AI Writing Detection 2026: Tropes.fyi for Social Creators

In 2026, AI-generated text on social platforms has become so common that a new category of web tool has emerged: the AI writing trope detector. These tools don't just give you a probability score. They tell you exactly which patterns in a piece of writing give it away as AI-generated. The most interesting example right now is Tropes.fyi, created by Ossama (ossama.is).

Tropes.fyi is a dynamic directory of AI writing patterns — the structural tics and rhetorical habits that large language models fall into when generating text. It catalogs these patterns across six categories, provides real examples, and ships four practical tools that social content creators can use to detect, audit, and fix AI-tainted writing.

This guide walks through what Tropes.fyi does, why it matters for social creators, and how to integrate its tools into your content workflow.

H2-1: The Six Categories of AI Writing Tropes

Tropes.fyi organizes AI writing signals into six categories, each with sub-tropes documented with real examples:

• Word Choice (4 tropes) — Overused words, faux-intellectual vocabulary, and improbable collocations that models default to. Examples include "delve," "navigate the complexities," and "it's worth noting that" used as filler.
• Sentence Structure (8 tropes) — The largest category. Covers negative parallelism ("It's not X — it's Y"), the em-dash addiction (every other sentence has an em-dash pivot), short punchy fragments, anaphora abuse (repeating the same word at the start of consecutive sentences), and the "here's the kicker" opener.
• Paragraph Structure (2 tropes) — One-point dilution (restating the same argument 8 times with slightly different words) and the template paragraph (topic sentence, explainer, "however" pivot, example, close).
• Tone (9 tropes) — The pseudo-enthusiasm problem (everything is "exciting," "game-changing," or "revolutionary"), the faux-humble intro, and the corporate-casual blend that no real writer uses.
• Formatting (3 tropes) — Bold-first bullets (every bullet starts with a bolded keyword), excessive use of numbered lists for simple concepts, and the "TL;DR" sandwich.
• Composition (7 tropes) — The fake counterargument ("Some critics say... but they're missing the bigger picture"), transition padding (however, moreover, furthermore at every paragraph break), and the conclusion that restates the introduction verbatim.

What makes Tropes.fyi different from a static list is that each trope links to real-world examples from published content. You can see the pattern in context, which makes it far easier to recognize in your own writing or in content you're evaluating.

H2-2: The Four Tools in Tropes.fyi

Beyond the directory itself, Tropes.fyi ships four practical tools that social creators can use immediately:

Tool 1: AI;DR — Reverse Engineering the Prompt

The AI;DR tool takes a long-form piece of content and attempts to reverse-engineer it back to the original prompt. Paste in a URL, and it distills thousands of words of "slop" into the likely 30-word prompt the author typed. This is brutally effective at exposing content that was generated with minimal human input.

Example from the site: A 2,847-word Medium article titled "Why AI Will Change Everything..." was distilled to a single prompt: "Write a blog post about how AI should be used as an exoskeleton that amplifies human ability rather than replacing workers. Include statistics about real exoskeletons in manufacturing and military. Make it persuasive."

For creators, AI;DR serves two purposes. First, it lets you audit content you're citing or sharing — is it written by a human or generated? Second, it's a teaching tool: if your own AI-assisted writing can be reduced to a single prompt, you haven't added enough of your own voice.

Tool 2: AI VETTER — One-Click URL Audit

# Drop a URL into Tropes.fyi's AI VETTER
# Returns:
# VERDICT: Suspicious
# 7 tropes identified across 4 categories
# Heavy use of negative parallelism, anaphora abuse
# One-point dilution: single argument restated 8 times

The AI VETTER takes a URL and runs it through the trope detection engine. It returns a five-point verdict scale: Human, AI-assisted, Suspicious, or Pure slop. Below the verdict, it shows the specific tropes detected, how many were found, and which categories they belong to.

This is the tool social creators will use most often. Drop in a LinkedIn post you're not sure about. Drop in a Bluesky thread that feels off. The VETTER gives you an objective breakdown based on structural patterns, not gut feeling.

Tool 3: DESLOPIFY — Diff-Style Rewrite Suggestions

DESLOPIFY takes pasted text and produces a diff-style breakdown of every phrase, sentence, and structure that reads like AI. Each flagged section comes with a suggested rewrite. This is useful in two directions: to identify what makes existing content read as AI, and to learn what patterns to avoid in your own writing.

