Slop Mop Chrome Extension Sends LinkedIn Posts to Jev for Scoring

Slop Mop filters bad LinkedIn writing without claiming to spot AI​

The Register reported that Frazier released Slop Mop as a Chrome extension on Tuesday and also published its code on GitHub for anyone who wants to tinker with it. The main idea is simple, and the store listing says it plainly: it is a bad-writing detector which can often mean AI but is specifically not an AI detector.

Frazier put it bluntly in his comments to The Register: "Slop Mop flags poor human content equally to poor AI content… I feel that is the right decision because AI content can also be good and human content can also be bad." The GitHub repository uses the same wording, calling the project a bad-writing detector built as a Chrome extension plus a Vercel server, powered by Jev from Typesafe AI.

This framing is a real design choice. The extension does not try to prove that a robot wrote something, does not accuse authors, and does not put "written by AI" labels on posts. Instead of asking who wrote a post, it asks whether the post deserves your attention.

The timing lines up with LinkedIn's own work on the problem. On September 15, 2026, LinkedIn said more than one million members used the "Seems like AI slop" feedback option within the first two weeks after it launched in August. The company also said recent efforts have already helped reduce views of content classified as AI slop by 40%. Both numbers come from LinkedIn. LinkedIn defines slop as low-effort, likely AI-generated content that may sound polished on the surface but lacks a clear point of view, unique perspective, or substance. Slop Mop starts from a different place: it drops the AI part and judges the writing alone.

How Slop Mop's scoring questions and the Jev model work​

As you scroll, Slop Mop picks up posts before they reach your screen and asks twelve questions about the writing. Nine look for signs of slop. Two look for reasons to spare the post: whether it sounds like a person and whether it contains something genuinely useful. One asks how machine-written the post reads, and that answer nudges the score but never decides the verdict by itself.

The nine signs are:

1. Scroll-stopper formatting: a single short line followed by a dramatic gap.
2. Engagement bait such as "Agree?" or "Comment YES below."
3. Hype words like "unlock," "paradigm shift" and "game-changing."
4. Empty praise such as "powerful" or "insane" with nothing to back it up.
5. No-catch promises that sound too good to be true.
6. Definition by contrast, the "not X, but Y" construction.
7. Flowery prose where a plain sentence would have done the job.
8. Stiff phrasing such as "furthermore" or "it is worth noting."
9. A too-tidy anecdote that lands its lesson like a corporate keynote.

The Register traces these signs to research by SEO firm Graphite Growth into patterns that appear disproportionately often in AI-generated writing. Frazier says the library began with Graphite's research on AI-writing tells, then got renamed because they're really slop tells. Plenty of humans can hit all nine without mechanical assistance.

The two protective questions matter a lot in practice. The listing says a slightly formulaic post that actually teaches you something can survive the mop, and that's deliberate because useful beats authentic. One detail is worth flagging: Frazier's own site describes asking Jev "eleven questions," while the store listing says twelve. The store listing is the most recently dated record, and The Register's account of a 12th, AI-likelihood question matches it.

All of the judging runs on Jev. The Register describes it as designed to make probabilistic decisions for other software systems and accessed via an API, not something you chat with. Other open-source projects built on Jev show the same pattern. One of them, Jev Slop Guard, describes one typed { slop, not_slop } choice per post, no free-form LLM text. That's a useful contrast, though Slop Mop itself scores several separate questions per post.

What Slop Mop's Hide and Highlight modes do to your LinkedIn feed​

Slop Mop never deletes anything. It changes what you see in your own browser. In Hide mode, suspected slop gets wrung down into a thin strip showing its score and two strongest signals. Click once and the post unfolds in place. In Highlight mode, everything stays exactly where LinkedIn put it. Questionable posts get a colored icon: yellow for possibly slop, red for likely slop.

Both modes offer Mild, Moderate, and Aggressive sensitivity, and whatever is already on your screen updates immediately when you change a setting. There are four verdicts: likely slop, possibly slop, "not sure" (shown grey and hatched), and "looks fine," which leaves the post untouched.

The failure behaviour is the most reassuring part of the design. The listing says if Slop Mop can't make a judgment confidently, it assumes the best not the worst. Timeouts, network problems, rate limits and unreliable answers all leave the post alone. So if the service goes down or you hit the daily cap, your feed simply goes back to normal. Posts don't disappear.

Every post also gets a vote control. Your vote overrides the model for your own feed. Vote No on something Slop Mop folded and it opens back up. The scoring should be treated as early work: the current thresholds were fitted on a relatively small number of votes and are marked provisional in the source code. Since the extension only changes your view, The Register notes that flagged posts won't actually be removed from a LinkedIn feed and can still be reported using the "Seems Like AI Slop" button.

Where your LinkedIn post text goes when Slop Mop scores it​

This is the part to read before you click Add to Chrome. Slop Mop runs in your browser, but the judging doesn't. The listing says post text and public engagement counts are sent to Slop Mop's server at api.slopmop.lol and then to Jev so the writing can be judged. According to the developer, your name and the author's name are not sent.

The GitHub README gives more detail than the store listing. The server keeps a hash of the text (not the text itself), the post's LinkedIn id, Jev's scores, engagement counts, and vote tallies. It doesn't store author names. Votes are tied to a random install id, never to your name or LinkedIn account, and only a salted hash of that id is stored. Logs of each check (time, token count, latency, outcome, hashed install id) are kept for 90 days to run an admin dashboard. The README's mention of the post's LinkedIn ID doesn't appear in the store listing's list of retained data. It's a small gap, but readers who care about privacy should know about it.

