Wikipedia: Signs of AI Writing
Wikipedia: Signs of AI Writing
[1] DANIEL:
Please write this page as a list in markdown for us to study. Signs of AI writing at Wikipedia
[2] GROK:
Here is the content from Wikipedia:Signs of AI writing (the official Wikipedia page dedicated to this topic) reformatted as a clear, structured markdown list for study purposes. This draws from the page's key sections and common indicators as described across reliable sources referencing it. Note that the page emphasizes these are descriptive observations (not strict rules), not all matches mean AI was used (since LLMs train on human/Wikipedia text), and some signs are Wikipedia-specific.
Important Preface from the Page
- LLMs tend to have an identifiable writing style.
- This is a field guide to help detect undisclosed AI-generated content on Wikipedia.
- Not all text with these indicators is AI-generated.
- The list is descriptive, not prescriptive — observations from thousands of instances.
- AI detectors are unreliable; human judgment is key.
- Check earlier revisions if categories/templates seem broken (reviewers sometimes remove them).
Main Categories of Signs
1. Language and Tone
- Overemphasis on importance/symbolism/legacy (inflated claims of significance).
- Examples: "pivotal moment", "enduring legacy", "rich cultural heritage", "a broader movement", "iconic", "seminal", "groundbreaking".
- Promotional or booster language (touristy/puffery tone).
- Examples: "must-see", "breathtaking", "stunning natural beauty", "must-visit destination", "hidden gem".
- Editorializing or non-neutral commentary.
- Examples: "It is important to note/remember/consider", "it is worth noting", "no discussion would be complete without", "it should be emphasized".
- Weasel words / vague attribution.
- Examples: "Industry reports", "Observers have cited", "Some critics argue", overgeneralizing perspectives.
- Overused verbs/adjectives (tics from training data).
- Verbs: "delve" (esp. "delves"), "underscore", "highlight", "showcase". - Adjectives: "swift", "meticulous", "adept", positive/generic descriptors that sand down nuance.
- Formulaic rhetorical patterns.
- Heavy use of em dashes (—) for interruptions/emphasis. - Rule of three (tricolons): saying things in triplets. - Negative parallelisms: "It's not X, it's Y" (e.g., "It's not a product launch—it's a paradigm shift").
2. Style and Structure
- Excessive bullet/numbered lists (even where paragraphs would fit better).
- Format: "Term: Definition" style in every item.
- Emojis in headers or body text (unusual for encyclopedic tone).
- Superficial analyses or generic positive framing.
- Regression to the mean: replaces specific/nuanced facts with generic, common, positive descriptions.
3. Communication Intended for the User (leftover from
prompts)
- Direct address or sycophantic/cheery replies left in.
- Prompt fragments or instructions accidentally included.
- "As an AI language model..."-style disclaimers (rare now but seen early on).
4. Markup and Formatting (Wikipedia-specific)
- Hallucinated/non-existent templates (esp. fake infoboxes or parameters).
- Appear as red links; non-existent params in real templates do nothing.
- Markdown-formatted text instead of wikitext (e.g., bold instead of '''bold''').
- Unusual or broken categories/categories deleted by reviewers.
- Hallucinated shortcuts or explained shortcuts in a non-standard way.
5. Citations and Sourcing
- Fake/fictitious references (hallucinated sources).
- Citations unrelated to the topic or mismatched.
- Overly detailed minor media mentions to inflate notability.
Additional Notes for Study/Use
- Many signs overlap with poor writing in general (e.g., promotional tone violates Wikipedia's neutral point of view anyway).
- For new pages: long, competent-looking prose created suddenly by new/low-edit accounts is suspicious.
- Wikipedia has a speedy deletion criterion (G15) for obvious LLM pages without human review.
- Related resources: WikiProject AI Cleanup, template {{AI-generated}}, noticeboard for reporting.
This captures the essence in list form. The original page is much longer with specific diff examples — check https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing directly for the full, up-to-date details with real Wikipedia excerpts. Useful for studying how LLMs deviate from encyclopedic style!