Your intellectual fly is open when you use an LLM to author a post (2025) | AI Paper Digest

TL;DR Highlight

Posts written by LLMs are easy to spot, and readers stop reading them.

Who Should Read

A developer who lets an LLM write blog posts, LinkedIn posts, or internal announcements

Core Mechanics

LLM prose gives itself away
- The author lists emojis, one-sentence paragraphs, "it's not just... but also" phrasing and em-dashes as tells.
- He says anyone who has seen a bit of LLM text can't miss them.
The problem is trust, not embarrassment
- When the voice doesn't sound like the author, he can't tell if the content is real or generated fanfic.
- So he stops reading, and he says many others do too.
LLMs are great editors
- He finds them useful for brainstorming, comprehending text and editing.
- You can ignore their suggestions freely, unlike with a human editor.
LLMs can't be you
- He says they are lousy writers and, most importantly, not you.
- Real content wrapped in LLM style makes people skim or doubt your authenticity.
Write your own content

Community Reactions

Writing is thinking, one commenter argued
- They said their views often changed substantially while writing.
- They argued you can't outsource understanding, especially for design docs.
Disclosure is the real issue, a counterargument
- One commenter said that even if LLM writing improves, undisclosed use would still feel misleading.
- Another said they distrust text the writer doesn't own and endorse.
Tuned LLMs can be decent, per one experience
- A commenter said giving writing samples and skill files plus editing gives pretty good results.
- They said it works best for procedural writing.
Internal docs are flooded with LLM text, per one experience
- A commenter said launch emails got longer and thinner, and almost nobody clicked the links.
- They also described AI-written project plans with bizarre timelines.

How to Apply

Scan your own writing for LLM tells
- Check your recent announcement or blog draft for emojis, one-line paragraphs, "not just... but" and em-dashes.
- Rewrite any you find yourself.
Limit the LLM to the editor role
- Write the first draft yourself and ask the LLM only to point out problems.
- Apply only the suggestions you accept, by hand.
Disclose where you used an LLM
- Add a one-line note at the end for parts like summaries.
- Include a link to the original source.

Try It

copy & run

# Paste into Claude or ChatGPT as-is (the article's method turned into a prompt)

You are an editor, not a ghostwriter. Do not rewrite the draft below. Only point out problems.

[Draft]
{{your own draft}}

[Instructions]
1. List, by number, places where the logic breaks or evidence is weak.
2. Point out where a reader would stop reading.
3. Do not write sentences for me.

Next: revise by hand based on the notes, paste it again, and run the same check.

Terminology

LLM tellsHabits that repeat in LLM writing and give it away, like emojis, one-line paragraphs and em-dashes.

em-dashThe long dash (—) used mid-sentence. It has become a symbol of LLM writing.

toupee fallacyBelieving all toupees are obvious because you only notice the bad ones. A commenter used it to argue the author never notices well-done LLM text.

Related Papers

Related Resources

https://bcantrill.dtrace.org/2025/12/05/your-intellectual-fly-is-open/