The developer also lists things the extension doesn't do. It doesn't change anything on your LinkedIn account, read your messages, read your profile, read your connections, or post, like, comment, or connect on your behalf, and it claims no cookies, no browsing-history collection and no advertising vendors. All of these are the developer's own statements. The Chrome Web Store's disclosure section lists "User activity" and "Website content" as the categories of data the extension handles. That is a self-declared label, not an audit result.

A hash doesn't change the basic fact that the full text of every feed post you scroll past is sent to two outside services for scoring. For personal use on a public feed, that may be fine. For anyone using LinkedIn on a managed work machine, it's worth weighing before you install.

Why Slop Mop is free: Frazier is paying for Jev himself​

The Register reports that the whole service is running on Frazier's Jev account, and thus is running requests through his API key. To keep his bill under control, up to 250 posts can be analyzed per installation per day, and the store listing gives the same limit. Frazier also shared his early costs: "From my early install base I am seeing a cost per user of ~$0.005 per day… If this project goes viral, I might re-evaluate the free service portion of it."

Those figures are his early estimates, and he has made no promise about the long term. The site calls Slop Mop a research project, not a product, and the listing says there's no paid tier. If the free hosted version goes away, the MIT licence is the backup plan.

The open-source code has an important limit, though. The README says the tuned weights stay on my server. This repository ships every sign at equal weight. That's enough to run and to build on, but production scores use weights I've tuned, which is why they can differ from yours. A self-hosted copy works, but it won't score posts the same way the Chrome Web Store version does unless you tune it yourself.

Running your own Slop Mop server​

The repository includes both parts: the Chrome (MV3) extension and the Vercel Functions backend: Jev scoring, a registry of content and votes, per-install limits, and the admin dashboard. The README describes this setup for loading a locally built copy:

1. Build the extension, then open chrome://extensions, turn on Developer mode, choose Load unpacked and pick slopmop-extension/dist.
2. To run everything on your own machine, cd slopmop-server && npm run dev, then build the extension with the default URL.
3. From there you can run your own server, point the extension at it, change the weights or the thresholds, replace the sign library, or build something else.

You'll still need your own access to Jev for scoring. A curated list of Jev projects describes Slop Mop's setup as free hosted use with daily check limits, or self-host MIT with your own key. Self-hosting means the post text goes to your own server instead of Frazier's before it reaches Jev.

Slop Mop is one of several Jev-powered LinkedIn filters​

Slop Mop isn't the only tool of its kind. Since Jev launched, developers have released a string of similar extensions, most of them requiring you to bring your own API key. Jev Slop Guard covers X and LinkedIn: every post gets a small pill with a slop probability. Another project marks AI-generated and AI-slop posts while you scroll X / Twitter, LinkedIn and Reddit. A third, AI Slop Detector for LinkedIn, analyzes visible LinkedIn post text for low substance writing patterns using your own Jev API key.

Slop Mop stands apart in two ways. It's the one with a hosted, no-key setup and a Chrome Web Store listing. It also frames itself as a bad-writing filter rather than an AI detector. Some of the alternatives label posts as AI-generated or human, which is exactly the kind of authorship verdict Slop Mop avoids.

Frazier is also upfront about the limits. The README says Slop Mop can be gamed. Once people know the signs, some will write around them. That's part of what's worth measuring. It supports desktop Chromium only. It doesn't run on mobile or in Firefox or Safari. Microsoft Edge is built on Chromium, so it should fall under "Chromium-compatible" in principle. That's an inference, though: the listing doesn't name Edge, and nobody has reported testing it there.

What this means for you​

If you read LinkedIn on a desktop and are tired of engagement bait, Slop Mop is a low-risk trial, as long as you're comfortable with feed text going to Frazier's server and on to Jev. Highlight mode lets you judge its calls before you let it fold anything away. The main reasons to hold off are workplace policy and privacy: on a managed PC, sending feed content to an outside service is your organisation's decision to make, not just yours.

• Slop Mop 0.2.1 is free on the Chrome Web Store and works only with desktop LinkedIn in Chrome or Chromium-compatible browsers, with no mobile, Firefox or Safari support.
• Start with Highlight mode on Mild or Moderate. Hide mode folds posts into clickable strips but never deletes them, and a "No" vote reopens a post for your feed.
• Post text and public engagement counts go to api.slopmop.lol and then to Jev. The developer says it keeps a hash of the text rather than the text itself, and keeps check logs for 90 days.
• The hosted service is capped at 250 checks per installation per day. When it can't score a post because of the cap, a timeout or an error, the post is left visible.
• The MIT-licensed GitHub code gives every sign equal weight, so a self-hosted copy will score differently from the store version until you tune it.
• LinkedIn's built-in "Seems like AI slop" option under a post's "…" menu is still the only way to send feedback to LinkedIn itself. Slop Mop changes only what you see.

Slop Mop's biggest strength is that it knows its own limits. It admits it's uncertain, fails safe, and lets you overrule it. The weak point is Frazier's willingness to keep paying for everyone's Jev usage. If installs climb well past the early numbers, the thing to watch is whether he changes the free hosted service. If he does, anyone who wants to keep using it will need to run their own server with their own Jev access and tune the weights themselves.