Or to put it bluntly: AI writing is trash.
It is disrespectful to expect ME to read something YOU could not even be bothered to write (or likely even read). The lingering human connection that remained resilient on the internet - through years of SEO optimisation, propaganda bots, ads, regional firewalls, politics, censorship - is now being diluted even further.
Many of the posts I eagerly click on HN (especially sorting by new) are now disappointingly, ragingly, completely AI generated. This includes several that reach the front page on a daily basis. It's shameless. Unfortunately, many of you educated readers are oblivious to the tropes and tells or too eager to discuss the topic to notice.
To combat this (and because I'm petty) I (and ironically Claude) built a directory and several tools to name and shame these patterns. Both to educate those are not AI-literate and to spread awareness. NOT TO HARASS.
tropes.fyi is a catalogue of the recurring tropes, tics, structures and word choices that give away AI-generated writing. Each trope is documented with a name, description, real-world examples and the category it falls under (word choice, sentence structure, tone, formatting, composition).
There are currently 32 tropes. Some highlights:
- Negative Parallelism: "It's not X — it's Y." The SINGLE MOST COMMON AI writing tell. One per piece is fine. Ten in a blog post is insulting. AND I HAVE SEEN TEN IN A BLOG POST.
- Tricolon Abuse: The compulsive rule of three. "Products impress people; platforms empower them. Products solve problems; platforms create worlds."
- Em-Dash Addiction: A beautiful punctuation. Once the sign of a brilliant writer now too many of these is a red flag. I still use them, you won't break my spirit.
- "Delve" and Friends: Perhaps the most infamous. "Delve" went from uncommon to appearing in a staggering percentage of AI text. See also: "leverage", "robust", "harness".
full directory: https://tropes.fyi/directory
ai;dr - Paste a link to any article. It will strip the AI fluff using our trope detection rules, then feed the skeleton to a lightweight llm which reverse-engineers it back to the prompt they likely* typed :D. 3000 words of slop → 30-word prompt + whatever unique insights (if any) survived the bloat. try it: https://tropes.fyi/aidr
AI Vetter - Paste a URL, get a verdict. The AI Vetter fetches article content, runs rule-based pattern matching against all 32 tropes and returns a score on the scale: Human → AI-assisted → Suspicious → Barely Legible → Pure AI Slop. try it: https://tropes.fyi/vetter
tropes.md - A single file you can drop into any AI assistant's system prompt so it avoids these patterns. Sorry for helping them become invisible, I'd just rather not see it.
"You're a hypocrite I'm sure you use AI for writing code!" — AI generated code is fine, it's one computer (LLM) talking to another (compiler). How else do you think AI is so good at writing code? They're speaking the same language! But prose? That's from human to human, it's sacred and meant for other people. Using AI for that is deceitful.
