Here’s an uncomfortable truth I learned after a year of testing AI writing tools: the tool is almost never the problem. The prompt is.
Type “help me write my novel” into ChatGPT or Claude and you’ll get beige, generic mush — the average of every writing guide ever published. But ask it to attack your premise the way an agent’s assistant would at 11pm with fifty more queries in the pile? Now you have something no writing group meets often enough to give you: a ruthless, tireless critique partner.
The mindset shift: assistant, not author
The novelists getting real value from AI aren’t asking it to write. They’re asking it to interrogate. The words stay yours — the pressure-testing gets outsourced. Three examples of the difference:
Weak: “Give me feedback on my chapter.”
Strong: “Give me a triage report ONLY: the 3 biggest problems in order of damage done, each with a one-line diagnosis. No line edits, no praise sandwich.”
Weak: “Help me with my villain.”
Strong: “Write my antagonist’s private justification in first person — the 2-minute speech where they’re the hero. Make me agree with at least one sentence.”
Weak: “Is my plot good?”
Strong: “Rewrite my chapter summary as a ‘therefore/but’ chain. Flag every place events connect with ‘and then’ — those are my dead spots.”
Same tool. Wildly different output. The pattern: specific role, specific format, specific constraint, and your actual material in the brackets.
Five prompts to steal right now
1. The Premise Stress Test. “Here is my premise: [PREMISE]. Attack it: Where is it derivative? Where does the conflict fizzle? What’s the most obvious version a lazy writer would produce? Then show me the version that avoids all three.”
2. The Wound Interview. “Interview my protagonist about the worst day of their life — one question at a time, in their voice. Push past deflection. After 8 exchanges, summarize the wound, the lie they believe, and the mask they wear.”
3. The Midpoint Fixer. “My middle sags. Here’s my outline: [PASTE]. Propose 3 midpoint reversals that turn my protagonist from reactive to proactive, each raising stakes in a different dimension — physical, emotional, moral.”
4. The Echo Hunter. “Scan this chapter for my crutches: repeated words, recycled gestures (nodding, sighing, smiling), duplicate sentence structures, and my apparent favorite word. Rank by frequency.”
5. Chapter One, Draft Ten. “Answer as an agent’s assistant with 50 more submissions in the pile tonight: Where did you stop reading and why? What would have kept you? Be brutal, be specific.”
The full toolkit
Those five come from The Novelist’s Prompt Vault — 100 prompts I’ve refined across the full life of a novel: premise, structure, character, worldbuilding, scene craft, revision, querying, and launch. Every prompt works with ChatGPT, Claude, or Gemini, and every one is built on the same principle: AI pressure-tests, you write.
If you’re more of a planner, it pairs with The Novel Architect Workbook — a fillable PDF that walks your novel from logline to revision checklist, with a matching prompt in the Vault to stress-test every section you fill in.
Both are launch-priced this fortnight — code LAUNCH20 takes 20% off at checkout.
One honest caveat
AI output is raw material, not finished prose. Readers can tell. Agents definitely can. Use these tools to think harder, cut deeper, and finish faster — then make every sentence yours. That’s the whole method.
Leave a comment