Everything Is Skimmable Now

Everything is skimmable now. But what’s worth stopping for?

I re-read The Shallows the way most of us re-read anything: one eye on the text, one ear waiting for the ping. Twelve pages in, my thumb twitched toward a tab that was not the book. A few seconds of switching. A small detour to look up something that did not matter. Then another.

That’s the problem Carr points at. Not that screens are cursed or that your attention proves your character. His claim is more annoying because it’s closer to physics:

You get good at what you practice. Your tools decide what you practice. And you won’t notice the training while it’s happening [1].

Carr wrote The Shallows in 2010. A tenth-anniversary edition arrived in 2020, pulling smartphones and social media into the frame 1 . That update matters. In 2010, “the internet” often meant a laptop. Now it lives in your pocket. It vibrates. It has opinions.

This piece walks through Carr’s core arguments with margin notes from research published after the book. Read it as a summary. Or as a design review of the environment we built and then moved into.

Carr’s map Link to heading

The mechanism is neuroplasticity. The brain changes with use. Repeated behaviors get faster, more automatic. What starts as choice becomes default 1 .

That sounds motivational-poster-ish until you apply it to attention.

Carr treats the internet as an “intellectual technology,” not a channel that delivers content, but a system that shapes the posture you hold while engaging with it. The web is built for speed, branching, interruption: links, tabs, feeds, alerts, infinite scroll, and the quiet implication that something better is one click away 1 .

Even when you’re reading, you’re navigating.

Deep reading is different. Sustained attention, working memory, inference, integration. You carry earlier pages forward, build a model of what the author is doing, keep enough context in RAM to notice when something doesn’t fit 1 . That’s a skill you build. It’s also a skill you can starve.

Margin note: Adam Garfinkle calls this “deep literacy,” an extended dialectic with an author, where you anticipate, interpret, and produce new insights rather than collect facts 3 . Useful framing. It stops us from treating “deep” as a personality trait.

Carr’s fear is not that we’ll forget trivia. His fear is that we’ll lose the capacity to form and hold complex internal representations, because the environment keeps pulling us into short, discontinuous bursts 1 .

That’s not nostalgia for books and quiet rooms. It’s a question about default cognitive style: what kind of attention does your day rehearse?

What the research says Link to heading

Carr wrote like a columnist with a stack of studies on his desk. Since then, the stacks got taller. We still don’t have a single clean “internet causes X” result; cognition rarely cooperates that way. But the patterns fit his model.

Screen reading is not paper reading. A meta-analysis by Delgado et al. found a small but consistent comprehension advantage for paper, especially for expository texts under time pressure 4 . Clinton’s review adds a design-relevant detail: people are worse at judging how well they understood when reading on screens 5 . If you don’t realize you understood less, you don’t slow down. You walk away with an inflated sense of “done.”

Small effect sizes are not comforting. Small effects, multiplied across millions of sessions, become the background radiation of a culture.

Interruptions are not a side effect. They’re a feature. Stothart et al. found that receiving phone notifications, without interacting with them, disrupted performance on attention-demanding tasks 7 . Your brain pays a tax just from the knock on the door, even if you don’t open it. Mark et al. found that people compensate for interruptions by working faster, without a quality drop, but with increased stress, frustration, and effort 8 . The system can look productive while quietly frying the operator.

Ward et al. go further: the mere presence of your smartphone reduces available cognitive capacity, even when it’s face down and you’re not using it 6 . The device becomes a background process. You may never open the app, but the OS is still running.

Cognitive offloading changes what memory does. Sparrow et al. tested what happens when people expect access to information later: they remember the information less well and remember where to find it more well 9 . Transactive memory, with search engines as the external partner.

This is not automatically bad. Offloading can be smart allocation. The problem is when retrieval becomes default. You can fetch a fact without building a mental model that explains it.

Margin note: Polanyi’s line “we know more than we can tell” is useful here. Expertise is tacit, pattern recognition, not explicit recall. Deep reading feeds that layer. Pure retrieval does not.

Everything is skimmable now. But what’s worth stopping for?

