Keep a reviewed exception list for automated fiction checks, with the exact term or construction, its intended use, and the reason it is acceptable. Apply exceptions narrowly and revisit them when the manuscript changes. The list should prevent repeated false alarms without hiding genuine errors or allowing a tool to decide character voice and story meaning on its own.
Separate a detected pattern from an editorial error
An automated check identifies a pattern according to its rules or model. That pattern may be an actual mistake, a stylistic preference, or an intentional feature of the story. A character's nonstandard grammar, an invented place name, or a repeated phrase used for emphasis should not be accepted or rejected merely because the tool flags it. Read the passage and identify what function the language serves.
For example, a character may consistently use a shortened expression that a grammar checker expands. The editorial question is whether that usage supports a clear and intentional voice, not whether the expanded version sounds more formal. Record the decision with enough context to distinguish the character's dialogue from narration. A blanket instruction to ignore the phrase everywhere could hide an accidental error in a different part of the manuscript.
Define exceptions with boundaries
For each accepted exception, record the exact spelling or pattern, the character or context where it applies, and an example from the approved draft. Include a note about cases that should still be flagged. An invented city name may be accepted, while a near-identical misspelling remains an error. A repeated line may be intentional at two important moments but accidental in a third location.
Keep the list small enough to understand and maintain. Do not add every dismissed suggestion without explanation. Some alerts are dismissed because the tool misunderstood one sentence, not because the underlying pattern should always be ignored. Distinguish a local dismissal from a reusable exception. This prevents the exception list from becoming a growing collection of permissions to overlook problems the author has not actually reviewed.

Preserve the reason behind voice decisions
Record why a character's language differs from the surrounding prose. The reason might involve register, rhythm, a deliberate catchphrase, or the narrative perspective. Avoid reducing a character to a mechanical list of quirks that must appear in every line. The exception record supports an editorial choice; it does not prescribe repetitive dialogue or replace the need to read the conversation as a scene.
If several editors are involved, share the approved examples and the limits of the decision. Otherwise, one editor may restore a feature that another repeatedly removes. Keep uncertainty visible when the choice has not been settled. A tool-generated suggestion should not become the house style merely because it was accepted once during an early pass. Resolve important voice questions deliberately and update the reference when the author changes direction.
Test exceptions against genuine mistakes
After adding an exception, check whether the automation still catches nearby errors you care about. Use a copy of representative passages rather than experimenting on the only manuscript file. For an invented name, verify that the correct spelling is accepted while a transposed letter is still visible through your review process. If the tool's exception control is too broad, a manual reference may be safer than suppressing the entire category.
Inspect the results after a tool update or a major configuration change. Do not assume that an old exception behaves identically forever. Keep the application and settings recorded with the review pass when practical. The goal is not to guarantee that every error will be detected, but to understand what the automated pass contributes and where human checking remains necessary for the manuscript's particular language choices.

Reconcile the list with the final editorial pass
Review the exception list against the current draft before final approval. Remove or retire entries for characters, names, or stylistic choices that no longer appear. Revisit exceptions that have spread beyond their intended context. A phrase originally approved in one character's dialogue may have entered the narrator's voice through later revisions, and the earlier decision does not automatically justify that new use.
Run the automated checks as one part of the final process, then read the manuscript for meaning, continuity, and voice. Keep the resulting issue list separate from the accepted exceptions so unresolved alerts remain visible. Preserve the list with the approved draft for future corrections or sequels. It should help the next editor understand intentional choices while leaving room to identify fresh errors and reconsider decisions when the story changes.
Frequently asked questions
Clear answers for this publishing decision.
Should I ignore a flagged phrase everywhere after approving it once?
Not automatically. Record the context and reason for the approval, then distinguish a local dismissal from an exception that genuinely applies throughout the manuscript.
Can an exception list replace a fiction style sheet?
It can support one, but it serves a narrower purpose. A full style sheet may also document names, formatting, continuity, and broader editorial decisions beyond automated alerts.
When should the exceptions be reviewed again?
Revisit them after major manuscript changes, altered voice decisions, or important tool updates. Confirm that they still apply and do not conceal errors outside their intended context.
What must I check, disclose, and verify myself when AI helped produce the book?
Review facts, rights, and source material yourself, record which parts were AI-assisted or AI-generated so you can answer Amazon’s disclosure question accurately, and treat every automated check as a starting point rather than a verdict.
See what the compliance checks cover