Is Your AI Slop Showing?

My LinkedIn feed has been full of commentary on “AI slop.” When someone takes an AI response, skips the review, copies and pastes, and puts their name on it, voilà: AI SLOP. Using AI shouldn’t be a secret or something to be ashamed of. Not using your judgment and review skills is. Whether you announce that you used AI or keep it to yourself, you’re responsible for what you share. You should have read it, questioned it, and decided that it actually represents what you think.

I use AI to talk through my thoughts, ask questions, and help me organize what I’m trying to say. Sometimes I’m still figuring out what I think as I’m talking, and having something that reflects those thoughts back to me is useful, but I still scrutinize what it gives me. I’m not going to pretend like what my robot (shoutout Maverick and Claude!) has written is me just because it sounds good. If we want less AI slop, we have to stay involved in what we’re creating.

Here are my best tips:

1. Give it your thoughts and a clear assignment.

Tell it what you actually think and why, including your experience, the relevant background, and anything you’re still considering. You don’t have to arrive with a perfectly organized idea. I’m wordy when I’m thinking out loud, and sometimes I change my mind as I’m explaining something. That’s part of working through it. Give it direction, too. Who is this for? What should they understand? What information should it use? Treat it like an intern you’re assigning work to. You wouldn’t expect them to know everything you know about your business or your audience without explaining it.

2. Review what it means.

I am not gentle with my AI’s work. I question whether I agree with the answer, whether it’s making assumptions, and whether it actually addressed what I asked. I also check whether it made me sound more certain than I am. If I say we’re considering a new service and the draft says we’re introducing one, it has changed the meaning, even if the sentence sounds perfectly professional. Tell it specifically what isn’t working, then read the revision, too. Asking for a correction doesn’t mean the correction was made. A well-written version of something I didn’t mean still needs to be corrected.

3. If it doesn’t sound like you, don’t copy and paste it.

There are things I read and immediately think, “I would never say that.” I also get annoyed by some of those familiar AI phrases. Since when did everyone start saying “hype and fluff” all the time? Those words alone don’t make something bad, but I notice when a draft sounds nothing like the person supposedly speaking. Explain what sounds off, change the language, and read it out loud if that helps. You should recognize your own perspective in the finished piece. Using AI to help you write does not mean you have to accept whatever personality it gives you.

4. Check the facts before you repeat them.

If you’re quoting someone, go find where they said it and check the wording, attribution, and context. If you’re defining something, check an authoritative source and make sure the definition applies to what you’re discussing. The same goes for numbers, dates, and other factual claims. Open the sources and read them. A citation is something to verify, and the source needs to support the specific claim you’re making. “It sounded right” isn’t much of an explanation if someone asks where your information came from, and neither is “AI told me.” If you can’t verify something, leave it out or make the uncertainty clear.

5. When an answer seems wrong, check where it came from.

I see this with Ardi. Sometimes a client asks why it responded a certain way, and when we go back to the source document, the answer reflects the information they provided. Since Ardi only generates answers from client sources, we can trace the response back to the document it used. If that document needs an update, it gets corrected, and then we check the answer again after the updated content is in use.

That’s the “garbage in, garbage out” issue I’ve written about before. Finding it should help us resolve the problem. It’s different from posting an unreviewed AI draft, but they belong in the same family: we have to pay attention to what we give AI as well as what we do with the result.

You Still Have to Think

I don’t blame AI for a post someone chose to publish without reviewing. I hold the person using it responsible. AI is a tool, not a replacement for your thinking, your experience, or your judgment. If you decide the output is good enough to share, then that decision belongs to you. Human judgment matters. You can’t say that and then skip using it.

As an executive in this space, I think “We don’t accept AI slop” is a reasonable company standard, but we need to explain what that means. Before someone submits or publishes work, they should have checked its factual claims, confirmed that it communicates the intended meaning, and made sure it answers what the audience needs to know. They should be able to explain their conclusions and identify anything that still needs verification. Those are expectations an employee can follow and a manager can review.

The company standard is simple: “use AI to help you do your job, but we hired you for your judgment, and we expect you to use it.” That expectation has to apply to leadership, too. We can’t ask employees to review their work while passing along our own unchecked output. Be the example you wish to see. We all have a part in reducing AI slop, starting with the work we’re willing to put our names on.