I joined the Google Plus team shortly after its inception in 2010, transferring from the Blogger team. I spent three years on the project, working at first on the profile team, then transferring to the Growth and Engagement Team (GET), and finally ending up on Project Madonna and Project Zorro, where I helped with the repeal of the “real names” policy.

As a lowly software engineer level five, I was a relatively minor participant in the project, but I did witness a lot of the internal decisions made during the initial development of Google Plus, and I believe I can identify a number of critical mistakes made early in its history. This isn’t merely a case of hindsight being 20/20—even back then, I had a strong feeling these choices would prove harmful in the long run. What has changed since then is that I am now much more able to clearly articulate my arguments.

I don’t claim that the issues I’m going to describe were the only problems that Google Plus encountered, but I believe they were significant contributors to its eventual demise. Also, everything I am about to say is my own personal opinion, and not that of my current employer or anyone else.

Asymmetric following model

A symmetric following model (according to the terminology that we used back then) is one in which both parties have to agree in order to establish a connection. This is the model used by both Facebook and LinkedIn.

The founders of Google Plus wanted rapid growth of the social graph. They knew that beating Facebook was going to be a matter of scale.

An asymmetric following model is where one party can unilaterally establish a following relationship without the other party’s interaction. This is the model used by Twitter and Google Plus.

The founders of Google Plus wanted rapid growth of the social graph. They knew that beating Facebook was going to be a matter of scale. After all, people don’t go to social networking sites with the best features — they go to the ones that all their friends are on.

They believed that the asymmetric choice would produce a faster growth of the social graph, since it did not require a “handshake” where both parties agreed to be friends. And they were right—at least initially.

At the same time, they also wanted Google Plus to be a “close ties” network, one that you use to remain in touch with your friends and family.

However, the asymmetric model has a key weak point, which is that it’s a poor channel for bidirectional communication. When you “follow” someone, it is a signal that you want to listen to what that person has to say. But just because you follow someone doesn’t mean they are listening back.

This unidirectional model is actually a poor choice for a close-ties network because close-ties relationships depend on frequent bidirectional communication.

“Hi, sis, how’s it going today?”

“Oh, just fine, how are the kids?”

I can add my sister Christie to my Google Plus circles, which means that I will start to see content that she posts. However, there’s no guarantee that she will see things that I post.

Humans are motivated to communicate. But they are only motivated to do so if they think that people are actually listening to them. Simply shouting into the void is, for most people, highly demotivating. What is also demotivating is the lack of feedback. If there’s no signal that says someone received your message, then the natural, human response is to stop using that communication channel and find another one.

One of the things that became clear early on in Google Plus was that messaging was highly unreliable. Not in the sense that Google Plus would lose the message or that it wouldn’t be delivered, but rather that there was a good chance that no one would bother to read it.

This isn’t a problem with email because, even though there’s no signal that tells you that the recipient has read your message, you know how email clients work; you know that the message will stay in their inbox until they do something, so you have a high confidence that your message will be read, sooner or later.

This also isn’t an issue with Twitter because Twitter is not a close-ties network. With Twitter, communication is a statistical phenomenon: You don’t care if a given message is received by every one of your followers; what you care about is how many followers you have and that a high percentage of them received the message.

Rank-based presentation

Another factor that contributed to the overall unreliability of communication on Google Plus was the use of ranking. When you viewed content in your Google Plus stream, the content was ranked using a sophisticated algorithm so that “important” or “interesting” content appeared at the top, while lower-priority communication appeared lower down on the page.

In addition, the page also supported “infinite scrolling,” meaning there was no actual end to the page. As long as you kept scrolling, it would continue to show you more and more posts. What this meant is that you could never actually be “done.” There was no finish line, no sense of closure. Which in turn meant that each individual user would eventually have to stop, and where they stopped proved unpredictable.

For important things — like that letter from Mom — you absolutely want total control over what you read.

The combination of these two factors was, in my opinion, a fatal blow to the “close ties” goal. If my sister sent me a message, there’s no guarantee it would appear at the top of my stream, and if it appeared lower down, there’s no guarantee I would get to it before I stopped reading. And this is less of a problem for me than it is for my sister, who won’t know whether I read her message or not.

