words, displaying only relevant results for a query is better than returning every result that could potentially be relevant (at the penalty of including many low-value hits).
Similarly, many digital products appear to be prioritizing precision over recall in personalization as well — trying to make sure that recommended content is definitely relevant to the user. As a consequence, they adhere strictly to what they already know about users (topics of interest, liked content, etc.). However, the exact same strategy that’s great for searching is less useful for browsing and monitoring.
For example, let’s imagine a user of an app for browsing images and video. This user has two primary interests — she loves both cats and dogs.
In this imaginary app, there are 100 cat posts and 100 dog posts that could be shown to the user (plus thousands of posts about other topics). She would love to see any of the 200 cat or dog posts, but she’s only liked or shared cat posts so far. This platform doesn't have any indication that she likes dogs. As a result, it shows her only the cat posts.
The user is getting a very precise, relevant set of posts, but she’s missing out on the potential diversity of content topics. Ideally, she’d like to see both cat posts and dog posts, but the app isn’t optimizing to its fullest potential and allowing her to discover new and interesting content. Furthermore, maybe this user would think that a platypus was the cutest thing ever if she ever saw one, but the handful of platypus photos in the app will never be shown.
Overpersonalization is dangerous both for organizations and users: (1) users see the same type of content again and again, with little chance of expanding their horizons and interests; (2) companies get fewer opportunities to learn about their users.
This approach treats users like one-dimensional characters, rather than the complex multifaceted individuals that they are.
A Homogenous Experience
We’ve seen above that overpersonalization can produce poor recall (in the information-retrieval sense) and that it may cater to just a few of the many interests that a user may have. But do people care? How do they react to overpersonalization?
Content Fatigue
We [the authors] are starting to get bored with our Instagram feeds. All we see are cats and jewelry. Yes, we both like cats and jewelry, and Instagram has certainly taken note. Our interactions with these types of posts have created a snowball effect. Much of the content that is algorithmically promoted to us by the platform fits within these categories, so over time our feeds have become saturated with the same homogenous content. (Instagram does include a Discover feature for finding new types of content, but it is separate from the primary feed experience, and must be consciously sought out by users.)
Our study participants expressed similar feelings; in some cases, this level of overpersonalization has caused users to lose interest in these platforms or abandon them altogether.
Several US study participants scrolled through Instagram and Facebook without even reading the posts, noticing the lack of new and interesting material in their feed. Comments included:
“This is boring”
“boring scan this morning,”
“I do think I see the same people on Facebook all the time; no doubt about it.”
One user even noted that because the content was boring she continued to scroll looking for something that was interesting, “I don't find anything interesting on Facebook tonight but what's funny is that I will keep scrolling until I do; it's addicting.” This behavior is related to the Vortex phenomenon, which refers to people feeling sucked into the online world almost against their will through sticky design techniques (like continuous content feeds). Users seek the emotional payoff they get from a good piece of content. In these cases, the phone turns into a mini slot machine: they keep pulling the lever coming across dozens of losers in hopes of finally getting a winner.
Several participants in our China studies for the Life Online project reported using TikTok (a Chinese music- and video-sharing social platform) heavily in the beginning: discovering, sharing, and engaging with the rich content on the platform was fun and almost addictive. Some users recounted that they would sit and flip through videos for up to an hour at a time. But then they got bored. Everything started to look the same, so they stopped using the app. “It’s pretty much the same content being posted over and over again. I don’t have much of a desire to keep watching. It gets boring over time,” said one study participant.
Content fatigue due to overpersonalization doesn’t only exist on social channels. In our research on recommendation systems, an Amazon user noted stale and boring product offerings in the Recommended for you section of the page. As she browsed through the filmstrip of products, she commented, “Some of these things like, Argan oil, I looked up years ago… So it’s not very updated. Literally when I was looking for sunscreen that was like 2 years ago. This is really stretching. The last time I ordered something like this might’ve been like 3 years ago. I’ve definitely looked at products since then and it doesn’t replace them [the older product recommendations]. It’s not as live-time as Sephora. You’re going back quite a few years, but my tastes have probably changed. I bought the lactic acid probably like 4 years ago. It feels very outdated.”
This example shows how placing users’ interests into a narrow segment can cause content to feel boring and even dated. As our participant noted, interests are often fluid, changing over time. In addition, Amazon may miss the opportunity to continue learning about these interests and promote fresh and more relevant products.
In fact, some users who have experienced overpersonalization in the past are beginning to fear interacting with sites that they know will use this information to customize future content. Another user browsing through Amazon’s Explore section, started hearting things she liked. “I liked the squid and that's giving me more squid like things […] here's a problem with liking things, what if it narrows [out] other things you might like. I don't want it to suggest only octopi even though I think octopi are pretty cool. What if it becomes like when you were a kid and you tell your mom you like cats and everything you get for Christmas is cat shaped; you're like, I'm not defined by cats, I'm this other person, too.” Thus, overpersonalizing content could actually influence people’s interactions with content, which could impact the validity and comprehensiveness of information gathered about these users interests.
