Let’s kill the comfortable myth first: you cannot reliably spot AI-generated text by style. The em dash counting, the hunt for the word “delve,” the online detectors with their confident percentages. None of it holds up, because well-directed AI writing reads fine and plenty of human copy reads like it was written by a fax machine. Anyone selling you a detector is selling you confidence, not accuracy.
But here’s what your customers can detect, without tools, often without consciously noticing they’re doing it: content that nobody was accountable for. That’s what slop actually is. Not machine-made. Unowned. And unowned content has tells that have nothing to do with word choice.
So instead of playing detective with sentence style, audit your own feed with five questions. Fair warning: we ran this on our own old content while writing this post and did not enjoy the experience. It works.
1. Could a competitor have posted this word for word?
Pull up your last ten posts. Swap your logo for your closest competitor’s. Does anything break?
Slop is interchangeable by nature, because it’s assembled from the general instead of the specific. If your content survives the logo swap, it isn’t saying anything only you could say, and your audience files it under background noise whether a human wrote it or not.
The fix: every piece should contain at least one thing that’s yours alone. A real customer moment, an actual number from your business, an opinion someone on your team would defend at a dinner party.
2. Is there a verifiable specific anywhere in it?
Slop speaks in confident generalities. “Studies show.” “Experts agree.” “In today’s fast-paced digital landscape,” a phrase that has appeared in marketing copy every year since roughly the invention of the fax machine, which we’ve now mentioned twice, so we’ll stop.
Real content has handles you can grab: a named source, a dated event, a number that traces back to something.
This one matters beyond taste. AI fabricates specifics fluently, so an unverified specific is worse than none at all. The question isn’t “does this sound authoritative.” It’s “could I show a customer where this came from without sweating.”
The fix: one enforceable rule. No claim without a source someone on your team has actually seen. If you can’t source it, cut it, or soften it into honest opinion. “We think” is an underrated phrase.
3. Would you say this sentence to a customer’s face?
Read your captions out loud. Actually out loud, not in your head where everything sounds fine. Slop has a particular weightlessness: sentences that are grammatically perfect and socially impossible. No human being has ever said “we’re thrilled to unveil a game-changing solution that empowers your journey” to another human being, and if one did, the other one left.
Your customers hear the difference because they spend all day talking to actual people.
The fix: the out-loud test before anything publishes. If it makes you cringe on your own behalf, it doesn’t go out, no matter who or what drafted it.
4. Do your images show anything real?
Generic AI imagery is the most visible slop there is, and it does damage way out of proportion to how easy it was to make. Stock humans with creative anatomy and products that ignore physics tell your audience one thing: nobody here cared enough to photograph the real thing.
There are legitimate uses for generated imagery. Passing it off as your actual product, team, or shop is not one of them.
The fix: real photos of real things beat polished renders of fake ones, every single time. Even shot on a phone. Especially shot on a phone, honestly. Slightly imperfect is what real looks like.
5. Did a named person approve it?
The root cause behind every failure above is the same one: no individual human was on the hook. Content pipelines where posts flow from prompt to publish without a person whose name is attached to the approval, that’s the machine that manufactures slop. And it manufactures it whether or not AI is anywhere in the building.
The fix: one rule that quietly fixes most of this post. Nothing publishes without a named person’s sign-off. Not “the team.” A name. Accountability is the only slop detector that actually works.
The uncomfortable summary
Notice that none of the five checks asks “was AI involved?” That’s on purpose. AI-assisted content with a human owner passes all five. Fully human content churned out carelessly fails them. The problem was never the tool. It’s whether anyone stood behind the work.
We wrote about where we draw that line in our own shop in What We Let AI Do, and the thinking behind it lives in The Credible Middle. Sorting your own workload the same way takes about ten minutes with a coffee, which is less time than you’ll spend regretting your old captions. Speaking from experience.
Run the five questions on your last month of content. If it passes, your process is fine, whatever tools are in it. If it doesn’t, congratulations: you found the problem before your customers had to tell you.