A new form of of SEO for feeds

Or: why your content library is currently failing you

As any creator will tell you, unless you’re already very successful, shouting your work into the void of your distribution channels (i.e. Twitter, LinkedIn, TikTok) isn’t going to net you much of anything. But as I mentioned last time, feedback, especially from the people you’re trying to reach, is a crucial ingredient to producing good content. So how do you get the Feedback Loop (in blue) spinning in the first place? And once it’s going, how do you keep the momentum?

The top substack writers without pre-existing fame provide useful data. These people have one thing in common: they are absolute beasts at Twitter and LinkedIn. Take Fabricated Knowledge, a newsletter about the semiconductor industry. The Fabricated Knowledge Twitter account tweets about 10 times a day: a mix of promotion of his own content, retweets of related news, replies in relevant conversations, and tidbits about what he’s working on and what’s in the news.

Liberty’s Highlights, another business-focused newsletter, whose anonymous writer describes it as ‘an outlet for sharing interesting things I find’, is similarly prolific on Twitter. Liberty is particularly good at engaging with other people about their shared work: here’s Liberty engaging with Byrne Hobart, a writer with an order of magnitude more paid subscribers, about FTX’s collapse, a topic that they’ve both been writing about recently.

This is the rule, not the exception, and it cuts across distribution platforms. As platforms have moved further toward engagement-based, video-heavy ranking (and away from a follow-based model), the lifecycle of any given post has gotten shorter and the need to post more frequently has increased.

And it’s not just about volume. Most of Liberty and Fabricated’s tweets are replies, @s (where they tag another account in a post) and retweets. In other words, they’re conversations with other users, many of whom are also writers and creators. Given the shift away from a follow-based model, this makes a ton of sense: if turning a viewer into a follower isn’t going to help you reach them again, your best bet is to engage in the conversations that your potential fans are most likely to see.

This is a kind of wrinkle of the Feedback loop: while your audience itself can provide valuable feedback, sometimes the best feedback—and exposure—comes by way of the conversations you participate in.

Some people, like Fabricated and Liberty have a knack for this. But I’ve also met a lot of creators of really interesting, worthwhile content for whom the promotion-and-networking process is at best, a distraction from where they want to spend their time. At worst, it’s an insurmountable hurdle to finding and, in a world of increasingly implicit ranking, keeping their audience.

So what can be done about it?

1. AI + content libraries could solve the post volume problem.

Firstly, I think most creators’ existing content libraries could work a lot harder for them. Artists and visual bloggers are further along on this front. Smitten Kitchen, a food blog I’ve followed for over a decade now (!), writes a new post on her blog every couple weeks, but posts a recipe from her library on Instagram once or twice a day. These posts, which are seasonal and often relate to current trends on food Instagram, don’t require any effort other than the image selection and the one-sentence caption.

For audio and video creators, the calculus is different. On one hand, a long-form video can provide more (and more diverse) snippets than a single recipe. Acquired, a US top 10 technology podcast, produces at least one TikTok video per day, most of which are focused on their most recent episode. On the other hand, it’s MUCH harder to produce the assets: while images are ready-made for sharing to fast-scrolling feeds, an hourlong video podcast is decidedly not.

This is one of the selling points for my old employer, Descript: quickly transcribe your video, cut your favorite snippets, and put snazzy templates around them before sharing on social media. Descript assumes that you will manually identify the moments you want to share, but other products are starting to make this selection process easier too. Munch and Vidyo, for example, both use some mixture of AI and heuristics to recommend the best clips to share.

Having played with OpenAI’s ChatGPT chatbot on a few creative partners’ content lately, I can say that the model’s ability to pull out interesting moments is impressive, even with little tuning. As an example, here’s what the model returned when I told it to “Find the best 5 quotes containing approximately 150 words from the text below:” in the transcript of one of my favorite This American Life episodes of all time: the one about a community theater fiasco

"And in this particular production, there's a big bed with all the three kids in it. And off to the left, I remember, is a big, huge wardrobe. And there's a large window there, and a little bureau. And Peter Pan comes in and has the little speech where he says, 'Anybody can fly. Why, with just a little magic dust, one can fly.'"

