SlopShape looks for patterns in how a commercial blog post organizes and presents information—not just words or a raw-HTML fingerprint. In its current preprint, the structure-only classifier reached 97.0 macro-F1 on a controlled test of pre-ChatGPT human posts and AI-generated counterparts, and 96.1 after a specific model-self-rewording test. Those are strong results for that benchmark, not proof that a real-world post was written by AI.
What SlopShape measures
SlopShape is a research classifier for distinguishing human-written from AI-generated B2B commercial blog posts. It represents each post through judgments about its document-level structure: how it orders information, presents evidence, frames its argument, and uses voice. It does not simply inspect the page’s HTML structure or search for a list of telltale phrases.
As an Amazon Associate I earn from qualifying purchases.
The current instrument contains 203 features across 11 dimensions. Of those, 176 are structural and 27 concern writing style. The headline structure-only result uses the structural features, leaving out the instrument’s writing-style features.
- Purpose and audience
- Structure and flow, explanation, and evidence
- Voices and actionability
- Commercial integration and timeliness
- Page format and writing style
The paper describes a recurring combined pattern in AI posts as “tidy, self-announcing”: a post states its thesis early, previews its organization, contrasts an older approach with a newer one, ends with a summary, and adopts a confident institutional voice. The classifier uses combinations of features; no single habit establishes authorship.
#1 Best Overall
How the classification pipeline works
An LLM is part of the measurement process. It converts a post into structured feature judgments, and a separate classifier uses that representation to predict a source class. This is not a deterministic parser that reads a DOM tree and counts layout elements.
- Represent each post. The pipeline creates a JSON-like template for the document using a commercial-content schema.
- Find candidate features. The researchers compare human posts with AI mirrors to identify potentially distinguishing characteristics.
- Screen the features. Candidate features are assessed for whether they can be answered, whether evidence can be located in the post, and whether judgments are reliable.
- Score and encode. The final feature instrument is applied to each document, and its answers are encoded for classification.
- Test on unseen company domains. Classifiers are evaluated using company-domain-disjoint splits, so posts from test companies are not represented in training.
The paper also runs a separate six-way task: distinguishing human writing from posts generated by each of five tested AI systems. That source-attribution task is different from the binary human-versus-AI classification.
What the current paper reports
Jochen Madler’s version 3 preprint, revised September 28, 2026, reports the following benchmark and validation results:
Free tools Windows power users keep installed
One-click scans. No signup required.
| Result | What it measures |
|---|---|
| 97.0 macro-F1 | Structure-only human-versus-AI classification on held-out company domains. |
| 96.1 macro-F1 | The same structure-only task after each AI test post was reworded by the model that generated it. |
| 68.6% correct | Six-way source attribution: human writing or one of five generating models. The reported chance level is 16.7%; this is not the binary detection score. |
| 0.939 Cohen’s kappa | Agreement between the two human annotators in the gold-annotation session. |
| 0.951 mean Cohen’s kappa | Mean agreement between the human annotators and the LLM scorer on the annotated feature items. |
In this corpus, the structure-only classifier outperformed the style-only classifier by 8.8 macro-F1 points. The two approaches also made errors on mostly different posts, which supports the paper’s claim that structural features contribute information beyond surface wording. The reported agreement values provide validation for feature scoring, but they come from a limited annotation session.
Rank #3
![MixPad Free Multitrack Recording Studio and Music Mixing Software [Download]](https://m.media-amazon.com/images/I/71ltIxIuz1L._SL160_.jpg)
- Create a mix using audio, music and voice tracks and recordings.
- Customize your tracks with amazing effects and helpful editing tools.
- Use tools like the Beat Maker and Midi Creator.
- Work efficiently by using Bookmarks and tools like Effect Chain, which allow you to apply multiple effects at a time
- Use one of the many other NCH multimedia applications that are integrated with MixPad.
