What is AI slop?
A dictionary made it word of the year, which tells you the backlash arrived faster than the technology it is about.
Key figures
Macquarie Dictionary defines it as low-quality content created by generative AI, often containing errors, and not requested by the user.
It won both the committee and the public vote for 2025, only the fourth time those two have agreed.
The defining feature is not that it is machine-made. It is that nobody asked for it.
What the word actually means
Macquarie Dictionary's definition, in choosing AI slop as its 2025 word of the year, is worth reading closely: low-quality content created by generative AI, often containing errors, and not requested by the user.
Three conditions, and the third is the interesting one. Slop is not simply content made with AI. It is content that arrives uninvited, in a feed, in a search result, in an inbox, positioned where something someone wanted used to be.
The word came from the community it describes rather than from a style guide. It circulated online through 2023 and was popularised in a May 2024 post by the programmer Simon Willison, who drew the boundary precisely: not all AI-generated content is slop, but content mindlessly generated and thrust upon someone who did not ask for it is.
It won both the committee's choice and the people's choice. The dictionary noted that this was only the fourth time the two have agreed, which is itself a measure of how little disagreement there is about the phenomenon.
How to tell slop from AI writing
| Characteristic | AI slop | Legitimate AI-assisted work |
|---|---|---|
| Who asked for it | Nobody; it fills a slot | Someone with a question or a need |
| Relationship to facts | Unchecked, often wrong | Checked against a source that is named |
| What it adds | Restates what is already everywhere | Something not otherwise available |
| Accountability | No identifiable author or publisher | A person or organisation that can be held to it |
Google’s own spam policy arrives at the same line from the other direction. Its scaled content abuse rule targets producing large amounts of unoriginal content to manipulate rankings, and states that this applies no matter how the content is created, whether by automation or by people.
The useful implication is that the tool is not the test. A carefully researched piece drafted with AI assistance is not slop. A hand-written article that restates what is already everywhere, cites nothing and exists to occupy a search result is slop in everything but the manner of its production.
Why it appeared so fast
The economics changed in one step. Producing a plausible thousand words used to cost an hour of somebody's attention, and that cost was the only thing limiting how much low-value content existed. When it fell to a fraction of a penny, the limit disappeared, and everything downstream that assumed the limit — search ranking, feeds, review sections, inboxes — was suddenly holding a filter designed for a world with a hundredth of the volume.
Macquarie's committee made a related observation: people had learned to be search engineers to find useful information, and now find themselves prompt engineers wading through slop.
What it costs everyone
The obvious cost is the reader's time. The less obvious one is that slop makes genuine work harder to recognise, because the surface signals people used to rely on — it is well written, it is confident, it has headings, it looks researched — are exactly the signals a language model produces for free.
What survives as a signal is the specific and checkable: a number with its source named, a date of measurement, a method, a stated limit. Those cost something to produce, which is the entire reason they still mean something.
How not to produce it
One question is enough. What does this page contain that the reader could not get anywhere else?
If the answer is a measurement you made, a source you checked, a document you read properly, or an argument that resolves a real disagreement, it is not slop, whatever drafted it. If the answer is that it exists, it is slop, whoever wrote it.
Sources and method
- AI slop crowned word of the year 2025, Macquarie Dictionary, reported 25 November 2025. Source of the definition, the word-of-the-year selection and the note that the committee and public choices coincided.
- Slop is the new name for unwanted AI-generated content, Simon Willison, 8 May 2024. Early popularisation of the term and source of the boundary quoted here: not all AI-generated content is slop, but content mindlessly generated and thrust upon someone who did not ask for it is.
- What web creators should know about our March 2024 core update and new spam policies, Google Search Central, March 2024. Google’s scaled content abuse policy, which defines the abuse as producing large amounts of unoriginal content to manipulate rankings no matter how it is created, whether by automation or by people.
- Other dictionaries selected related terms for 2025; this page cites Macquarie because its published definition is the most precise of them.
- The characteristics table is this page's own summary of how the term is used, not a quotation from the dictionary.
- Every source here was opened and checked on 25 September 2026. Where a question has no measurement behind it, this page says so rather than dressing an opinion as data.