SOURCES.md

# Sources

Every marker in [`markers.json`](markers.json) carries `sources` ids that map to the
list below. This catalogue is a distillation of public work; credit belongs to the
authors and editors cited here.

## Primary reference

- **Wikipedia: Signs of AI writing** (`wikipedia_aisigns`) — the most detailed public
  field guide, maintained by volunteer editors with real examples.
  https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
- **Wikipedia: WikiProject AI Cleanup** (`wikipedia_aicleanup`) — the project behind
  the guide; also the source of the "check the edit date before you conclude AI"
  discipline (text predating 30 Nov 2022 is almost never LLM-generated).
  https://en.wikipedia.org/wiki/Wikipedia:WikiProject_AI_Cleanup

Wikipedia text is licensed CC BY-SA.

## How detectors work / accuracy

- **Kompozy — AI content detection in 2026** (`kompozy`).
  https://kompozy.io/ai-content/content-detection
- **TextSight — How AI Detectors Work (2026)** (`textsight_how`).
  https://www.textsight.ai/how-ai-detectors-work/
- **eyesift — Perplexity and Burstiness in AI Detection** (`eyesift_pb`).
  https://www.eyesift.com/blog/perplexity-and-burstiness-ai-detection/
- **HumanizeMyAI — Are AI Detectors Accurate? 2026 benchmarks** (`humanizemyai`).
  https://humanizemy.ai/ai-detector-accuracy
- **thesify — How Do Professors Detect AI in 2026?** (`thesify`).
  https://www.thesify.ai/blog/how-professors-detect-ai-writing-2026-guide
- **Liang et al., "GPT detectors are biased against non-native English writers,"
  *Patterns* (2023)** (`stanford_toefl`) — the key false-positive study: 61.3% of
  genuine TOEFL essays flagged as AI vs 5.1% for native-speaker essays.
  https://doi.org/10.1016/j.patter.2023.100779

## Perplexity / burstiness and remediation

- **Leap AI — Perplexity vs Burstiness** (`leap_pb`).
  https://www.tryleap.ai/learn/perplexity-vs-burstiness
- **Vortenza — Perplexity and Burstiness 2026 Guide** (`vortenza_pb`).
  https://www.vortenza.com/guides/perplexity-burstiness-ai-writing

## Overused vocabulary and phrase lists

- **Alston Antony — 300+ Overused ChatGPT Words (2026)** (`alston`).
  https://alstonantony.com/ai-seo/avoid-chatgpt-words-phrases-seo/
- **Walter Writes AI — Most Common ChatGPT Words to Avoid in 2026** (`walterwrites`).
  https://walterwrites.ai/most-common-chatgpt-words-to-avoid/
- **AI Phrase Finder — 100 Common ChatGPT Phrases** (`aiphrasefinder`).
  https://aiphrasefinder.com/common-chatgpt-phrases/
- **AI Free Forever — How to Tell If Something Was Written by ChatGPT** (`aifreeforever`).
  https://aifreeforever.com/blog/how-to-tell-if-something-was-written-by-chatgpt

## Note on secondary sources

Several sources above are commercial blogs (some sell "humanizer" products). They are
useful as compilations of observed patterns, but their accuracy claims and any product
benchmarks should be read with that conflict of interest in mind. Where a claim is
load-bearing (false-positive rates, the perplexity/burstiness definitions), it is
corroborated by the peer-reviewed or primary source cited alongside it.