The tool flags patterns like "The technology is extraordinary, the demos are dazzling, and the marketing is relentless" — a triple-parallel construction that's classic AI — and suggests alternatives that sound more human: shorter, more specific, less symmetrical.

Tool 4: TROPES.MD — The System Prompt File

Perhaps the most practical tool for creators who use AI assistants: a single Markdown file you can add to your AI assistant's system prompt or context to help it avoid common AI writing patterns.

# Add this to your AI assistant's system prompt:
# Source: tropes.fyi
#
# ## Negative Parallelism
# Avoid "It's not X — it's Y" patterns
# Avoid the em-dash addiction
# Avoid bold-first bullets for every list item
# Avoid transition padding: however, moreover, furthermore

This is a concrete, actionable way to improve AI-assisted content. Instead of telling your assistant "write like a human" (a vague instruction it will fail to follow), you give it a specific list of patterns to avoid. The TROPES.MD file is MIT-licensed and lives on GitHub for easy inclusion in any project.

H2-3: Why Social Creators Should Care About AI Writing Tropes

For social content creators publishing on X, Bluesky, LinkedIn, and newsletters, AI writing detection matters at three levels:

1. Platform algorithm penalties. LinkedIn's algorithm explicitly demotes posts that read as AI-generated. X's feed ranking also deprioritizes content that shows low engagement relative to impressions — a signal that readers detect AI patterns even if they can't name them. Bluesky's community-driven moderation and feed algorithms increasingly filter content flagged for AI patterns.

2. Audience trust erosion. Readers are getting better at spotting AI writing. The "em-dash addiction" — where every other sentence pivots with an em-dash — has become so recognizable that Twitter users regularly call it out in replies. Once readers identify a creator as an AI-poster, engagement drops permanently.

3. SEO and AI citation value. Google's helpful content system evaluates content based on original reporting, analysis, and human expertise — all signals that AI-generated text struggles to produce. Meanwhile, tools like Tropes.fyi make it easier for readers to detect AI content, which means even search-driven traffic won't stick if the writing doesn't feel human.

H2-4: A Practical Workflow for Detecting and Fixing AI Tropes

Here is a 4-step workflow for social creators who want to audit their content and their sources:

Step 1: Vet your sources. Before citing or sharing an article, drop its URL into Tropes.fyi's AI VETTER. If the verdict is "Suspicious" or "Pure slop," find a human-written source or do your own reporting. Don't amplify AI-generated content in your feed.

Step 2: Audit your drafts. Before publishing a long-form piece on X Articles or LinkedIn, paste the body into DESLOPIFY. Look at the flagged patterns. Are you using negative parallelism? Em-dash addiction? Bold-first bullets? Revise the flagged sections.

Step 3: Train your AI assistant. Download TROPES.MD from Tropes.fyi and add it to your AI assistant's system prompt. This is the most effective single action you can take to improve AI-assisted writing quality. The assistant will avoid the most common tells by default.

Step 4: Archive with version history. Use ThreadGrab to capture your published content as Markdown. When you revise a post based on trope detection feedback, the original Markdown is preserved so you can see what changed. Over time, you build a personal database of patterns you've eliminated from your writing.

# Example: automate trope checking in your workflow
# Use Python to fetch rendered text and pipe to analysis
curl -sL "https://example.com/article" | \
  python3 -c "import sys; print(sys.stdin.read())" > /tmp/article.txt
# Then open Tropes.fyi DESLOPIFY and paste the content
# Or add TROPES.MD to your AI assistant and have it review drafts

H2-5: The Broader Trend — AI Detection Is Becoming a Standard Tool

Tropes.fyi is part of a larger movement in 2026 toward transparent, explainable AI detection. Unlike black-box tools that return a single number (GPTZero, Originality.ai), Tropes.fyi names each pattern and shows you where it occurs. This transparency has three benefits:

• Education: Creators learn what to avoid, not just that something is wrong. The directory format turns detection into a learning resource.
• Verification: You can confirm the detection yourself. Instead of trusting a probability score, you can read the flagged text and decide whether it's truly a trope or a legitimate stylistic choice.
• Improvement: The DESLOPIFY tool and TROPES.MD file provide a closed feedback loop: detect the problem, understand what it is, and fix it in future content.

For social content creators, the takeaway is clear. AI-generated text is not inherently bad — but the patterns that give it away are becoming visible to everyone. Understanding those patterns, avoiding them in your own writing, and using tools like Tropes.fyi to audit your output is becoming a basic competency for publishing on the social web.

H2-6: Frequently Asked Questions