---
id: con-ai-writing-tells
schema: concept/1
layer: garden
status: emerging
name: AI writing tells
aka: ["AI slop", "sounds like ChatGPT", "AI tropes", "generic AI voice", "AI-isms"]
summary: >
  The recognisable habits of AI-generated text, such as stock phrases, tidy groups of three, empty
  enthusiasm and vague claims, that make writing feel generic and erode readers' trust.
first_seen: unknown
relations:
  builds_on: [con-large-language-model]
  evolved_from: []
  related: []
  parallels:
    - idea: "corporate boilerplate"
      field: "business writing"
      note: "Phrases that fill space without saying anything specific, which readers learn to skim past."
    - idea: "spotting a form letter"
      field: "communication"
implementations:
  - { provider: every, name: "draft-review-kit guardrails reviewer", note: "Checks a draft for AI-sounding phrasing and evidence gaps; see fld-every-draft-review-kit." }
  - { provider: peter-yang, name: "no-ai-slop skill", note: "See fld-no-ai-slop." }
competencies: [cmp-check-the-result, cmp-use-responsibly]
origin:
  kind: ai-proposed
  by: "seed agent 2026-10-16"
  rationale: "A fast-growing reader reflex; editing for specifics is both a quality and a trust issue for anything sent under your name."
---

Readers have learned to spot AI-written text. The giveaways vary by tool and change over time, but common ones include stock openers and closers, lots of grand but vague words, lists that always come in threes, sentences built around a contrast that wasn't needed, cheerful enthusiasm about everything, and paragraphs that summarise what was just said.

The bigger problem underneath is generic content. AI fills gaps with safe, average phrasing because it doesn't know your specifics. When a reader senses that, they trust the message less, even if it's accurate.

Fixing it is mostly editing for specifics. Replace general claims with the real number, name or example. Cut sentences that could appear in anyone's email. Read it aloud and change anything you would never say. Giving the AI samples of your own writing up front prevents much of this.

## Example

An AI-drafted fundraising appeal opens with a sweeping line about "making a difference in today's world." The writer replaces it with the real story of one family the program helped last month, cuts the closing summary, and the appeal reads like her.

## How it connects

- A side effect of how a **large language model** defaults to likely phrasing.
- **Examples in prompts** are the best prevention.
- Reading for tells is a form of **verification** for voice and substance.

## Try it

Take a recent AI draft and underline every sentence that could appear unchanged in someone else's email. Rewrite one of them with a detail only you know.