Every technology wave invents a word for its own waste product. Email got spam. Generative AI got slop. The word is older than the technology — in the 1700s it meant soft mud, in the 1800s pig feed, and later anything of little value. Since 2024 it has settled into a precise new meaning, and in 2025 Merriam-Webster named it Word of the Year, defining it as “digital content of low quality that is produced usually in quantity by means of artificial intelligence.”

Developer Simon Willison, one of the term’s early proponents, frames it by analogy: the way spam gave everyone a word for unwanted email, slop gives everyone a word for unwanted AI-generated content — material that is mindlessly generated and thrust upon people who didn’t ask for it. This page explains what AI slop is, why there is suddenly so much of it, and — most relevant here — how it relates to the way you prompt.

What makes something slop

Slop is not simply “content made by AI.” An article drafted with an LLM and then carefully fact-checked, edited and owned by a person is not slop. Slop is defined by what is missing, not by what tool was used. The recurring hallmarks:

• Produced in bulk. The point is volume — hundreds of articles, images or videos — not the quality of any single one.
• No human review. Nobody read it, tested it or verified it before it was published. Willison’s blunt version: sharing unreviewed AI-generated content with other people is rude.
• No real intent. It was not made because someone had something to say; it was made to fill a slot, farm clicks, or game an algorithm.
• Plausible but hollow. It looks the part — fluent sentences, polished images — but says nothing, and often contains quietly wrong facts, invented citations or impossible details.
• Unwanted. It arrives in feeds, search results and inboxes without being asked for, and it costs the reader time to discover that.

The result is content that shifts its cost onto others. It is nearly free to generate, but every reader pays — in attention, in fact-checking, or in trust slowly eroded. That asymmetry is the whole problem.

Why there is suddenly so much of it

Slop is an economics story before it is a technology story. Generating a passable article, image or video now costs fractions of a cent, while the attention economy still pays out per impression. When production cost collapses and distribution is free, volume wins — unless quality is enforced somewhere. So feeds fill with absurd videos, off-kilter ad images, fake news that looks real, and junky AI-written books, each one cheap to make and expensive to sort through.

There is also a quieter, workplace version, sometimes called workslop: the polished-looking report, summary or slide deck a colleague generated in seconds and forwarded without checking. It looks like progress, but the real work — reading, correcting, deciding — has simply been handed to whoever receives it.

Slop in software

Code has its own strain of slop. It shows up as pull requests nobody truly read, plausible-looking functions that call APIs which do not exist, and sprawling generated files that grow a codebase faster than anyone’s understanding of it. The pattern is identical to the written kind: bulk output, no review, no ownership — and the cost pushed onto the reviewer, the maintainer, and eventually the user.

This is where prompting comes in, because slop is usually a prompting failure before it is a model failure. Vague instructions (“write me a blog post about X,” “add feature Y”) produce generic filler, and generic filler shipped unreviewed is the definition of slop. The practices on this site exist precisely to prevent that: a prompt built with real context, a clear role and a concrete goal gives the model something genuine to execute, and a disciplined verify step — ideally not by the same agent that produced the work — is what stops bad output from leaving the building. Frameworks like CRAFT (for Software Development) encode the same idea: humans define the boundaries and own the result; the AI only accelerates the execution.

How not to produce (or ship) it

• Start with intent. If you cannot say what the piece is for and who it serves, the model cannot either — you will get filler.
• Feed it real context. Specifics, sources, constraints and audience are what separate content from noise.
• Review before it leaves your hands. Read it, run it, fact-check it. Unreviewed output is a draft, never a deliverable.
• Own what you publish. Attach your name and credibility to it. If you would be embarrassed to defend it line by line, it is not ready.
• Publish less, better. The honest answer to near-zero generation cost is not more output — it is more judgement per output.

Sources and further reading

This page is our own summary of a term that entered working vocabulary in 2024 and reached mainstream recognition by the end of 2025.

• Simon Willison, Slop is the new name for unwanted AI-generated content (May 2024) — simonwillison.net. The spam analogy and the “unreviewed, unwanted” framing used above.
• Merriam-Webster, 2025 Word of the Year: Slop — merriam-webster.com. The dictionary definition quoted in the introduction.
• The Guardian, Spam, junk… slop? The latest wave of AI behind the ‘zombie internet’ (May 2024) — theguardian.com.
• Wikipedia, AI slop — en.wikipedia.org. An overview of the term’s usage and examples across media.