macquarie dictionary made ai slop its word of the year for 2025, defining it as low-quality content created by generative ai, often containing errors, and not requested by the user. the edits that stop it are concrete enough to run through a draft in ten minutes. read that definition closely and you learn how the judgement is being made. nothing in it is about which tool produced the text. it is about errors, and about nobody having asked for the thing.
that is the useful frame for anyone editing a draft. slop is not a style you can wash off with a thesaurus pass. it is text that carries no verified specifics and answers no actual request, and the surface tells are just the smell coming off that.
the tells a volunteer cleanup crew wrote down
wikipedia editors dealing with a flood of machine-written articles compiled the most detailed public list of what the output looks like. the signs of ai writing page names overused constructions including stands as a testament, plays a crucial role, underscores its importance and reflects broader, along with the contrast pattern that presents something as not only one thing but also another.
its vocabulary list is the one worth taping to your monitor: landscape, pivotal, meticulous, showcase, boasts, garner, vibrant. the grammatical tell is subtler and more reliable. models avoid a plain "is" and reach for serves as, stands as, functions as, represents. when you see three of those in a page, you are reading an inflated draft.
the formatting tells on the same page cover heavy em dash use, boldface scattered for emphasis, curly quotation marks, title case in section headings, and closing sections shaped like challenges and legacy or future outlook.
deleting the words does not fix the text
here is where most anti-slop advice stops, and it is why so much edited slop still reads as slop. swap showcase for show and you have a sentence that says nothing, in slightly better clothes.
the vocabulary tells are symptoms of the problem. a model reaching for stands as a testament is a model with no fact to state at that point in the paragraph, filling the slot with cadence. delete the phrase and the empty slot is still there. the repair is to put something in it: a number, a name, a date, a thing that happened.
try this on a paragraph of your own. underline every sentence that contains a proper noun, a figure, or a date. if fewer than a third of them do, the piece has a substance problem that no word list will touch.
the scale version, and what search does about it
the same failure at volume has a name in google's spam policies. google defines scaled content abuse as when many pages are generated for the primary purpose of manipulating search rankings and not helping users, and lists using generative ai tools or other similar tools to generate many pages without adding value for users as an example of it.
notice what is being policed. the offence is pages produced for the ranking rather than for a reader, which lines up exactly with the not requested by the user half of the dictionary definition. two very different bodies arrived at the same test.
disclosure, where somebody is relying on you
academic publishing took the other half of the problem, which is accountability.
responsibility does not thin out because a machine drafted the sentence. if a hallucinated citation ships under your name, it is your hallucinated citation. social posts are not preprints and nobody needs a disclosure line on a joke, but the same rule decides whether you check a claim before you post it.
the pass that works
run these in order on anything you are about to publish.
- check every proper noun and every number against a source you opened today. cut what you cannot verify.
- delete each sentence that would survive being deleted. slop is padded with sentences that restate the previous one.
- replace serves as, stands as and functions as with is, then look at whether the sentence still has a point.
- strip the closing section if it summarises what you already said. end on the last real thing you have to say.
- read it aloud. uniform sentence length is the tell no word list catches, and your ear finds it in one pass.
the second step does most of the work. machine drafts are long because length was the default, and a human edit that cuts the padding removes the parts nobody asked for.
the thing you are editing toward
a draft stops being slop at the point where it contains something the reader could not have guessed. one verified number, one specific example, one opinion with a reason attached. macquarie's committee also gave honourable mentions to clanker, for an ai-driven robot doing human tasks, and to medical misogyny, and the people's choice list added attention economy.