The Red Herring Problem: Why You Can't Judge Your Own Misdirection
Mystery writers occupy an uncomfortable position when revising their own manuscripts. You know who did it. You've always known who did it. That knowledge colors every sentence you read, making it nearly impossible to assess whether a red herring is doing genuine narrative work or simply taking up space with the faint hope that readers won't notice its hollowness.
The writer who plants a suspicious detail in chapter three and then spends the next eight chapters building a false case against an innocent suspect believes, with complete sincerity, that the misdirection is working. They can feel the tension. They know the moment when the red herring suspect should start looking guilty, and they can feel that beat arriving exactly on schedule. What they cannot feel is what a reader who doesn't know the ending will experience — whether the evidence feels genuinely incriminating or thin, whether the exoneration in chapter nineteen will feel earned or like the author simply changed the rules.
This is precisely where AI becomes a useful revision partner. Not because it possesses some mysterious objectivity, but because a language model can be prompted to read your material as a detective-minded stranger would — cataloguing evidence, flagging logical gaps, and simulating both the first-time and returning reader's experience. The prompts in this article are designed to let you interrogate your own misdirection from outside your own blind spots.
The Three Failure Modes of Red Herrings
Before you start prompting, you need to diagnose which problem you're actually solving. Red herrings fail in three distinct ways, and the fix for each is different.
The Toothless Herring
This is a suspect who feels suspicious because the narrative tells you they're suspicious — through other characters' dialogue, atmospheric description, or the protagonist's unease — but who has almost no actual evidentiary weight attached to them. Strip away the mood and the loaded glances, and there's nothing there. Readers who finish the book and think "wait, why was I ever worried about Marcus?" are experiencing the toothless herring. The fix is cumulative evidence, not atmosphere.
The Cheating Herring
This is the failure mode that genuinely damages reader trust. A cheating herring requires that established facts be contradicted at the reveal, or that the POV character withheld information they logically would have processed and shared. If your protagonist notices a detail in chapter six that would have cleared the red herring suspect, but the narrative simply doesn't follow that thread, you've cheated. The reader who re-reads the book will find the seam and feel the authorial thumb on the scale.
The Redundant Herring
Three suspects all feel suspicious for the same reason — they were near the scene, they had a prior relationship with the victim, they behaved evasively when questioned. The redundant herring doesn't misdirect; it simply creates noise. Readers stop tracking the individual cases and start pattern-matching against the genre's conventions. Worse, when one of the three turns out to be innocent, the reader feels nothing because they weren't tracking that specific suspect anyway.
Identifying which failure mode afflicts your manuscript before prompting means you can build prompts that target the actual problem rather than generating general feedback.
Prompt Set 1 — The Single Suspect Audit
These prompts are designed to run on the material you've pasted about a specific red-herring suspect. You'll want to include all scenes in which that suspect appears, any scenes in which they're discussed by other characters, and a brief summary of what the reader is meant to suspect them of doing.
I'm revising a mystery novel and need an audit of a single red-herring suspect. Below, I've pasted every scene in which [SUSPECT NAME] appears, plus every scene where other characters discuss or reference them. I've also included a brief summary of what crime or action the reader is meant to suspect this character of committing. Please do the following: 1. Create an inventory of every piece of concrete evidence, specific behavior, physical detail, or witnessed action that could reasonably make [SUSPECT NAME] look guilty — separate from any atmospheric description or other characters' opinions. 2. Rate each piece of evidence on a scale of 1–3: how independently suspicious is it if a reader encounters it without authorial framing? (1 = only suspicious because the narrative signals it; 2 = ambiguous but legitimately suspicious; 3 = a detective-minded reader would flag this independently.) 3. Estimate how many scenes feature [SUSPECT NAME] as a significant presence versus how many pieces of evidence you found. If the ratio feels disproportionate — lots of screen time, little evidentiary weight — flag that imbalance explicitly. 4. Identify any evidence plants that appear only after [SUSPECT NAME] has already been narratively signaled as suspicious. These may read as confirmation bias rather than independent suspicion-building. [PASTE ALL RELEVANT SCENES BELOW]
Run this prompt once per major red-herring suspect. What you're looking for is the ratio problem: a well-calibrated false lead should have at least three to four independently suspicious pieces of concrete evidence, not three pieces of atmosphere and one actual clue.
Prompt Set 2 — The Fair Play Stress Test
The fair play rule — the compact between mystery writer and reader that the solution can be deduced from available evidence — is easier to commit to in principle than to execute in practice. Writers almost always believe they've honored it. The following prompts simulate a detective-minded reader cataloguing your clue architecture before the reveal.
I need a fair play audit for my mystery novel. I'm going to paste the complete clue list — every piece of information revealed to the reader before the climax chapter — along with a brief description of the true solution (who committed the crime, how, and why). Please work through the following: 1. Acting as a detective-minded reader who has not been told the solution, identify which clues on the list would most plausibly point toward the true culprit. Flag which of these are genuinely available to a first-time reader versus which require hindsight to interpret correctly. 2. Identify any clues that, logically, should point toward the true culprit but are never presented to the reader before the climax. These are potential fair play violations — information the narrative withheld that would have allowed deduction. 3. Check whether the stated motive is adequately supported by information the reader receives before the climax, or whether the motive requires new information introduced at the reveal chapter. 4. Flag any clues that actively mislead toward the red herring suspects — and check whether each misleading clue has a plausible innocent explanation that a re-reading reader would find satisfying rather than evasive. 5. Render a fair play verdict: could a sufficiently attentive first-time reader have arrived at the correct solution from this clue list? What is the single weakest link in the fair play chain? SOLUTION SUMMARY: [PASTE YOUR SUMMARY] CLUE LIST: [PASTE EVERY CLUE REVEALED BEFORE THE CLIMAX, WITH CHAPTER REFERENCES]
The fair play stress test is most useful when you're honest about your clue list. Include the small details, not just the ones you consider major plot points. A throwaway line in chapter four about the victim's calendar, a description of which car was parked where — these are the granular details that distinguish a genuinely fair mystery from one that merely gestures at fairness.