Where Carr overreaches Link to heading

Carr sometimes writes as if “the internet” is one technology doing one thing to everyone’s mind. Clean prose, messy science.

Effects differ by genre: narrative and expository texts behave differently 4 . Effects differ by context: time pressure matters 4 . Effects differ by measurement: “attention” is not one thing but an ecosystem of skills and states. Parry and le Roux’s meta-analysis on media multitasking is not “multitasking melts your prefrontal cortex.” It’s more cautious: effects are small, measurement is inconsistent, causality is unsorted 10 .

That caution is an upgrade. Designers should prefer models that match reality: multiple pathways, context sensitivity, feedback loops.

Carr also underrates skimming. Skimming is not a disease. It’s a navigation skill. You need it to survey a landscape, identify trails, avoid irrelevant detail.

The failure mode is not skimming. The failure mode is losing the ability to shift gears.

If the environment keeps forcing skim mode, you stop treating it as a tool and start treating it as the only mode you have.

The hermeneutic angle Link to heading

If you want a deeper frame, hermeneutics is oddly practical.

Gadamer’s tradition treats understanding as interpretation. You don’t download meaning; you meet it. You carry prior assumptions, the text pushes back, and through that loop your horizon shifts. Less like a data pipeline. More like an iterative design review.

Deep reading has a spiral shape. You return to the same idea at higher resolution. You notice what you missed. You revise your model. You integrate.

Skim mode breaks the loop. It keeps you in first-pass interpretation forever.

Most digital environments are not designed for continuity. They are designed for throughput.

What this means for builders Link to heading

If you build digital products, you are in the attention business whether you admit it or not. You don’t just deliver content. You train the reader’s posture toward content.

That training can be accidental, “we just followed best practices.” It can be deliberate. Either way, the system has an effect.

Treat attention as a system. People respond to signals, rewards, friction, and delays. If the signal is “new thing, new thing, new thing,” you get switching. If the reward is fast novelty, you get novelty-seeking. You don’t have to preach about focus. You can change the instrumentation.

Build mode shifts. Skimming is a mode. Deep reading is a mode. Don’t force users to do deep reading inside a skim-first interface. If the experience includes long-form material, give it an environment that signals continuity: stable typography, fewer branching cues, fewer temptations to exit.

Design for calibration, not just consumption. If people misjudge comprehension on screens 5 , add supports that help them detect “I didn’t actually get this.” Self-check prompts at natural boundaries. Summaries the user generates, not just receives. The point is not to slow them down for fun. The point is to restore accurate feedback.

Treat interruptions as a cost center. Your notification system is a lever. It can help users coordinate with other humans, or it can create persistent partial attention. If your product adds interruptions, be honest about what you’re billing the user.

For AI features: be careful what you automate. If a tool does the interpretive work for the user, you might improve short-term output while degrading long-term competence. That’s not an argument against assistance. It’s an argument for preserving the loop where the user makes meaning. Remove engagement, and you get fluent users who can’t explain what they’re doing, and can’t notice when the system is wrong.

Margin note: Stafford Beer’s law of requisite variety is a useful gut-check. If the environment throws high variety at a user, the user needs ways to manage that variety. Otherwise, they collapse into shallow heuristics. Most feeds increase variety while removing the user’s control over pacing.

What I take from Carr Link to heading

His book is not a perfect model of cognition. It’s a field guide written by someone watching the weather change.

The big thing he got right is the framing: attention is trained by environments. The unit is not the individual mind. The unit is the person plus the system they operate inside 1 .

The big thing he sometimes got wrong is the implied uniformity: one web, one outcome. Reality is patchier 10 . But patchy does not mean harmless. It means design matters.

If you build tools, you’re not just making features. You’re laying trails. People walk the trails you make easiest. After enough repetition, the trail becomes the landscape.

Everything is skimmable now.

This piece was designed to be read. To stop for. Whether it earned that is not my call.

But here’s what I know: you made it here. That means something. Most didn’t. And the trail you walked is the one you’ll walk again.

Choose your trails.