I caught myself writing like Claude last week.
I was writing out the diagnosis of why a complicated situation was getting more complicated, and started a bullet point with “The real culprit: [pithy mic-drop explainer].”
This came on the heels of another fail in which I’d asked my handy desktop AI tool to transcribe a recording, but didn’t notice it truncated each segment. When I gave the results a quick glance-over, they seemed legit. Then I asked the tool to generate a summary, which I also reviewed in passing (“Yep! Good enough!”) and distributed accordingly. It was an accurate high-level summary, but rather milquetoast and skipped over details that were kinda important.
A colleague rightfully called my attention to the slop, which was really embarrassing considering what I write about here.
Not that it’s an excuse, but I’d just spent the weekend doing coursework for an AI class I’m taking. We’re learning how to set up and run an agentic operating system, which is cool in concept and WAY complicated in reality. Claude, myself, and my chief-of-staff agent Willow spent the better part of two days diagnosing all manner of technical glitches, which included a lot of “Approve once?” gatechecks that eventually all blurred together.
I didn’t hit the YOLO switch, but I might as well have for all the thought I was putting into it. Decision fatigue leads to surface-level assessments, which leads to low-quality outputs, intentionally or not. Add time or productivity pressure into the mix, and it gets worse.
The good and bad news for writers and creatives like myself is that, after a scary stretch of “welp, I guess it’s time to find a new career,” audiences everywhere revolted against the steady diet of AI-generated pablum being thrown at them.
The good part: Authentic human creations are once again prized.
The bad part: Everyone still expects you to create at a lightning speed that pretty much requires you to use AI on the non-authentic parts. Research. Business processes. Multi-channel management. That operating system I’m building.
And then you start to internalize the Claudeisms and skip over the details, diminishing your authenticity.
Damned if you do, damned if you don’t
A variant of this dilemma is occurring among web publishers, who’ve been inundated with slop posts, articles, videos, comments, DMs, and quote-on-a-wall genericisms. They did this to themselves by cramming in AI everywhere to boost engagement. But now, users are fed up, and the ban hammer’s swinging. A few recent examples:
- YouTube tightened up its definition of “inauthentic content” to stop folks from monetizing their AI-powered content farms and videos of not-a-real-person experts.
- Substack introduced a tool readers can use to measure how much of a page’s content was AI. TBD on how well this will work. It’s banking on an AI tool to tell you what is and isn’t AI, sort of like a snake eating its tail.
- LinkedIn implemented an algorithm that penalizes the visibility of posts deemed to be low-quality AI filler. Last week, they also introduced a button that allows you to flag slop that slipped through, and they ditched the little sparkle icon in the authoring interface.
You have to admit, it was a long time coming.
Yet when you visit LinkedIn, there are still AI tools among the job boards, offering to rewrite resumes and profiles. Substack was also quick to note that their intent is for writers to voluntarily disclose their process, and not to punish AI-generated content. But you can bet false positives will end up doing unjust financial or reputational damage to an as-yet-unknown number of creators. Same with LinkedIn. Imagine the onslaught anyone tagged as a LinkedIn Lunatic is about to face.
To recap … Using AI during the creation process in an obvious way is bad, you’ll be punished, and people will shout Hallelujah from the rooftops over it. But even if you don’t use AI at all yourself, you’re reliant on AI-powered algorithms to make sure your authentic content isn’t getting buried or associated with the inauthentic stuff, which could still punish you.
Choices, choices
It’d be a whole lot easier if this were an all-or-nothing decision. Purists, for example, might win the moral high ground, but business-wise, they can’t keep pace with competitors who’ve integrated AI in ways that boost their productivity. On the other side of the spectrum, the AI Everything! folks are likely to churn out slop at some point (if they’re not already) and will face the consequences.
What it boils down to, then is knowing what, and what not, to use AI for. These are straight-up judgment calls. Personal values, your purpose in creating things, your audience’s tolerance, and the rules your distribution path sets will all play a role.
Let’s walk through a few scenarios:
Situation: Your target audience is 100% anti-AI and will shun you if you use it in any way, shape, or form. For the creative process itself, the choice is obvious. No AI. Where it gets murkier is in the before-and-after parts, especially with AI getting forced into more tools. Case in point: When you’re researching, it’s getting harder to avoid AI in even basic Google searches. Going fully offline is an option, but it could impact your bottom line if you’re releasing work less frequently than others. For artists, this won’t be an issue. For personal-brand-builders, a “100% human” label can be a powerful selling point but, depending on your niche, could also be a disadvantage in the visibility economy where frequent presence makes a difference.
Situation: You distribute primarily through social media channels, and repurpose content to connect with different audiences in different ways. The anti-slop crackdowns mean you’ll need to be extra mindful of the quality of your work. If you’re see-and-sending AI-generated content variants (text, video, imagery, music), that’s not the path to success. But using AI for video touchups, extraction and reformatting, proofreading, and similar yeomanlike tasks could speed up your workflow. The key is making sure your outputs are meaningful and not simply well-polished. And yeah, this is easier said than done.
Situation: You’re creating a presentation deck for an important work meeting. This scenario is likely to be more tolerant of AI-generated content than others. In fact, I suspect AI-generated starter drafts will become the norm for these documents, with your team’s branding and guidelines baked in. Your biggest decisions, then, will be in how you describe the concepts you want to present, and what context you provide so the AI tool can do them justice. Business slop is still slop, so don’t do a see-and-send on this stuff either (like my failed transcript notes!). Review the output and revise the parts AI got wrong or that call attention to themselves as non-meaningful. If the whole deck is a disaster, then rethink your prompt and context files.
These are all content-creation scenarios, but the same rule of thumb applies to any decision about using AI: It’s critical to know what, and what not, to use it for. And — coming full circle — this is the operating system part. You’ll have to experiment with what does and doesn’t work, but once you have the system in place, the laundry list of judgment-call decisions will start to get easier.
While the real you still shines through.
All opinions here are my own. All text is my own, too, including the em dashes. I welcome constructive comments and discussion on LinkedIn and Bluesky.
Leave a Reply