Someone asked me a question recently:
What is the difference between AI slop and a hallucination?
It sounds like a word game. It is not. I think the difference between these two words explains most of what goes wrong when companies start using AI. So here is the short version.
A hallucination is when AI is wrong. Slop is when AI is right, but bad.
Let me take them one at a time.
Hallucination: confidently wrong
A hallucination is when AI states something false as if it were true. A statistic that does not exist. A case study that never happened. A quote nobody said. The AI is not lying on purpose. It is making a confident guess, and the guess is wrong.
Hallucinations sound scary, and they can be. To be honest: they are the easier problem.
Why? Because a hallucination fails a fact-check. If you read the output and ask "is this true?", you catch it. It takes discipline, but the test is simple. True or not true. Most teams I meet have already learned this. They check the facts before anything ships. Good.
Slop: correct, and worthless
Slop is different. Slop passes the fact-check.
Every sentence is true. The grammar is clean. The structure is fine. And the whole thing says nothing. It is the blog post that could have been written about any company in your industry. The LinkedIn post you scroll past without noticing you read it. Content that is right about everything and interesting about nothing.
That is what makes slop the more dangerous problem. There is no simple test that catches it. You cannot ask "is this true?" because it is true. You have to ask a harder question: "is this any good?" And that question needs a human with taste and a point of view.
I wrote before that correct is not the same as distinctive. Slop is what happens when you settle for correct.
Why this difference matters
I keep seeing a pattern: Teams build guardrails for hallucinations. Review steps, fact-checks, source requirements. All good.
Almost nobody builds guardrails for slop.
So the wrong output gets caught, and the bad output ships. Then six months later someone asks why the content is not working, even though "we checked everything."
You checked whether it was true. You did not check whether it was worth reading.
And right now the pendulum makes this harder to see. The same voices who were selling custom GPTs for writing LinkedIn posts a year ago now say they write everything by hand, because readers can tell. Maybe. I am not fully convinced. But notice what both ends of that swing have in common: they treat AI or human as the question. It never was. The question is which failure you are guarding against, and who decides what good looks like.
The taste guardrail
The fix is not technical. It is a second question in your review, next to the fact-check. Before anything AI-assisted goes out, I ask:
- Does this say anything? If you removed your logo, could this be from any company in your field? If yes, it is slop.
- Would I say this? Not "is it acceptable." Would I, personally, stand behind this sentence in a room?
- Is there one thing here the reader will remember? One idea, one image, one honest line. Slop has zero.
None of this means avoiding AI. I use AI in almost everything I make. Even the images I use are AI generated, and for my brand that fits. I write about AI. For another brand, an AI face in a photo might be exactly the wrong move. Both choices can be right. What is not right is not choosing.
Because taste is not universal. It belongs to your brand. The real guardrail is that your brand knows what is okay for your brand, and it is written down and shared with the people in the organisation. What can be AI and what can never be. Who gives the human read before something ships. What you tell your customers about how things were made. Where nobody has decided what good looks like, slop walks straight through.
So know which failure you are checking for. The fact-check catches hallucinations. Only a human read catches slop.
A hallucination is when AI is wrong. Slop is when it is right, but bad. Wrong gets caught. Bad gets published. That is why the second one should worry you more.