AI-Generated Content is the New Floor

Anyone can make it. Most of it is at least OK. Some of it’s pretty good. A few folks are already sharpening their skills at prompting AIs to generate increasingly high-quality material. But, the core problem with content created by large language model (LLM) outputs is that it’s so cheap, so accessible, so ubiquitous—it’s not merely table stakes; AI content is the new floor.

In 2015, I made a very popular video about “10X Content,” i.e. content that was ten times better than anything else on the front page of Google. Moz republished it a few years after I left in their “best of the blog,” category, so I assume it maintained its popularity (despite my questionable hair and mustache). I included a large list of 10X web content: 121 examples to be precise. Most of those are still “better” than something ChatGPT4 or Google Bard could create today. Given a few more years… who knows?

What we can say for certain is this:

If your content isn’t better* than what AI can produce, it’s not worth making.

This wasn’t always the case. In the relatively short history of the web, the minimum bar for content that’s worth creating, publishing, and distributing has only ever gone up. But that history is filled with years of slight, incremental increases and ones with massive, step-function changes, e.g. Google’s launch, the emergence of professional SEOs, the emergence of social networks and the content mills that (briefly) fed them, etc.

ChatGPT3, 4, and the inevitable 5-X are harbingers of another step-function change. Whether that’s bigger than what happened in the 2001-3 era (as Google emerged as the default search engine) or the 2017-2019 era (when the Big Tech players moved fully into their zero-click paradigm) remains to be seen.

For now, the LLM AIs have limitations on what they can do well.

Via OpenAI’s FAQs about ChatGPT, which far too few people have read

For example, ChatGPT doesn’t know anything that happened after 2021, doesn’t cite its sources, and regularly makes up facts and information with no warning about its degree of (in)accuracy. That’s because it’s not a large fact and logic model. It’s a large language model. It’s trained to predict and output words people have previously written after other words, and it’s so exceptionally good at doing so that many users falsely assume it has near-human-like intelligence.

To know how to do better than AIs, human creators need to first understand how LLMs work. There’s no better primer, in my opinion, than Stephen Wolfram’s immensely detailed What is ChatGPT doing and why does it work? But, it’s a lengthy, technical, at times overly complicated piece. I’d probably suggest AssemblyAI’s 6-minute long How ChatGPT Actually Works to get a more accessible, faster sense of what these models do.

In essence, you’re competing with what we at SparkToro have been calling “spicy autocomplete.”

Via Wolfram’s ChatGPT post

Large language model AIs have been trained on, you guessed it, large amounts of language. Ginormous would probably be a better word choice here, as we’re talking about hundreds of billions of documents. Everything that journalists, content creators, and marketers have ever published on the web, everything users have ever contributed to Reddit, Wikipedia, Twitter, Google News, and anything else a web crawler can reach is fair game.

Yes, ChatGPT is using your own content to compete with you.

Moaning about it probably won’t help. Unless you’re Italian. Creating things other humans who use ChatGPT prompts can’t (or won’t) create is the only path forward. In my experience, the three biggest advantages human creators have over AIs (for now) are:

  1. Emotion – ChatGPT can’t be vulnerable. It isn’t scared. It feels no empathy, nor can it convey true regret. It isn’t humble or prideful, distraught or loving. When you prompt it to communicate using these emotions (e.g. “Say that again, but more empathetically?”), the results feel inauthentic. Spicy autocomplete almost never elicits the emotional weight that good, human writers can.
  2. Novelty – If an idea, a bit of data, a data source, an amalgamation of information, or an event didn’t exist before 2021, ChatGPT isn’t going to produce content about it. Technically, it can create new works, but these will always be derivative. If you ask ChatGPT what to write about to please an audience of X, it can only tell you what they might have cared about in the past.
  3. Creative Insight – After reading the output of a LLM AI, you will almost never hear someone exclaim “Oh my god… that’s a great point!” or “Whoa… I’ve never thought of it that way.” Nor will you see the AIs get artistic or inspirationally motive with their replies. “Oooo… I just thought of a great way to visualize that,” or “I bet we could make a really cool video game based on that premise,” aren’t responses the machines can compete with (yet).

Via Jerome Choo on Mastodon

There are emerging plugins promising to ameliorate some of ChatGPT’s limitations. Last week, for example, OpenAI announced a slew of new offerings that enable ChatGPT to directly book flights for a user on Expedia or make a dinner reservation on OpenTable. Plugins that scrape the news and can summarize current events, those that mine social posts, access your email, or send phishing scams on LinkedIn are probably not far off.

But even in the extreme cases, I’m not seeing LLM AIs that can do those three things above better than a human. There are probably other advantages flesh-and-blood types still have, too. And if you’re creating content, you’d better leverage what you can do better than the Terminator’s great-grandparents. There’s literally no point in creating something that’s on par with what $0.000004/hour, 0-hours of rest-requiring machine can do in milliseconds.

E.g. thank goodness I can do better than this:

Side note: has my writing ever been that tediously generic? I’m slightly offended. (And before you ask, yes, immediately following this, I did try a number of secondary prompts to improve the output, but thankfully got back equally “meh” responses)

The floor for content has never been higher. But the ceiling is still higher than most of us can imagine.

The only good news is, that’s always been true.

* “Better” is a complicated word when it comes to content, but I plan to write about the many ways content can be better to its intended audience(s) in an upcoming post, so stay tuned.