AI & Writing

Why Writing Matters More in the Age of AI, Not Less

Quick Takeaways
  • 1.5+ billion users on ChatGPT, Claude, Gemini, and others send 4+ billion prompts each day
  • Prompt engineers earn $123,803 on average, with a 14% bonus for writing skills
  • AI models break down when trained on AI text; human writing is now rare and vital

The Writing Paradox

ChatGPT hit one million users in five days after its late 2022 launch. It was the fastest growth for any app. Instagram took 2.5 years. TikTok took nine months.

Today, AI writing tools serve over 1.5 billion users:

  • ChatGPT: 800 million weekly users, 2.5 billion daily prompts
  • Google Gemini: 450 million monthly users
  • Claude: 300 million monthly users, 70% yearly growth
  • Perplexity AI: 100+ million queries per month

This boom made a new field. A job that did not exist in late 2022 now pays $123,803 on average. LinkedIn lists over 5,000 prompt jobs. The field grows 31.6% per year through 2032.

Yet Google searches for "how to write better" held steady. People still want to write well, even with AI help.

The paradox: The tool built to help us write shows how much writing skill still matters.

What 4+ Billion Daily Prompts Reveal

A look at 100 prompt jobs shows writing skills come up the most. Jobs asking for "strong writing" paid $18,000 more. That is a 14% pay bump for writing skill.

Research

LLMs are "shockingly sensitive" to how well we write prompts. A vague prompt like "Summarize this" opens endless paths. A clear prompt like "Summarize this in 150 words for policy makers, focus on budget" steers the model the right way.

Four billion daily prompts mean a lot of writing practice. Every prompt is a small drill in clarity.

The Model Collapse Crisis

In July 2024, Nature showed that AI models "collapse" when trained on their own output. Tests on many model types found the same thing: "Careless use of AI-made content causes lasting defects."

The Data Exhaustion Timeline

  • Top-grade data: Used up before 2026
  • Best-case scaling: Enough through 2028
  • 100x overtraining: Already gone

AI now writes 50.3% of new web pages. And we can't yet spot all AI text to keep it out of training data.

The twist: Just as AI makes writing help cheap and everywhere, human writing becomes rare and vital for AI to keep working.

The 5 Writing Skills That Transfer to AI Prompt Engineering

Skill 1: Task Splitting (58% of job posts need this)

Breaking "write a thesis" into sections. In prompts: breaking "analyze this data" into steps.

Skill 2: Knowing the Reader (73% of job posts need this)

Shifting tone and depth for each reader. In prompts: telling the AI who will read it.

Skill 3: Working Within Limits (51% of job posts need this)

"Write exactly 100 words" maps to "give 3 bullet points, 50 words each."

Skill 4: Judging Output (45% of job posts need this)

Spotting strong vs. weak writing. Finding gaps. Knowing what to fix. Checking if AI output meets the goal.

Skill 5: Revising in Loops (89% of job posts name this)

Write, judge, revise, improve. Knowing how to make it better when results fall short.

The Data Points to Opportunity

The Evidence
  • 1.5+ billion users → Writing help is everywhere
  • 4+ billion daily prompts → Huge writing practice
  • $123,803 average pay → Writing skill pays well
  • 50.3% AI-made content → Human writing grows scarce
  • Model collapse shown → AI needs human thought

Who Wins in This Environment

  • Tech workers who can explain hard ideas in plain words
  • Those who see the math: Good writer × AI = Great results
  • Those who grasp that AI tools are free for all, but writing skill is still rare

The New Literacy

  • 20th century: Typing speed, good spelling, neat format (computers fixed this)
  • 21st century: Clear thought, precise writing, sharp judgment (AI needs these from us)

AI makes clear thought visible and prized. The tool sold as a writer's replacement depends deeply on real writers.

References

Shumailov, I., et al. (2024). AI models collapse when trained on recursively made data. Nature, 631, 755-759.

Schulhoff, S., et al. (2025). The prompt report: A systematic survey of prompting techniques.

Share