No one writes a plot hole. Plot holes are emergent — each scene was locally sensible when you wrote it, and the contradiction lives in the space between scenes written months apart. That's why "read it again carefully" is such weak advice: your brain wrote both scenes, and it will happily re-supply the connective tissue the page never had.
Finding plot holes systematically means knowing what kinds exist, because each kind hides from a different search.
The four species of plot hole
Continuity holes are factual contradictions: the burned-down inn that hosts a scene two chapters later, the autumn that lasts five months, the dead horse that gets ridden in the finale. Mechanical, unambiguous, and the most screenshot-able by readers.
Causality holes break cause and effect: the villain's plan requires information he had no way to get; the siege is over but nothing on the page resolved it. These read as "wait, why did that work?"
Knowledge holes are the subtlest: a character acts on something she hasn't learned on the page. You, the author, knew the letter was forged — so it never occurred to you that Mara doesn't. Readers experience these as telepathy.
Motivation holes are characters bent to the plot's will: the cautious character who suddenly gambles everything because act three needs her in the castle. Technically not a contradiction; emotionally, the most damaging hole of all.
Each species needs a different net. Here are the four passes, in the order that wastes the least work.
Pass 1: The timeline audit (catches continuity)
Build — or update — a master timeline from the manuscript, not from memory: every dated event, every "two weeks later," every age claim, one line each, in story order. Most writers find their first contradiction before finishing the extraction. Travel times deserve special paranoia; fictional geography compresses whenever a deadline approaches.
This audit is brutal to do by hand at 100K words, which is exactly why it's the most automatable pass. If your facts already live in a structured story bible, a consistency checker can cross-reference each chapter against canon as you write — Creader's Guardian runs this continuously and flags contradictions with chapter references, rather than waiting for your revision week. See how it works.
Pass 2: The knowledge ledger (catches knowledge holes)
For each POV character and each major secret, trace: who knows it, and which scene taught them. Write the scene number down — "she obviously knows by now" is exactly the assumption being tested. Every action motivated by a secret gets checked against the ledger: did the character learn this before acting on it, on the page?
This pass is tedious and unglamorous and finds the holes beta readers describe as "something felt off but I can't say what."
Pass 3: The causality chain (catches causality holes)
For each major plot event, ask two questions: what on the page caused this? and what did this cause? An event with no on-page cause is a hole (or an unrevealed cause you forgot to reveal — check your foreshadowing threads). An event that causes nothing is dead weight that revision should cut anyway.
Villains' plans deserve a dedicated walk-through from the villain's POV: at each step, does he have the information and resources the plan requires, from inside his own knowledge ledger? Most thriller plot holes die in this one exercise.
Pass 4: The motivation read (catches the deadliest species)
One full read per major character, skipping every scene they're not in. Read only their scenes, in order, and ask at each decision: would this person, as established, do this? Out-of-character pivots stand out vividly when you remove the surrounding noise of other plotlines — and they're invisible in a normal front-to-back read, which is why this pass exists.
When you find one, the fix is rarely "change the action." It's usually "earn the action": add the pressure, the new information, or the prior crack in their caution that makes the gamble believable.
What to automate vs. what to read for
The honest division of labor: machines are better than you at passes 1 and 2 — cross-referencing thousands of facts without fatigue is precisely what AI tools built for fiction do well, and continuity checking is a solved problem when your canon is structured data. Passes 3 and 4 are judgment calls about human behavior; no checker should make them for you, though a good one can surface the evidence (every scene where the character appears, every mention of the secret) in seconds instead of an afternoon of ctrl-F.
Run the four passes at the end of each act rather than once at the end of the book — holes are exponentially cheaper to fix before more plot is built on top of them. The goal isn't a flawless manuscript; it's making sure the reader's trust, once earned, is never refunded.