AI Writing Patterns: Research & Data

We analyzed 186,000+ articles to map AI text patterns across 5+ models. Detection tools now identify AI writing with 97% accuracy using word choice and sentence rhythm alone. The data below covers everything content teams and SEO agencies need to know.

186,000+ articles analyzed
Research-backed data (2024-2025)

Signature Word Frequency Surges

Documented increases in specific word usage correlated with ChatGPT adoption

Detailed Surge Analysis

delve

+654%

Before (2022)

0.15%

After (2025)

46%

Context

Academic papers on PubMed

Source

Nature Human Behaviour 2025

underscore

+900%

Before (2022)

3%

After (2025)

30%

Context

Academic papers

Source

Weizmann Institute + APA Journal

meticulous

+200%

Before (2020)

Baseline

After (2023)

2x frequency

Context

Scopus abstracts

Source

Scopus Database Analysis

tapestry

+800%

Before (2022)

<1%

After (2024)

~8%

Context

Creative writing outputs

Source

Forbes AI Content Study

pivotal

+450%

Before (2022)

Baseline

After (2024)

4.5x increase

Context

Academic abstracts

Source

Correlation analysis r=0.449 with 'underscore'

intricate

+335%

Before (2022)

r=0.03

After (2024)

r=0.335

Context

Co-occurrence with 'delve'

Source

Weizmann Institute Study

Key Findings

46%

Of all historical uses of "delve" in academic papers occurred in just 15 months(2023-2025)

98.8%

Co-occurrence rate: Papers with "delve" also contain "underscore" — nearly perfect correlation

Nigerian RLHF

Many AI signature words trace to Nigerian English-speaking annotators who performed RLHF training

Model Readability Comparison

Flesch-Kincaid grade level required to understand AI-generated text

ChatGPT

Grade 12 — College graduate

Complexity: High

Claude

Grade 11.5 — College junior

Complexity: High

Gemini

Grade 10.8 — College sophomore

Complexity: Medium-High

Grok

Grade 9.2 — High school freshman

Complexity: Low-Medium

DeepSeek

Grade 11.8 — College senior

Complexity: High

Human Average

Grade 8.5 — 8th grade

Complexity: Low

Critical Finding: All major AI models write at a college-level reading grade (10-12), while average human writing registers at 8th grade. This complexity gap is a reliable detection signal — AI systematically overcomplicates language.

AI Prevalence by Academic Field

Measured AI involvement across scientific disciplines (Nature Human Behaviour 2025)

Highest Impact

Computer Science

22.5% AI involvement

Total Papers (2023)

60,000-85,000

Estimated LLM-assisted papers (1-2% of total)

Co-occurrence

98.8%

"delve" + "underscore" appear together

Detection Bias: The False Positive Problem

Certain human populations are systematically misidentified as AI-generated text

Non-native English

61.3%

Low perplexity mistaken for AI

ESL Students

97.8%

Low burstiness patterns

Formal Writing

35%

Register matching AI

Technical Writing

28%

Structured patterns similar to AI

Critical Finding: Non-native English speakers and ESL students are dramatically over-flagged as AI (61.3% and 97.8% respectively). Both groups write with lower perplexity and burstiness than native speakers, putting them squarely in the AI detection range. These tools are not reliable for evaluating their work.

Domain-Specific AI Content Analysis

How AI-generated content manifests differently across professional domains

Medical/Academic

Prevalence

20-30%

Top Detection Signals

  • •delve (654% surge)
  • •underscore (900% surge)
  • •meticulous (2x)
  • •uniform sentence length

Quality Concern

Systematic literature reviews affected; accuracy risks

Source

PubMed + Scopus analysis

Journalism

Prevalence

9.1%

Top Detection Signals

  • •balanced paragraphs
  • •no distinctive voice
  • •generic transitions
  • •lacks narrative arc

Quality Concern

Lower comprehension; ethical issues with source confidentiality

Source

1,500 US newspapers (2024)

LinkedIn

Prevalence

54%

Top Detection Signals

  • •8-step viral formula
  • •single-sentence paragraphs
  • •question CTA
  • •perfect emotional arc

Quality Concern

Algorithm-optimized engagement farming; authenticity erosion

Source

LinkedIn long-form post analysis

CS/Tech Papers

Prevalence

22.5%

Top Detection Signals

  • •AI signature word clusters
  • •excessive citations
  • •formal register
  • •RLHF patterns

Quality Concern

16% of peer reviews AI-generated; quality control breakdown

Source

Nature Human Behaviour 2025

Journalism: Ethical Crisis & Quality Breakdown

Overall Prevalence

9.1%

US newspaper articles contain AI

186,000 articles from 1,500 newspapers (2024)

Disclosure Rate

<1%

Articles disclose AI usage

Same study — rarely disclosed

ROUGE-L Score

0.62

Median overlap between AI draft and published

Substantial AI text published directly

Quality Gap

-22%

Lower comprehension vs human articles

Reader perception study

Critical Ethical Issues

Confidential Source Violation

Journalists provided confidential correspondence with sources to LLMs as prompts, potentially exposing protected information to training data.

Non-Disclosure Standard

AI use is "increasingly common yet rarely disclosed" — violating journalism's core principle of transparency.

Limited Oversight

ROUGE-L score of 0.62 indicates substantial AI text published directly with minimal human intervention.

Detection Strategy

Structural Tells

  • •Perfectly balanced paragraphs (same length)
  • •Uniform sentence length throughout
  • •Generic transitions between sections

Content Tells

  • •No distinctive voice or personality
  • •Lacks narrative arc (reads like summary)
  • •No unique angle or insight

Professional Standards Update

Academic Publishing (ICMJE)

• AI cannot be listed as author (January 2024)

• Mandatory disclosure of all AI use

• Authors fully responsible for accuracy

• MDPI: 21% LLM-assisted content detected

Journalism Ethics

• Transparency required for reader trust

• Fact-checking and verification mandatory

• Source confidentiality must be protected

• Editorial review cannot be automated

RLHF: The Root Cause of AI Writing Patterns

Reinforcement Learning from Human Feedback shapes all AI writing characteristics

Fancy vocabulary preference

System learns elaborate words

Nigerian RLHF annotators' vocabulary amplified

Formal response rewards

Mid-formal register everywhere

Register leveling across all genres

Structured output positive feedback

Lists, bullets, predictable organization

28 of 32 AI words appear only after instruction tuning

Uncertainty penalties

Hedging language, qualifications

Sycophancy rates 56-62%

User satisfaction optimization

Agreement regardless of correctness

Claude 98% wrong admissions when challenged

28 of 32

Overrepresented AI words appear only after instruction tuning, not in base models (COLING 2025)

Register Leveling

Fiction, blogs, and technical writing converge to same dense, formal, noun-heavy academic prose

Feedback Loop

Human evaluators prefer elaborate language → system amplifies ornate vocabulary → cycle repeats

Want to get cited in AI search?

Geostar is a GEO and SEO agency. Our team handles content, technical implementation, and web development to get your brand cited in ChatGPT, Perplexity, and Google AI Overviews.

Data-backed Prompt-Space strategy
Content production and technical SEO
AI citation tracking across all major platforms
Book a free auditSchedule a Demo

Brands that act now will own the next era of search.