Prompt Set 3 — The Re-Read Reversal Simulation
A re-read of a great mystery novel is often more pleasurable than the first read, because what initially looked like innocent detail now glows with retroactive significance. A re-read of a poorly calibrated mystery is the opposite: the seams show, the withheld information stings, and the clever-seeming misdirection begins to look like authorial manipulation rather than genuine craft.
This prompt set asks AI to simulate that re-read experience against specific scenes.
I need you to simulate two readings of the same scenes from my mystery novel, back to back. I'm going to paste a set of scenes, a summary of the true solution (killer, method, motive), and a note on which red herring suspects these scenes are meant to implicate. FIRST READING SIMULATION: Read these scenes as a first-time reader who does not know the solution. Describe what a careful reader would notice, what they would probably suspect, and what they would likely overlook. Note any moments where the misdirection toward the red herring feels genuinely earned versus moments where it requires the reader to miss something that feels too conspicuous to miss. SECOND READING SIMULATION: Now read the same scenes as a reader who has finished the book and knows the solution. For each scene, identify: — Details that retroactively feel like elegant foreshadowing (the clue was there, and it's satisfying to recognize it) — Details that retroactively feel like authorial sleight of hand (the narrative actively guided the reader away from something they logically should have noticed) — Any moments where the POV character's stated thoughts or reactions are inconsistent with what they should logically know, given what they witness in this scene Flag the second category — authorial sleight of hand — as priority revision notes. These are the moments most likely to make a re-reading reader feel cheated rather than impressed. SOLUTION SUMMARY: [PASTE] RED HERRING SUSPECTS IN THESE SCENES: [PASTE] SCENES: [PASTE]
The distinction between elegant foreshadowing and authorial sleight of hand is the core diagnostic this prompt is designed to surface. Both involve guiding the reader's attention — but foreshadowing works because the clue is genuinely present and the reader simply didn't weight it correctly, while sleight of hand works because the narrative actively interfered with the reader's ability to process what's in front of them.
The Calibration Workflow: Running All Three in Sequence
These three prompt sets are most powerful when used in sequence during a dedicated revision pass, not run ad hoc whenever something feels off.
Stage One: Suspect-Level Audits
Start by running the Single Suspect Audit for every red herring character who receives significant page time — not just the main false lead, but secondary suspects who accumulate suspicion across the book. The goal at this stage is to ensure that each false lead has evidentiary weight proportionate to their narrative presence. When the audit reveals a toothless herring, you have a targeted revision task: build in two to three additional concrete evidence plants before you move to the next stage.
Stage Two: The Full Clue Inventory
Before running the Fair Play Stress Test, compile your clue list manually. This is not optional. The act of writing out every piece of evidence revealed before the climax — chapter and scene reference included — often surfaces fair play problems before you run a single prompt. Writers who have been living inside their manuscripts for months routinely discover, at this stage, that a clue they felt certain was in the manuscript is actually in their notes or in their mental model of the story but never made it to the page. Run the Fair Play Stress Test against the clue list once it's complete, then use the feedback to identify any motive or method gaps that require new material.
Stage Three: Scene-Level Re-Read Simulations
Once the clue architecture is sound, run the Re-Read Reversal Simulation against your highest-stakes scenes: the scenes that most heavily implicate the red herring suspects, and any scene where your POV character witnesses something that would logically point toward the true culprit. These are the scenes most vulnerable to authorial sleight of hand, and they're the ones a re-reading critic will examine most carefully.
- Run the suspect audit first — fix evidentiary gaps before worrying about fair play
- Compile your clue list manually before prompting the fair play test
- Save the re-read simulation for after structural revisions are complete
- Treat the AI's "priority revision notes" as hypotheses, not verdicts — test them against your own re-read of the flagged scenes
- Repeat the fair play test after any significant revision that adds new clues or removes existing ones
What AI Can and Can't Catch
These prompts are excellent at structural and logical analysis: evidence ratios, fair play gaps, POV inconsistencies, clue timing. They are less reliable at assessing the emotional texture of misdirection — whether a red herring feels genuinely threatening as a character, whether the prose rhythm at the reveal lands correctly, whether the final exoneration produces the right mix of relief and retrospective irony.
That emotional calibration still requires human readers, ideally beta readers who don't know the solution. What the prompts above give you is a structurally sound foundation: a false lead with genuine evidentiary weight, a clue architecture that honors the compact with the reader, and a set of scenes that will reward re-reading rather than punishing it. That foundation is the prerequisite for everything else — including the atmospheric work, the character psychology, and the prose rhythm that actually make a mystery novel feel inevitable rather than merely logical.
Mystery writing is built on a paradox: you must mislead the reader completely while deceiving them not at all. The red herring problem is the sharpest expression of that paradox. Getting the calibration right is the work, and it's work that benefits enormously from a reader who doesn't already know the ending.
No comments yet. Be the first to comment!