The problem with ranking in general is that it takes control away from the reader. After all, who is to decide what is and what is not important and interesting? The leaders of Google Plus were convinced that ranking was the key—after all, the entire success of Google was based on ranking algorithms—and that any problems could be solved by coming up with a better, more personalized algorithm.

Ranking can be a great help in browsing low-priority information (“grazing” behavior, such as reading the newspaper) because no one wants to have to make hundreds of decisions about what to read and what not to read. Having an automated “curator” makes sense in this context.

But for important things—like that letter from Mom—you absolutely want total control over what you read. (This is also why “Google Inbox” never caught on.)

As a result, users quickly learned that sending a post to a friend or family member had a high probability of never reaching the recipient’s attention. And thus, they went back to using email or Gchat.

Later on in the life of Google Plus, the company added a “notifications” feature that provided a separate user interface, visible in all Google applications, that displayed a list of messages sent specifically to you or that were otherwise deemed high priority.

But this had two problems: First, it was simply too late, and second, by creating a separate independent interface instead of fixing the core problem, Google weakened the overall experience. It meant that if you wanted the full Google Plus experience (both the stream and the intimate personal messages), you ended up having to read a lot of messages twice.

There were a few more problems with the ranked, infinite scrolling view I should mention. Since there was no “mark as read” button, the only way to detect whether you had read a post was to monitor when that post scrolled into view. Needless to say, this is a highly unreliable signal. A false positive means it thinks you have already read the article and won’t show it again; a false negative means you get to see the same post over and over.

The reason for this is that the user interface designers wanted to simplify the interface as much as possible and not require the user to click “done” for each post. Unfortunately, I think they fell in love with their own cleverness and did not see the downsides of that decision.

Of course, Facebook’s news feed has many of the same design elements that I have described here, including ranking. However, important messages sent directly from one person to another have their own dedicated channel that works more like email, with strictly chronological ordering rather than ranking and with an explicit “mark as read” function. Ranking and algorithms are only used for nonessential postings.

The designers of Facebook clearly recognized that when it comes to social interaction, there’s no single one-size-fits-all medium that works for all situations. There’s a difference between essential and nonessential communication, between urgent and casual, especially when it comes to maintaining interpersonal relationships, and these should not be casually thrown together into a single bucket in the interest of simplifying the user interface.

Nonorganic growth

Despite the problems mentioned above, when Google Plus was first launched, it actually felt pretty fresh and vibrant. There were a lot of highly engaged users and diverse voices talking about interesting topics. I remember the excitement I felt at being able to follow Wil Wheaton for the first time. The network was growing—slowly, organically, but still growing.

However, that changed once they started onboarding Google users onto Google Plus in mass numbers.

Again, the executives were concerned about the problem of scaling up to beat Facebook. And they already had billions of users—that is, users who had Google accounts. How do you get all those Google users to start using Google Plus?

Project Hancock was the internal code name of the project designed to do this. It was going to set up a Google Plus account for every Google user. This is actually much more complicated than it sounds; it took a team of engineers somewhere around three months to accomplish it.

All of a sudden, Google Plus had millions of new users. And many of us started adding those users into our Google Plus circles. Only there was a problem—a lot of those new users didn’t respond or interact in any way.

It was shortly after this when I started hearing people refer to Google Plus as a “ghost town.” Only, it wasn’t a ghost town in the traditional sense of an empty, abandoned settlement. Rather, it was a town inhabited by ghosts. Specifically, “ghost accounts.”

You see, Hancock was supposed to create millions of new Google Plus users, but what it actually did was create millions of accounts. And these accounts were a problem.

As part of my work on GET, we cooperated closely with the analytics team next door. Each week we would get a one-hour presentation from a member of that team, a deep dive into some statistical metric or analysis. It was always interesting.

One of the presentations was about the effect of inactive accounts. The common wisdom was that “more friends = better.” That is, the larger your circles were, the more likely you were to be an engaged user. However, what the data showed was a much more nuanced picture. The real situation was “more active friends = better.” In fact, there was a slight negative correlation between user engagement and the number of inactive friends you had in your circles.

Essentially what this meant was if you had lots of inactive friends in your circles, it made for a poor overall experience, leading to disengagement. It was actually better to keep your circles small and limit them to active users only. Unfortunately, there was no easy way for users to know which of their friends were active and which were not.

