Engagement Was Always a Proxy

"Make your content more bite-sized, it will get better engagement" they said.

And I sat with that for a minute. Because they're probably right. Shorter videos do get better engagement metrics on most platforms right now. Carousels on LinkedIn outperform plain text posts. Reels beat long-form. The algorithm rewards the snack.

What keeps nagging at me, though... Did anyone actually learn anything?

Because engagement and learning are not the same thing. And somewhere along the way, the content world decided they were close enough. They aren't. And the data is starting to make that very clear.

The Metric That Ate Itself

Engagement started as a reasonable idea. If people are interacting with your content, or community, then something must be working. Likes, comments, shares, swipes... all signals that the thing landed, right?

The problem is what happens when you optimize for the signal instead of the thing the signal was supposed to represent.

You get content designed to trigger a reaction instead of transferring knowledge. The substance is gone, and the platforms made sure of it.

Facebook's own head of marketing science, Brad Smallwood, published findings in the Journal of Advertising Research stating that engagement has "no more chance of predicting actual business outcomes than a random guess." Their own internal research confirmed: "industry research has repeatedly shown that engagement rates do not correlate with the outcomes ultimately used to judge the success of marketing efforts."

Forrester piled on with a report literally titled "Stop Measuring Engagement" recommending companies track customer loyalty and lifetime value instead. And yet here we are, still building strategies around the metric everyone already knew was broken.

The Numbers Are Getting Worse

Even by its own terms, engagement is collapsing.

Rival IQ's 2025 Social Media Industry Benchmark Report analyzed 4 million posts and 9 billion interactions across every major platform. Facebook engagement dropped 36%. TikTok dropped 34%. X/Twitter dropped 48%, with multiple industries now at rates approaching zero.

Buffer's State of Social Media Engagement report (52 million+ posts analyzed) found Instagram's median engagement rate dropped 26% in a single year. The steepest decline in their dataset.

'Engagement' is migrating from public signals to private ones. The visible metrics are getting quieter while the invisible ones grow. Which makes the public engagement dashboard look worse and worse, even when something is actually resonating.

Your Brain on the Scroll

OK so the platform metrics are a mess. But what about the deeper question - "did anyone learn anything?"

The cognitive science here is pretty damning.

There's a well-established finding in memory research called the generation effect. A meta-analysis of 86 studies found that self-generated information is remembered nearly half a standard deviation better than passively consumed information. Active engagement produces meaningfully better retention than passive scrolling. Every time.

Social media is designed to make consumption effortless. That's a feature, not a bug, from the platform's perspective. But effortless consumption is, cognitively speaking, the worst possible condition for actually retaining what you just saw. Gavriel Salomon's research on "Amount of Invested Mental Effort" found that when people perceive a medium as easy, they invest less mental effort and score lower on comprehension tests... while simultaneously feeling more confident they learned (Frontiers in Psychology).

The usual excuse for producing more short-form content is that that our attention spans are shorter now. Except, that's completely made up. Neuroscientist Edward Vogel has been measuring attention in college students for over 20 years. His finding: "It's been remarkably stable." Attention isn't shrinking. It's becoming more selective.

The Long Game Is Winning

While short-form platform engagement is collapsing, the channels built around depth and trust are going in the opposite direction.

A two-wave experiment with 663 participants found the newsletter group significantly outperformed social media groups on actually remembering what they'd read. Higher multitasking during social media use correlated directly with weaker recall. The format wasn't just more enjoyable. It worked better.

The market reflects this. Substack grew from 2 million paid subscribers in 2023 to over 5 million by early 2025 — a 2.5x increase in under two years. The platform hit a $1.1 billion valuation after a $100M Series C in July 2025 and reached positive cash flow in Q1 2025. These are people voluntarily paying for long-form writing. Many of them pay for several publications at once.

Depth is surviving. Depth is what gets returned to.

So What Does This Mean for Communities?

Attention is commoditized. The walled gardens are high and getting higher. You can't simply post your way to relevance anymore.

Engagement as a metric has the same problem, just one layer deeper. Engagement told us people were paying attention. Attention told us people were present. Neither one told us people were actually getting something…

I watched this play out firsthand at a previous employer. A team running webinars measured themselves on two things: registrations and attendance. Did someone say they'd show up, and did they show up for any amount of time. That was it. That was "engagement."

Nobody was asking how long attendees stayed. Whether anyone made it to the end. How many questions got asked. Whether any attendee answered another attendee's question... which is a completely different signal than did they turn up.

Registration and attendance are symptoms. They tell you a body was present. They tell you nothing about whether anything actually transferred.

Start asking harder questions. Did this member do something differently because they were here? Did the person who asked a question leave with something they could actually use? Did anyone's behavior change?

It's not easy, and the tools here aren't great yet. But you can start crudely. At Microsoft, we didn't just track whether a question got marked answered… We spot checked and followed up. Did you actually implement it? Could you move on? The difference between "answered" and "done" sounds small. It isn't. One tells you the thread is closed. The other tells you someone's life got easier!

Trust compounds when people get outcomes, not content, and certainly not the feeling of being 'engaged'.

Outcomes make people come back. Build for outcomes.