Slop MopHow Slop Mop judges a post
Click the mop on any post and you get a breakdown: a score, a verdict, the Slopprint and three blue bars. Here's what each part means, what moves it, and what it can't tell you.
Reading the breakdown
Every part of the details panel, top to bottom. Pick one of three posts to see how the panel changes: one that folds away, one that's borderline, and one that reads fine. Try voting on them.
- The score, out of 100. Higher means more slop-like. It's worked out after the three counter-signals have taken their share off, so it's the final number. It never arrives alone: the signs behind it are always listed.
- The verdict. Looks fine, Possibly slop or Likely slop, each with a sentence saying why. When the model isn't confident, the verdict is Not sure and the post is left alone. Where the lines fall depends on your sensitivity setting (Mild, Moderate or Aggressive).
- The range bar. Where this post sits between fine and likely. In Hide mode a line marks where posts start folding away.
- Your vote. Not sure, No, Maybe or Probably. Not sure is where every post starts. A vote overrides the model's score in your feed and dims it in the panel, with or without an account. There's no vote on your own posts.
- The Slopprint. Nine tells of slop, one per spoke. The further out a point sits, the more of that tell the model found. All nine, with examples.
- In its favour. Three counter-signals: sounds like a person, useful to readers and reader response. More blue means more evidence the post is worth reading. They can only lower the score, never raise it.
- Community votes and checks today. How other Slop Mop readers voted on this post, and how many of your 250 daily checks are left. Votes from accounts, free or XL, count in full. Votes from a browser trying Slop Mop without an account count as a hundredth each, until that browser signs up.
Nine tells of slop
The model asks nine separate questions about the writing, one per spoke of the Slopprint. The shape tells you what kind of slop a post is, not just how much. Six tells come from Graphite's research on AI tells, which compared human and model writing. The last three are feed habits we added ourselves.
One spike isn't enough
Plenty of people have one over-the-top habit. When only one tell stands out, the score is cut. Slop usually shows up as several tells at once, and that's what gets flagged.
Unsure answers count for less
The model says how sure it is about each tell, and shaky answers carry less weight. That calibration is what it's designed for. If it isn't sure enough about the post overall, the verdict is Not sure and the post is left alone.
The weights are private
Some tells count for more than others. The exact weights on the hosted server aren't published, so nobody can write a post that games them.
Scroll-stopper dramatic hook, one-line paragraphs, a moral
Then everything changed.
Here's what I learned 👇
Engagement bait asking for comments, reposts and follows
Hype words marketing vocabulary
Empty praise evaluation with no detail behind it
No-catch promises all benefit, no cost
“Not X, but Y” defining things by what they aren't
Flowery prose flourish instead of the plain claim
Stiff phrasing stock transitions and hedges
Too-tidy story an anecdote that exists to prove the moral
counter-signal 1 of 3
Sounds like a person
Does this read like one particular person, or like a generic professional voice that anyone could have written? The model reads the text only.
What raises it
- Specific personal detail: names, places, dates or numbers from the writer's own life or work
- Word choices and asides that are clearly theirs
- Uneven rhythm: long sentences, short ones, fragments
- Contractions and casual qualifiers (“pretty”, “sort of”)
What doesn't count on its own
- “I”, “my” or “we”
- Naming the writer's company
- “I believe”, “in my experience” and similar framing
Anyone writing on someone else's behalf uses those too, so they don't prove there's a real voice.
What it does to the score
It takes a little off the slop score. It's a nudge, not an exemption: a very personal post stuffed with engagement bait can still be flagged.
This is not the same as “was it written by AI?” Slop Mop asks that separately, and the answer can only soften the score of a post that reads as typed by a person. It never flags a post by itself.
counter-signal 2 of 3
Useful to readers
Could someone doing the job learn something from this post, or act on it? Who or what wrote it doesn't matter. Useful AI-assisted writing scores just as well as useful writing typed by hand.
What counts
- Specific facts and numbers
- Named, concrete examples
- First-hand experience, with the details
- A method or step a reader could follow
What doesn't count on its own
- An opinion or position, however well argued
- A framing, a manifesto or a call for change
- Citing a well-known source or disclaimer
- “More on this in my next post”
A thoughtful opinion piece can be worth reading. It just isn't what this bar measures.
Useful isn't the same as popular
A post nobody has commented on yet can score high here, and a viral post can score low. Popularity has its own bar, reader response. Usefulness is half of the shield: a genuinely useful post has a large share of its slop score taken off. Never all of it, so a useful post buried in bait can still be flagged.
counter-signal 3 of 3
Reader response
What readers actually did with the post: reactions, comments and reposts. This one isn't the model's opinion. It's counted from the numbers shown under the post.
Effort counts more than taps
A comment costs someone minutes, and a repost puts the post in front of their whole network. Both count for far more than a one-tap reaction.
Hollow engagement is discounted
Hundreds of reactions with almost no comments or reposts is what an engagement pod or a cheap hook looks like. Once a post has lots of reactions, a thin comment-and-repost rate earns it less credit.
The first responses move it most
The scale is logarithmic. A post's first comments and reposts move the bar a lot; the thousandth reaction barely moves it.
What it does to the score
Reader response is the other half of the shield, alongside usefulness. Together they can take a large share off a post's slop score, but never all of it. Posts are checked again as they pick up engagement, so a new post starts with an empty bar and can fill it later. Your own posts and drafts leave this bar out entirely.
How it adds up
From twelve answers to one score, in order.
- Nine tells. The model's answers on the Slopprint are weighted and combined. A single tell standing out on its own is cut; shaky answers count for less.
- Sounds like a person. Takes a little off.
- The shield. Half usefulness, half reader response. Takes a share off the score, up to a cap, so no post is cleared by the shield alone.
- Human-typed? If the post reads as typed by a person, the score is softened a little. This step can only lower the score; being AI-assisted never adds to it.
- Your threshold. The final score is compared with your sensitivity. Looks fine is left alone. In Highlight mode, the post's mop turns yellow for Possibly slop and red for Likely slop. In Hide mode, posts past your threshold fold away into a strip you can reopen.
When it gets one wrong
Click the mop beside the post's “…” menu and vote No, Maybe or Probably, or Not sure to clear your vote. Your vote overrides the score in your feed. When your vote disagrees with Slop Mop's verdict, the post's text is saved and a person at Artimovo reviews it; the ones it got wrong become training examples for tuning the tells and the score. You can't vote on your own posts: the panel has no vote on them.
You don't need an account to vote. Before you sign up, each vote counts as a hundredth of a community vote; signing up makes every vote that browser has already cast count in full. Try it on this folded post:
If a bar looks badly off, email hello@slopmop.lol with the post's link and which bar surprised you.