AI-Enhanced Series: AI Shortcomings

In the early days of ChatGPT, I ran an experiment.

I tried to write a novel with it. A 21-year-old woman discovers an interdimensional doorway. She starts naive — curious, uncertain, in over her head. I gave the AI the premise and watched what happened with no training, no voice document, no constraints. Just raw generation.

By the time it was done with her, she was destroying the universe.

Not a metaphor. Literally destroying the universe. I had written a character study about a young woman encountering something she didn’t understand, and the AI had turned it into a cosmic apocalypse. The drift was so complete that the character in the final chapters had no relationship to the character I’d started with. She’d been optimized out of existence and replaced with whoever the training data thought a protagonist in this situation should become.

That experiment introduced me to every failure mode this guide covers. Drift — the character becomes someone else across sessions because AI has no persistent memory of who she was. Hallucination — plot elements appear that were never established. Escalation — the AI defaults to bigger, more dramatic, more catastrophic because that’s what stories do in its training data, regardless of your intentions. Voice collapse — by the end it sounds like every other AI novel because your specific story got smoothed into statistical averages.

The most dangerous thing about all of it: the AI sounded completely confident the entire time. It didn’t hesitate. It didn’t flag uncertainty. It stated each invented plot development with the same assured tone it would use for something you’d actually asked for. Your brain trusts confident speakers. AI exploits that trust without meaning to.

This guide covers everything AI does poorly. Not to discourage you from using it. To help you use it with eyes open. Writers who only understand AI’s strengths produce work riddled with errors they never catch. Writers who understand its weaknesses use the tool without being used by it.

The Failure Modes Nobody Talks About

I tested AI extensively. About a third of academic citations it generates lead nowhere. They look perfect — proper formatting, plausible author names, realistic journal titles — but the sources don’t exist. The context drift is gradual. The damage is cumulative. You might not notice until your manuscript is riddled with inconsistencies.

Learn All the Failure Modes →

The Tells That Give You Away

AI has favorite words. “Delve” appears constantly. So do “tapestry,” “nuanced,” “landscape,” “multifaceted,” “leverage,” “navigate,” and “robust.” Real humans almost never use these words this frequently. AI uses them all the time.

Get the Complete Cleanup Checklist →

What’s Inside This Guide

• The Confident Liar — Why AI’s delivery means nothing about accuracy and how to protect yourself
• Hallucinations — Invented sources, fake details, nonexistent books, and the compounding problem
• Context Drift — Signs, causes, and defenses for long conversations
• The Echo Chamber — Getting genuine critique from a tool that defaults to agreement
• The Common Name Problem — Why every Chinese character gets named Wei Chen and how to fix it
• The Common Word Problem — AI’s favorite words and how to spot them
• AI Language Patterns — Transitions, hedges, intensifiers, formal vocabulary, em dashes, and more
• The Redundancy Problem — How AI repeats itself and how to cut the bloat
• The Nuance Problem — Irony, sarcasm, metaphors, and cultural context AI misses
• The Verbosity Problem — Why AI uses twice as many words as necessary
• The Helpful Assistant Voice — The tone that bleeds into everything
• The Same-Voice Dialogue Problem — Why all characters sound alike

Plus chapters on the emotion-telling problem, happy ending default, complexity avoidance, list addiction, arbitrary structure, when AI fails completely, the complete cleanup checklist, verification strategies, and using AI to fix AI.

Yes, I Need This →

Using AI to Catch AI’s Mistakes

This sounds paradoxical. If AI makes errors, how can AI find them?

AI generating content and AI evaluating content are different tasks that trigger different patterns. Ask it to identify AI-typical language in a passage and it will flag many of the patterns covered in this guide. Ask it to check for redundancy and it spots where it repeated itself, because redundancy is a pattern recognition task. Ask it to evaluate verbosity and it identifies its own filler and suggests tighter alternatives.

The feedback loop combines generation and analysis. AI generates. AI analyzes. You direct. AI revises. Each pass improves the output. It isn’t perfect. But used well, with human oversight, it makes AI-assisted writing better than either tool or human working alone.

113 Books. Months of AI Testing. One Guide.

I’m Richard Lowe. I’ve published over 113 books and ghostwritten dozens more. My ghostwriting clients have secured over $30 million in venture capital, landed traditional publishing deals, and hit bestseller lists.

I’m also AuDHD. My brain doesn’t accept “this is just how it works” as an answer. When AI tools emerged, I spent months testing capabilities and, more importantly, limitations. My twenty years as Director of Computer Operations at Trader Joe’s taught me that understanding failure modes is essential to building reliable systems. You can’t trust a system until you know how it breaks. The same principle applies here.

The experiment with the universe-destroying protagonist was the beginning. This guide is everything I’ve learned since about where AI fails and what to do about it.

Questions

The woman with the interdimensional doorway became a universe-destroyer because I handed an untrained tool a premise and stepped back. Every AI-assisted draft you publish without understanding these failure modes carries the same risk — not necessarily at that scale, but the same mechanism. The confident tone. The gradual drift. The invented details that look real. The voice that sounds like everyone else’s AI because nobody told it to sound like you.

Know the failures. Catch the errors. Use AI without being used by it.

Part of the AI Writer’s Library Series. See also: AI Writing Partner Handbook | Revisions Handbook

P.S. — Still scrolling? If you’re using AI for writing and haven’t studied its failure modes, you’re trusting a confident speaker who doesn’t know when it’s wrong. This free guide takes an hour to read and will save you from errors you’d never catch otherwise.