With a symmetric following model like Facebook, this would be less of an issue because in order to “friend” someone, they would have to be active in order to accept your invitation in the first place.

In my opinion, Project Hancock started Google Plus down a path it would never recover from.

The combination of these three factors—asymmetric following, excessive reliance on ranking, and large numbers of inactive accounts—created a sense of uncertainty when composing messages to family and friends. In Facebook, I can have a one-on-one, rapid fire conversation with my sister, both of us having a high confidence that our messages will be received and read.

With Google, I had to go outside of Google Plus and use a different app, such as Gmail or Hangouts, to accomplish this. More importantly, I would have to go outside of the circles/following model and use a different set of connections, such as Gmail contacts. This weakness undermined the overall design goal of providing a “close ties” social network.

Meta: Groupthink

Google has some of the smartest people in the world working for them. Why didn’t they see these problems early on, and correct them?

In fact, many did see them and tried to get the execs to change course—to no avail.

Somewhere around early 2011, I had lunch with Vic Gundotra, the head of Google Plus, at one of the Google cafes. Gundotra wouldn’t normally have hung out with a midlevel engineer like me, but he had heard about my former career as a game developer and was curious.

He asked me, point-blank, what I would change about Google Plus, and I mentioned the asymmetric following model. He disagreed vehemently, claiming that, in fact, the asymmetric following model was one of the best decisions they had made in the project.

When the execs are extremely smart people making ten times what you do, there’s a tendency to give them the benefit of the doubt.

The other problems I mentioned—about ranking, infinite scrolling and nonorganic growth—were also raised by others to our managers and various decision-makers. It wasn’t that our objections were dismissed out of hand; the people in charge of the product vision had detailed, coherent arguments as to why their way was the best choice.

At the same time, we couldn’t be absolutely certain we were right. We couldn’t provide concrete evidence that the current direction was wrong. And when the execs are extremely smart people making 10 times the salary you do, there’s a tendency to give them the benefit of the doubt. Surely they must know what they are doing.

But those execs were deeply committed, both intellectually and emotionally, to their particular vision of the product. And they were surrounded by peers who all believed the same way they did. The result was a classic group-think scenario. There was a strong bias toward doing things a particular way—such as depending on smart, sophisticated algorithms to decide what was best for users instead of letting users actually choose for themselves—because that is what had always worked for Google and, to a lesser extent, Facebook.

It’s true that as years passed, it became increasingly clear that some of the initial design assumptions were flawed. But by then it was too late to change. Reworking the following model after billions of relationships had been established would have been too disruptive to the users. Once we had started down that path, we could not turn back.

Aftermath

By 2014, it was clear that Google Plus was never going to be the “next Facebook.” I sat in on many of the open strategy discussions hosted by David Besbris, who was now the leader of the project. Unlike his predecessors, he seemed open to the idea of fundamental change to the product.

However, the strategy they chose was somewhat of a downer. Essentially, Google Plus was going to completely give up on the goal of being a “close ties” network. Instead, they were going to double down on being an “affinity” network—that is, communities of people who share common interests but don’t necessarily know each other in real life. The reasoning was that many affinity-based communities in Google Plus were still doing well, and so perhaps the product could be successful by focusing on that audience.

In other words, instead of trying to be the next Facebook, they were going to be the next Tumblr.

By that point, I had left the Google Plus team and was working in the JavaScript systems group; although much of the work I did was related to Google Plus, I wasn’t part of the product team. I left Google in late 2015, so I don’t know what happened after that, but I gather that the product continued to decline before it was finally closed this year on April 2.

I wish that Google Plus had succeeded. As many people have observed, there was a lot to like about it, including many subtle design aspects that were superior to Facebook or any other social network out there. And I truly dislike Facebook, both as a product and as a company. My refusal to use Facebook means that I tend to miss out on a lot.

I think a lot of people are as dissatisfied with Facebook as I am. But the thought of someone coming along and making a new social network, one that doesn’t have Facebook’s flaws, seems hopeless. After all, if Google, with all its vast wealth and talent, couldn’t do it, then who could?

But that’s not the right way to think about it. Google Plus didn’t fail because Facebook is invulnerable. It failed because of deep flaws that were embedded in it from the start. And learning from those flaws is the first step to building something better.