The best AI for novel continuity checking, in my test, turned out to be the one most writers already have open in another tab: a general AI chat with a good prompt. Claude caught all 8 continuity errors I hid in a 3,463-word manuscript. The best dedicated checker, Novilot, caught 6. PlotLens and Pantser caught fewer on their free plans; Sudowrite and Novelcrafter wouldn't run a check until I paid or plugged in my own AI; and our tool, the AuthorFlows Story Analyzer, ranked last on facts while giving the best story-level notes.
Full disclosure: I'm a co-founder of AuthorFlows. I ran this test to see where our Story Analyzer actually stands, and I'm publishing the results exactly as they came out, including the part that stings.
Every planted error was worth 2 points for a clean catch and 1 point for a half-catch. A perfect score is 16.
Search for this topic, and you'll mostly find pages written by the tools themselves, each one quietly crowning itself. I couldn't find a single one that showed the manuscript it tested, the errors it hid, or what the winner missed.
So I built a test where I knew every answer before any tool saw a word.
The Salt Road is a 5-chapter fantasy novella, 3,463 words long. It opens in a cold office at the top of a guild tower:
"This you do not open until you are inside the walls of Cindermoor. Not on the road. Not if you're dying."
Mara Quell, a 29-year-old courier for the Harbor Guild, is hired to carry a ledger to the magistrates of Cindermoor, a walled city three days' ride away. The ledger proves the Harrowgate city watch has been skimming the crown's salt tax, so the watch captain, Idris Venn, wants it back. Mara travels with Tomas Reyne, a disgraced mapmaker who knows the road. She carries her late father's brass compass, its glass cracked straight across and never mended. And she hasn't set foot on a boat since her brother drowned ten years ago.
If you write fantasy, you know how fast this kind of detail piles up (our guide on how fantasy writers stay organized exists for exactly that reason). A cracked compass, a fear of water, a three-day road: every one of those is a promise the story has to keep. That's what made it a good trap.
Every tool got the exact same text. Where a tool had somewhere to put a story bible (character profiles, world notes, a codex), I gave it the same one-page bible.
I split them into three levels, from "a careful proofreader would spot this" to "you need the whole book in your head."
The easy ones: details that change
The sneaky ones: time and knowledge
The hard ones: the story contradicting itself
A good checker also has to know when to keep quiet, so I added two things that look like errors but aren't:
Flagging either one cost a point. For the record, no tool fell for them.
The full answer key: every planted error and decoy, with the chapter where each detail is set up and where it breaks.
I opened a new chat in Claude, pasted the full manuscript, added the prompt above, and hit enter.
It found all eight. It called Halvard's reappearance "almost certainly an error." It connected Mara's cheerful ferry ride to her drowned brother and the fact that she can't swim. It caught Venn turning into Vane, the grey and brown eyes, the brass compass going silver, the "only a day" slip, Tomas knowing too much too soon, and the letter nobody opens. Every finding came with the chapter and both passages quoted side by side.
What impressed me most was the judgment. It noticed that Tomas's early knowledge could be a deliberate hint before deciding the story never pays it off. It left both traps alone. And it found two mistakes I hadn't planted, which I'll come back to below.
The catch: a general chat remembers nothing between sessions, so you paste the book in every time, and the quality depends on your prompt. 3,463 words is also easy for any model to hold at once. A 100,000-word manuscript is a much harder job (more on that below). I'd expect ChatGPT or Gemini to do well with the same prompt, but I didn't test them, so I'm not scoring them.
Novilot is built around continuity, and it shows. The free plan gives you 50 one-time credits with no card. I pasted the five chapters into a project, ran the Continuity Guardian's standard scan, and about a minute later it had read all 3,463 words. I still had 42 of my 50 free credits afterward.
It came back with six issues, each tagged by chapter.
Novilot's Continuity Guardian flagged 6 of the 8 planted errors in about a minute, sorted by chapter.
Eye color, the compass, Halvard alive in Chapter 5, Venn becoming Vane, Tomas's early suspicion, and Mara boarding the ferry "despite her established phobia." Click any finding and it opens into a full explanation: both passages quoted, possible in-story reasons ruled out, a suggested fix, and a link straight to the chapter.
Each Novilot finding quotes both passages, explains why it's an error, and offers a fix.
It missed two: the "only a day" timeline slip and the letter that never gets opened. Both need a reader to hold the whole arc in mind, and that's where the dedicated tools struggled across the board.
On price, the Writer plan is $19 a month for 100 credits, and a re-run on my manuscript was quoted at about 5 credits. One warning: new free accounts get a single personal project. I used mine up by clicking New Project out of habit, so decide where your manuscript goes before your first click.
Pick Novilot if you want a dedicated fact checker that shows its evidence and costs almost nothing to try.
Pantser reads your first 30,000 words for free without an account, which is a generous way to start. I imported the manuscript as a .docx and waited while it read.
Here's the honest framing for this score: Pantser's dedicated continuity checker is part of its paid Storyteller plan ($10 a month, with a 14-day trial). What I tested was the free chat, using one of its ten free questions and the same prompt every chat-based tool got.