"And sure enough, the kids suddenly just lurch into the air. And it becomes clear right away that the people that they've hired to run these flying apparatuses really aren't quite clear on how they actually work. So instead of the kids sailing gracefully to and fro, they hang in the air like puppets, just dangling there and getting jerked up an inch or two or back and forth."

"No one is laughing. This is one of the great things about audiences, especially in a live theater production, is that they're very forgiving. They want the show to work. And so everyone is gripping their chair a little tightly. We feel for them. They're up there. They're embarrassing themselves for us."

"And then several of them start to circumscribe these circles in the air, where it's clear that the people running the machines have just set them off on these oval courses that spiral farther and farther out. And if you're sitting in the audience, there was clearly a sense of fear on the faces of these people."

"And so he's flying around in this circle. And the audience sees this coming. And there's a real sense of pain and gripping of the chair and white knuckle-ness as the kid suddenly does just an enormous splat into the wardrobe. And it's clear that he's hurt, and he comes off of it a little dazed. Then, of course, he's jerked up in the air a little bit, and often a little too high, so that he's suddenly in the workings. He left the stage itself. He's now up there with the lights. Then, all of a sudden he would just plummet back down to the stage and be caught up just before he hit the floor."

It’s not perfect. The first quote is just context setting. And I’ve found there’s a great deal of variability across different content inputs and prompts. It seems to work best for publicly available content (This American Life puts its transcripts online), even when I provide the text in the prompt.

But, having listened to this particular recording multiple times over the years, I can tell you that these quotes do represent some of the funnier anecdotes.

I think there’s room to make this whole process easier, from clip identification to the production of TikTok/Twitter/etc-ready snippets. But what’s missing from all of these media production solutions is context. As covered above, shouting your work into the void of social media is a low-efficacy strategy, even if you have endless, good-quality clips to shout. Instead, you need to find ways to plug into the right conversations with the right people.

2. Automations can help find and track the worthwhile conversations.

On this dimension, too, I think machine learning can help. The promotion-and-networking process is a protocol that goes something like this:

1. First keep a constant pulse on:

2. The other people who are interested in similar topics to what you’re currently covering (and stuff in your library).

   • The trends and news that matter to your audience and potential audience and the people who are interested in those topics.

3. Then make a guess about whether they (or their followers) might actually engage with you, based on their following, their work, your mutual connections, what they’re discussing right now, and what you have to say about the topic.

4. Then form coherent additions to the conversation.

5. Repeat.

Tools like Tweethunter are starting to get at this problem by providing ‘AI-selected inspiration’ in the form of popular tweets from others that align with stuff I’ve tweeted about in the past. And more traditional cross-channel CRMs like Hubspot have long enabled customers to track organic engagement with their own posts on social media. But the real power here, for a content creator, comes at the intersection of the content itself and what people are talking about right now.

Taken a step further, it’s not hard to imagine a world where, given access to both the content library and the distribution channels, the clips and the conversation suggestions are self-tuning:

“...this week, influential members of your target audience are talking about Twitter’s business model. Maybe you should promote this clip from an old podcast episode where you suggest ways for Twitter to move beyond ads?...”

- Content engine optimization bot of the future

Something like this could be useful for anyone with a content library; not just creators. I worked on ad products at Twitter, and I’m curious about the potential to apply the same techniques to video ad campaigns. If you could automatically find the right moments in your existing library of video content based on the current social context, and if your video creatives automatically adjusted themselves based on these findings, would that produce better results? That’s a topic for another post.

But back to the organic conversations. I was catching up with my friend Xiaolei about all this recently, and he had a metaphor that stuck with me: in the old world, before social media became practically synonymous with the internet for a huge swath of people, the only ways to get your blog or video noticed were: a) for another creator or publication to drive traffic to you, b) to buy ads and c) to rank highly in search engines. An entire multi-billion dollar industry developed around search engine optimization strategies and tooling. And while the tactics have changed, SEO remains one of the best ways to get found online, especially for businesses. But ephemeral content engines, like Twitter and TikTok, require a different optimization strategy, one that is always-on, context-aware, conversational, and optimized for audio and video. And I think there’s a lot of headroom for something that does all four well.