What was in the benchmark
The study paired 2,250 human B2B blog posts from 268 company domains with 11,250 AI mirrors. The human posts were archived from 2008 through 2022 and predate ChatGPT. The mirrors were generated in August 2026 from reverse-engineered briefs using GPT-5.4, Claude Sonnet 4.6, Gemini 3 Flash, DeepSeek V3.2, and Kimi K2.5. Evaluation splits held out company domains.
This design tests whether features learned from one set of companies distinguish the constructed human and AI sets at other companies. It does not recreate every way people publish online. The original human author had full business context, while the AI systems received briefs reconstructed from the published post; differences in what those briefs omitted may therefore contribute to the signal.
Rank #4
![Express Schedule Free Employee Scheduling Software [PC/Mac Download]](https://m.media-amazon.com/images/I/41yvuCFIVfS._SL160_.jpg)
- Simple shift planning via an easy drag & drop interface
- Add time-off, sick leave, break entries and holidays
- Email schedules directly to your employees
Where the result stops
The headline scores describe one corpus and protocol, not universal accuracy. The paper tests first-pass generations from five named systems and a particular rewrite condition in which each system reworded its own test posts. It does not establish performance after extensive human editing, human-AI collaboration, use of humanizer products, or deliberate attempts to evade structural detection.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →- Time period: The human posts are pre-ChatGPT, while the AI mirrors were made in 2026. The authors check for publication-year effects, but say a same-period human-versus-AI comparison remains open.
- Genre and geography: The corpus is software-heavy and US-heavy. Some 75.5% of the human snapshots came from 2020–2022. The authors do not claim that commercial writing is inherently easier to separate than fiction.
- Mixed authorship: The benchmark’s labels are source classes in a constructed experiment. It does not establish how the classifier performs on posts combining human and AI contributions or conventional editorial templates.
- Measurement: An LLM helps discover, screen, extract, and score features. Human validation and repeatability checks are reported, but the annotation session was limited and both annotators were company-affiliated.
- Format: Human pages and AI outputs entered scoring in different formats. The paper reports format-sensitivity checks and says affected features were removed, but the distinction is relevant when interpreting the pipeline.
These limits matter in particular for formulaic marketing and SEO content: a tidy structure can result from an editorial convention as well as an AI-generated pattern. The paper’s benchmark does not settle how often that kind of human writing would be flagged in current, same-period material.
Best Value
![WavePad Audio Editing Software - Professional Audio and Music Editor for Anyone [Download]](https://m.media-amazon.com/images/I/B1fcLEGCs6S.png._SL160_.png)
- Full-featured professional audio and music editor that lets you record and edit music, voice and other audio recordings
- Add effects like echo, amplification, noise reduction, normalize, equalizer, envelope, reverb, echo, reverse and more
- Supports all popular audio formats including, wav, mp3, vox, gsm, wma, real audio, au, aif, flac, ogg and more
- Sound editing functions include cut, copy, paste, delete, insert, silence, auto-trim and more
- Integrated VST plugin support gives professionals access to thousands of additional tools and effects
How to interpret SlopShape or a checker score
Treat a structural score as a triage or research signal, not proof of who wrote a post and not a sound basis for accusing an individual writer. A result from this experiment means the features aligned with the study’s source labels under its data and protocol; it does not resolve authorship for an arbitrary web page.
Sitefire’s Slop Checker is a related demonstration: it accepts one blog-post URL and returns a “Human score” based on ten structural signals from the study. Its product page says the study measured English posts and that scores for other languages are indicative. That makes the checker a screening aid, not an independent validation of authorship.
Paper release and commercial context
The public release repository says it includes code, prompts, the feature instrument, aggregate artifacts, and verification materials. The code is under PolyForm Noncommercial License 1.0.0; other release materials are all rights reserved and made public for audit or verification. Some document-level data are available to researchers under a noncommercial research agreement. The repository also discloses the author’s commercial interest in Sitefire and says commercial use of the code requires a separate agreement.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