Redundant Content
Additionally, the content homogeneity observed in social apps like TikTok and Instagram is exacerbated by community members creating derivative and duplicative content. Thus, on social media, popular posts end up being replicated by other users who seek the same popularity. In fact, there is an Instagram account (@insta_repeat) that identifies Instagram tropes — posts that are similar (for example, having the same type of landscape and the same viewpoint or camera angle), although created by different users or at different times.
We observed one Chinese user intentionally planning to reenact some popular images in her upcoming trip to Paris. She browsed through social posts on Redbook that had been taken by other users in Paris. The photos inspired her to visit the same locations, intending to recreate her favorite posts on her own account. She took screenshots of the photos to catalogue them and to use them during her trip:
“I want to imitate the travel experience of popular blog posts I saw online -- taking pictures at the same location with the same pose and from the same angle.”
Echo Chamber
Another major symptom of an overpersonalized homogenous experience is the echo chamber effect. An echo chamber is a situation created by hypertargeted content in which our beliefs are amplified and reinforced by the delivery of a single type of content. The echo chamber effect fuels social polarization between people.
For example, on Facebook people like and engage with content that resonates with them. So, over time, as the platform identifies their interests, the algorithm delivers posts targeted to them, creating a world in which they consume only the types of messages that reinforce their beliefs and interests. Therefore, people no longer see messages that challenge their beliefs or broaden their points of view.
If every experience is focused on matching our interests, we’ve lost the opportunity for a unifying experience. The result: people are divided as clearly in real life as the virtual segments within these content networks.
Before hyperpersonalized content, the shared experiences of consuming the same news helped create a sense of community. Our worldview was shaped by reading and experiencing the same information, and people were more united as a result. In March of 2017 The New York Times announced that it would begin experimenting with delivering personalized news to its online readers. This news elicited concern from many readers who felt they could be missing out on the shared experience of being exposed to the same news and stories as everyone else.
Being “Downright Creepy”
Personalization can easily be perceived as creepy in some situations. Though most consumers are aware they are being tracked for personalization purposes, it can be unnerving when a single action generates a barrage of related mailers, ads, emails, and articles.
We asked some participants in our Life Online study how companies determine which ads to show them. Most users knew that ads and other content were personalized based on prior interactions. They would often back up this observation by expanding on their feelings about this targeting, or telling us about a time they searched for one thing and suddenly continued to see it all over the web.
For instance one Chinese participant told us, “If I look up toilets on Taobao, there will be information about bathroom appliances everywhere. The downside of this is that if I already bought this thing, the ads still keep showing up. It's annoying, and I'll feel bad if I see a better deal after I bought it.”
Most of our participants were lay users — they did not have an accurate understanding of the kinds of data that companies use to personalize their experiences. Though they were aware that sites used interaction data on that particular site, they did not understand how sites could promote content or serve ads related to their activity on other sites or on other devices. When personalized content seemed to break the boundaries of what they think the site or devices should know, the practice seemed especially creepy. For example, one user in our China study commented that she understands why, on her phone, she may get content related to her mobile browsing activity, but that seeing ads related to her mobile activity on her computer is very “creepy”. A Raleigh user also found Facebook to be “creepy” because it seemed to know and take advantage of her Amazon and Google search data.
(While we do recommend seamless omnichannel user experience, the passing of state from one context to the next should be limited to cases where users expect and prefer an all-knowing backend system.)
Unnerving feelings can also be triggered by companies that detect and personalize content around sensitive information. For example, pregnancy information and products, funeral preparation, or embarrassing medical issues can cause strong emotional reactions when shown out of context. Some online interactions are meant to remain private or should be approached with care.
How to Avoid Overpersonalization
Now that we have the technical ability to track, analyze, and personalize to the very last detail, it’s becoming clear that friction certainly exists between the desire for personalization and some of the yucky drawbacks that come from overdoing it or getting it wrong. Perhaps it’s time to take a step back and evaluate our personalization efforts. Based on the pain points observed in our study, here are several recommendations that companies should consider to prevent overpersonalization:
- Broaden overly narrow segments. Consider widening segments to allow for some level of personalization but still provide variety. The New York Times’ Personalization FAQ page indicated that the organization seeks to strike a balance between delivering personalized stories and general content.
- Mix it up. Don’t rely too heavily on tagged interests. Create content strategies that purposefully introduce nontargeted content in a thoughtful way. Consider mixing targeted campaigns and content with introduction of new or popular products outside the user’s defined interests. Also consider how these messages are delivered. Outgoing messaging channels like email and SMS can be used to deliver content of known interest, while social channels may be better suited for mixing up the content — because in these situations users are likely more open to browsing and discovering new and unique information. With a screenful of posts, there’s very little penalty from including one item that’s generally interesting (e.g., gets lots of clicks from a broad user base), even if it doesn’t fall within the current user’s specific interests.
- Don’t use a blanket algorithm, finetune for special situations. Some topics can be more sensitive than others. Identify if any targeted content may fit this situation and take a second look at how you are personalizing around it. It may be worth being more conservative in your approach in regard to these subjects.
- Provide a way for customers to give feedback. Users should be able to indicate whether targeted content is problematic or annoying. Give customers some sense of control over the personalization.
Share this article: