What is AI Slop?

AI slop is low-care content produced with AI.

It can look polished, but it is often generic, inaccurate, misleading, or empty. It can be text, images, videos, slides, summaries, emails, comments, reviews, or even scientific-looking material.

AI slop is not “anything made with AI”.

The problem is when AI is used to produce content without enough human thinking, checking, editing, or responsibility.

Key idea: AI slop is fluent output without care.

Why should we care?

Large Language Models are very good at producing text that sounds confident. But confident does not mean correct.

A polished AI-generated paragraph can still contain:

Problem	What it looks like
Generic writing	“In today’s fast-paced world…”
Unsupported claims	“Studies show…” without a source
Fake details	Invented names, papers, numbers, or links
Empty summaries	Many words, little meaning
Misleading images	Beautiful, but scientifically wrong
Wrong emphasis	Uncertainty removed or hidden

This matters because slop wastes attention. In science, teaching, communication, and administration, we do not only need nice wording. We need accuracy, context, judgement, and care.

Stay in the driver seat

AI assistants can improve our workflow. They can help us rewrite, proofread, structure ideas, generate draft illustrations, and identify unclear sections.

But we should not delegate the core thinking.

Human responsibility	AI can help with
Purpose	Suggest options
Facts	Organise information
Argument	Improve clarity
Context	Adapt tone
Judgement	Flag possible issues
Accountability	Never delegated

Rule of thumb: use AI to support your work, not to replace your responsibility for the work.

If your name is on it, you are responsible for it.

A poor workflow

1. Ask AI to write a full text about a topic.
2. Copy the answer.
3. Add an AI-generated image.
4. Publish or send it without checking.

This is how we get AI slop.

The output may look complete, but nobody has properly owned the content.

A better workflow

1. Draft the text yourself

Start with your own message. Write the main point, audience, and context in your own words. It does not need to be perfect.

The important part is that the content starts from your thinking.

1. Ask AI for a supporting illustration

Use AI to support the text, not replace it. Give a careful prompt that describes the image you want.

Example prompt:

Create a simple illustration for a training slide about AI slop. Show a researcher sitting in the driver seat of a small car labelled “human judgement”, with an AI assistant in the passenger seat holding a map. In the background, show a pile of generic documents and images labelled “unchecked AI output”. The tone should be clear, friendly, and educational. Avoid scary robots, glowing brains, or exaggerated sci-fi imagery.

Then check the image. Does it support the message? Is anything misleading? Are there strange details? Does it match the tone of the course?

1. Ask AI to proofread the text

Use AI as an editor, not as the author.

Instead of asking:

Write this section for me.

Ask:

Please proofread this text. Identify sentences that are unclear, too generic, unsupported, or repetitive. Do not rewrite everything. Give specific suggestions and explain why each change would improve clarity.

This keeps you in control. AI can help you notice weak parts, but you decide what to change.

1. Always read the final output

This step is not optional.

Before sharing, publishing, or teaching AI-assisted material, read everything.

Ask yourself:

Final check
Is the meaning correct?
Are the claims supported?
Is the tone appropriate?
Is anything generic or empty?
Would I be comfortable putting my name on this?

Examples to discuss

When adding screenshots, useful examples include:

• an AI-generated image that looks realistic but contains impossible details;
• a social media post that is emotional but unsupported;
• a generic AI-written paragraph that sounds professional but says little;
• an invented citation or a paper summary that does not match the paper;
• an illustration that is attractive but scientifically misleading.

For each example, ask:

Question
What looks polished?
What looks suspicious?
What would we need to verify?
Who is responsible if this is shared?

The goal

The goal is not to avoid AI.

The goal is to avoid becoming part of the slop machine.

Good AI use looks like this:

Human idea -> AI assistance -> human review -> better human output

Bad AI use looks like this:

Prompt -> AI output -> copy-paste -> publish

AI should be the assistant, not the author of our judgement.

---