Every week brings a new AI writing tool promising to revolutionize how fiction gets written. Most of them will make your book worse. A few will make it better. The difference isn't about whether AI is involved — it's about what the AI is actually doing, and most current tools are doing the wrong thing.
This post is about that distinction. Not "AI good" or "AI bad" — those are unhelpful framings. Specifically: which uses of AI improve fiction, which uses degrade it, and how to tell them apart.
The generation trap
The default mode of most AI writing tools is generation. You type a prompt, the tool produces prose. The interface is built around getting more words on the page faster.
For fiction, this is the wrong primary mode, for reasons that go deeper than "AI prose is bad."
The words aren't yours. This isn't a moral point — it's a craft point. The voice of a novel comes from the writer's specific way of seeing and saying things. AI-generated prose has a smoothed, averaged voice. Drop it into your manuscript and the seams show. You can edit it heavily and bring it closer to your voice, but the easier the AI made it to produce, the more your draft drifts toward the AI's voice and away from yours.
Generation outpaces thinking. Writing fiction is mostly thinking. The slow process of figuring out what should happen, what the character would do, why this scene matters, is the work. AI generation lets you skip the thinking and produce text. The text exists. The thinking didn't happen. Later, in revision, you'll discover the thinking was the actual deliverable.
The plot solves itself badly. Stuck on what happens next? An AI will happily generate a plausible-sounding next scene. Plausible-sounding is the enemy. The choices that make a novel feel inevitable in retrospect are the surprising-yet-right choices — and those don't come from "what would plausibly happen next." They come from your specific understanding of your specific characters in your specific world. Generic AI generation gives you generic plot.
Characters become inconsistent. Most AI tools don't know your characters. They know "fantasy characters" or "noir characters" or "young adult protagonists." Ask them to write dialogue for Mara and you get a competent fantasy character speaking competently. You don't get Mara. Across enough scenes, the cumulative drift in character voice becomes a serious problem.
This isn't a critique of large language models as a technology. It's a critique of using them, naively, as generation engines for prose that's supposed to be yours.
What AI does badly in fiction
Let's be specific about the failure modes.
Generating prose in your voice. Possible with heavy fine-tuning and consistent re-prompting. Not what most tools offer. Default output is averaged and smooth in a way that flattens distinctive writing.
Generating plot. AI is good at producing plausible. Fiction needs specific. The two are different operations and current tools default to the wrong one.
Generating character dialogue without grounding. Without specific knowledge of who the character is, what they know, and what they're like, AI dialogue is generic. Generic dialogue is the death of character.
Making craft decisions. Should this scene be in past or present tense? Should this revelation come now or in chapter 17? Should this character live or die? AI can offer opinions, but they're statistical averages, not craft judgments about your specific book.
Replacing revision. AI tools that promise "polished prose" usually deliver prose that's smoother and worse — the texture and idiosyncrasy that make writing alive get sanded off. Revision is where a novel becomes itself. Outsourcing it loses the book.
What AI does well in fiction
Now the other side: AI is genuinely useful for specific tasks if you use it for those tasks.
Searching your own work. Hundreds of thousands of words of draft and notes are hard to search by keyword. Asking "where did I mention the gate seal?" and getting an actual answer from across your whole project is a small thing that adds up to hours of saved time.
Surfacing contradictions. AI is good at noticing when two things don't quite fit. A timeline says one thing, a scene says another — a system that catches that for you, instead of waiting for revision, is a real productivity gain.
Talking to characters. Not generating dialogue for the book — using the character as a sounding board. Asking "Mara, why did you trust this stranger?" and getting an answer grounded in what she knows and who she is, then writing the actual scene yourself with that thinking informed. The output isn't prose. The output is understanding.
Brainstorming options. "What are five ways a character with these specific constraints could escape from this specific situation?" is a question AI can usefully answer, especially when the AI knows the character and constraints in detail. You still pick. You still write. But the menu of options is bigger than what your tired mid-draft brain would have generated alone.
Continuity questions. "Has anyone established what color the inn sign is?" "What's the youngest character in the kingdom currently?" "When was the last reference to magic being silent?" Questions that take ten minutes to answer manually can take ten seconds.
Light research. Looking up the historical reality of medieval falconry, the chemistry of a specific poison, the geography of a real region you're basing a fictional one on. AI summaries are starting points, not endpoints, but they're useful starting points.
Notice what these have in common: AI is doing work that supports your writing without producing your prose. The generation is yours. The thinking is yours. The AI is making the support tasks faster.
The grounding question
The single biggest differentiator between AI writing tools is whether they're grounded in your specific work.
Ungrounded AI knows "fiction in general." It will help you with anything because it doesn't know anything about your book. The help is generic by definition.
Grounded AI knows your characters, your world, your timeline, your draft. When you ask a question, the answer is specific to your project. When you talk to a character, they respond from what they know in your world, not from a generic notion of fantasy characters.
This difference is enormous in practice. Asking ChatGPT "would my character do X" is useless — it doesn't know your character. Asking a grounded tool the same question, where the tool has your character's full history, relationships, secrets, and current state, produces actually useful answers.
Most current AI writing tools are ungrounded or weakly grounded. They might let you paste in some character notes, but the AI isn't deeply integrated with your worldbuilding. The output reflects that.
How to choose an AI writing tool (or whether to)
A short checklist:
Does it generate prose for you to drop into your book? If yes, be very cautious. This is the failure mode. You can use such tools usefully for first-draft scaffolding that you'll completely rewrite, but most writers don't completely rewrite. They edit. And edited AI prose is still mostly AI prose.
Is it grounded in your specific project? If no, the help will be generic. Generic help isn't useless, but it's not what most tools advertise themselves to be.
Does it support thinking, or replace it? A tool that helps you think through a scene before you write it is different from a tool that writes the scene for you. Both exist. The first makes books better. The second makes books worse.
Does it preserve your voice? Any AI involvement in actual prose generation is a voice risk. If the tool's primary mode is producing words that go in your book, your voice is being diluted whether you notice or not.
Does it help with craft work that's hard to do manually? Searching your draft, catching contradictions, talking to characters, surfacing patterns — these are tasks where AI is a force multiplier without being a replacement.
Where Inkwarden fits this picture
Inkwarden's AI is deliberately not in the business of writing your prose. It's grounded in your specific world — your characters, your timeline, your lore — and its job is to support the thinking that produces the writing.
You can talk to your characters in their own voice, from what they know. Not so they can write the scene for you, but so you understand them well enough to write the scene yourself. The system surfaces contradictions between your draft and your established canon, so you catch the slips before a reader does. You can ask questions about your own world and get answers from your actual notes, not from generic fantasy tropes.
The prose stays yours. The AI does the supporting work that makes writing the prose easier — the searching, the cross-referencing, the character-grounded brainstorming. This is what AI-assisted fiction looks like when the tool is built around making the writer better at writing, rather than producing words on the writer's behalf.
Join the Inkwarden waitlist for a different model of what AI in fiction can be.
The bottom line
AI in fiction is not one thing. The same underlying technology can be deployed in ways that help your book and ways that hurt it. Generation-first tools that produce prose tend to flatten voice, replace thinking, and accelerate the production of words that aren't really yours. Grounded, support-first tools that know your specific work tend to make the actual craft easier without taking it over.
The question to ask of any AI writing tool isn't "does it use AI" — that's not informative. The question is: what does it do with it, and does that make the writer better or just faster? Faster bad writing isn't a win. Better writing — actually yours, actually thought-through, with less friction in the craft work that supports it — is.