In plain terms
Spam was the price of free email: sending cost nothing, so inboxes filled up. Slop is the same thing for content. Producing a plausible article, image or report now takes seconds, so far more gets produced than anyone wants to read. Slop may contain no errors at all. What marks it is that nobody put thought into it, and that the effort the sender saved is spent by everyone who receives it.
Why it matters
It reaches a business from two sides. Outside, slop crowds search results, marketplaces and social feeds, so your own content competes with an endless supply of filler, and customers grow suspicious of anything that reads as machine-made. Inside, colleagues forward AI drafts they have not read properly and the recipients do the thinking; the nickname for that is workslop. Publishing AI text unedited saves an hour and costs credibility. The remedy is unglamorous: a named author who has read the piece, something specific to say, and fewer, better pieces.
Example
A project manager asks an assistant for a status report and, after two minutes of work, forwards the six pages to 40 people. Each reader spends about six minutes finding out that the report holds two facts they need. The sender saved perhaps an hour; the organisation spent four. The next month the team agrees a rule: AI drafts are cut and checked by the sender, and every report states its conclusion in the first three lines.
Most often confused with
Slop vs. Hallucination
A hallucination is an error of fact inside an answer. Slop is a failure of care around the whole piece: generic, padded, unchecked, produced because it was cheap. Slop often contains hallucinations, since nobody looked, yet a text can be accurate from start to finish and still be slop. Fact-checking catches the first. The second is cured by editing, and by the decision to publish less.
Origin: Merriam-Webster chose “slop” as its Word of the Year for 2025, and Australia's Macquarie Dictionary chose “AI slop” in the same year.
Under the hood
Forms: search-optimised articles produced by the thousand, synthetic product reviews, AI images and short videos made for engagement, generated books on marketplaces, bug reports and pull requests submitted in bulk to open-source projects, and internal documents nobody edited. Recognisable traits: generic phrasing, padding, a confident tone without specifics, repeated structures, invented details and no accountable author. Three consequences are technical. Search engines treat mass-produced content made mainly to rank as spam and demote it. AI answer engines draw on the web, so slop about your market can end up in answers about you. And models trained on large amounts of model output lose quality over generations, an effect researchers call model collapse, which raises the value of verified human-written data. Countermeasures inside an organisation: disclosure rules for AI-assisted work, an editing step with a named owner, quality criteria in the evals for generated content, and limits on volume. Detectors of AI-written text are unreliable and produce false accusations; judging the content on its merits works better.