Attention Spans

Are you the sort of person who loves stories that start with a question? Or maybe you prefer being intrigued by more mysterious openings that promise deep insights if only you keep reading.

Did you like that lede? I wrote it just for you.

Well, not you, personally. You, the broader pool of humans I hope are reading this. You, the humans who I hope will get sucked in by my magnetic writing and read all the way to the end.

But what if, in a near-future world, I’m not really writing for you? What if, in fact, this substack is consumed more by machines than by humans; by agents that will ingest, digest, extract and reformulate for you (the humans) — just the parts that matter to you, in just the style you want it. Does it really matter then whether the top of my piece entices you to read, or whether my sparkling wit makes you chuckle and keep going to the end? How badly do I need a kicker to close out the post to make my points stick?

All these thoughts — and more — came to me as I was reading a post by tech entrepreneur Rami Alhamad about how he set Claude off to ingest all 243 videos posted by the World Economic Forum at Davos and give him a 1,000-word deep dive and synthesis of the core ideas, positions, predictions and disagreements at the annual Masters of the Universe meeting. (Which may or may not have real value; the consensus is that the Davos consensus is almost always completely wrong.)

It’s a fascinating post about how information consumption changes; we’ve talked about this, and my colleague Adiel Kaplan has written about how she’s built a newsletter that consumes newsletters to give us a summary of what we care about. So the idea that machines will consume, and intermediate information, isn’t new.

But what’s a new thought — at least to me — is what that does to the battle for attention.

And how we — not machines — will be adapting to it.

Rami writes:

I could have watched all 243 videos. They’re free. They’re on YouTube. The constraint was never access, it was attention.

AI doesn’t give me more hours. It gives me leverage on the hours I have. The 30 minutes I spent on this experiment yielded more insight than 30 minutes of watching random sessions would have.

Sure, he didn’t get the full Davos experience — such of it as there is — and perhaps the machines may have missed some nuance. But arguably, he got something better: A clean, well-structured digest of information, containing not just summaries, but also points of disagreements, key positions, predictions and their probabilities. In a package he actually has time to ingest, without needing to drown in 60 hours of video. In effect, he traded the depth of experience, of listening to every word, for the breadth of understanding; and in his mind, that was a great bargain.

And that’s the trade all of us writers — creators — will be engaged in once AI agents become ubiquitous. Yes, we’ll still be fighting for your attention. But the battleground will shift. Right now I try to craft engaging ledes that will hook you and draw you in. I focus on making sure I have a narrative thread to keep you engaged and enough humor to leaven the process. I throw in a kicker that circles back to the lede to make my main point stick.

That’s a human-reading-centric strategy. But a machine doesn’t need to be hooked or intrigued. It’s going to read my post regardless, because, well, it’s been told to. It doesn’t need a kicker; it’ll remember what I’ve written whether I drop in a memorable line at the end or not.

In this world, what matters is clear structure and a well laid out argument, vivid examples, smart framing and compelling data to back up my points. That’s what the machine will see, and — assuming it’s something its reader will care about — extract and summarize.

That requires good writing, too; but a different kind of good writing.

And what that means is — just as machines are adapting to understanding our world so they can navigate it well, we’ll be simultaneously adapting to their world, and learning how to navigate how they see information.

If that sounds uncomfortable, well, it is.

What matters more in this world is less the ability to catch your eye and more finding ways to get on your list of sites for your agent to trawl, and to ensure my arguments are clear enough to make it to whatever summary you’ve instructed the system to look for. If what matters to me is getting my arguments or ideas in front of you, then how I make that clear to your agent, rather than you directly, becomes my north star.

(And yes, I realize this raises questions about how I might be compensated for my writing, if it simply becomes grist for AI-generated summaries; it’s a real question, but that’s a subject for another post.)

So far, I’ve been focused on more technical issues of how information needs to be structured for an AI-intermediated world — better metadata, more background information, bullet-pointed structure, and so on. But I’m realizing, belatedly, that this shift will also require us to reframe how we think about how we write.

To be sure, there will be people — hopefully me, too! — who can find an audience precisely for my prose, as well as for my insights and ideas and thinking. Those of us who can live and thrive in what Tony Haile calls the “archipelagos of trust,” the niche pockets where we’re valued for high-end smart writing, reporting, analysis or aggregation and where readers want to savor our words, un-intermediated. For the people who want the depth of experience rather than the breadth of information.

But those will be exceptions. In a world of ever-increasing information, we’ll turn more and more to systems that can process all that writing (and video, and audio) for us. And while we’ll want to make it legible to humans who may want to come to us directly, our priority will be worrying about how well machines understand our writing rather than what our fellow humans may prefer.

I’d put in a pithy kicker here, but I want the bots to like this piece, so I’ll just leave it as it is.