Pantser's free chat spotted the eye color change, then told me Halvard's fate was "accounted for."
It caught the eye color, the unopened letter, and Mara taking the ferry after telling Anwen in Chapter 1, "I don't take ferries." The letter matters: Pantser was the only writing tool to notice it. It also correctly read Tomas's lie as deliberate.
It missed the name change, the compass, the timeline slip, and Tomas's early knowledge. And it told me Halvard's fate was "accounted for," which is a strange thing to say about a man who dies in Chapter 2 and nods hello in Chapter 5.
Pick Pantser if you want to try an AI reader on your book with zero commitment, then decide whether the paid checker is worth a trial.
PlotLens takes a different route. It builds a canon from your material, then checks new text against it. I uploaded the manuscript and the story bible to a free project, and it pulled out 44 story elements and 90 canon rules. That part was genuinely impressive.
The full-manuscript check needs a paid plan (Plus is $15 a month), so I used the free Spot Check on the chapters where the errors sit. It caught the eye color and the compass, and both findings quote your text, cite the canon rule you broke, and suggest a fix.
PlotLens caught the eye color and the compass, citing the canon rule each passage broke.
Everything else slipped through. On several checks it reported that "40 rules were not checked (per-check limit)," which tells you the free tier is working with one hand tied.
On the free plan, PlotLens keeps its Real-time and Full Manuscript checks locked.
So treat this score as a score for the free plan, not for PlotLens itself. I couldn't test the paid full check, and I'd genuinely like to.
Pick PlotLens if you want a canon database with cited evidence and you're ready to pay for full checks.
This is the section I least wanted to write.
I set The Salt Road up properly in AuthorFlows: all five chapters, the seven characters with their profiles (Mara's fear of boats included), and the places and key items as world notes. Then I ran the Story Analyzer twice. Each run took about 40 seconds and used no credits.
The Story Analyzer scored The Salt Road 70/100, with chapter-specific notes on characters, plot, and world building.
On continuity, it earned a single point. It said the "travel logistics could be clearer to avoid confusion regarding timing," which points at the "only a day" slip without naming it. It missed all three fact errors (the eyes, the name, and the compass), plus Halvard's return, Tomas's early knowledge, and the letter.
The moment that really hurt: it listed "Mara Quell's backstory and refusal to use ferries" as one of the story's strengths. In a manuscript where she happily boards a ferry. I expected it to catch that one, since her fear was written into her character profile. It didn't, on either run.
It listed Mara's refusal to use ferries as a strength, in a story where she happily takes one.
Now the other side, because it's just as true. The Story Analyzer reads like a developmental editor, not a fact checker, and on that job it gave me notes no other tool did:
Note from the team: We built the Story Analyzer to judge whether a story works: arcs, pacing, setups, reveals. This test showed us something we hadn't caught ourselves: it doesn't yet flag fact-level slips like a changed eye color or a dead man walking back into chapter five. Now that we've found the gap, we're going to close it, and we'll update this post when we do. Yassine Rhouati, Co-founder, AuthorFlows
Pick the Story Analyzer if you want developmental feedback while you draft, alongside a separate continuity pass. If you're weighing it against other editing tools, our AI developmental editor comparison covers that side in depth.
I had a Sudowrite trial ready to go. Its editor-style Feedback tool, the one that covers plot holes and reads your Story Bible, is "exclusive to Professional and Max plans" ($22 and $44 a month). In the trial editor, it's simply marked PRO.
In the Sudowrite trial, the Feedback tool is marked PRO and can't be used.
What Sudowrite offers instead is Chapter Continuity, which links up to 25 chapters so its AI reads the previous 20,000 words before writing anything new. That's prevention while you draft with its AI, not a scan of mistakes already on the page.
Since I couldn't test it, here's what others report. A 2026 Sudowrite review scored its continuity 8/10 and credited the Story Bible, but warned that "credits can run out fast if you use heavy rewrite/expand loops." Sudowrite's own changelog keeps announcing models with "less plot drift" and "stronger continuity across chapters," which suggests drift is still something it's working on.
Pick Sudowrite if you draft with AI and want its Story Bible keeping new text consistent. We go through the full trade-offs in AuthorFlows vs Sudowrite.
Novelcrafter doesn't have a continuity checker. Its chat can read your manuscript, so I imported The Salt Road and opened a chat. It showed "AI Connection Required."
Novelcrafter's chat stays locked until you connect your own AI provider.
That's by design. Novelcrafter doesn't include AI of its own, so you connect a provider and pay for the usage yourself (bring-your-own-key starts on the $8 Hobbyist plan, after a 21-day trial). The imported manuscript also arrived with an empty Codex, the story bible that feeds its AI, so you'd fill that in by hand first.
Once it's set up, people like it. An April 2026 review said "the AI read my entire 40k+ word document and gave specific feedback referencing scenes, characters, and plot points," but told readers to "budget about an hour for the initial setup," with real costs around $13 to $40 a month including AI usage. Its Codex is widely praised for series, and Novelcrafter even publishes a guide to tracking previous books.
Pick Novelcrafter if you want full control over the AI model and a deep codex, and you don't mind the setup. We compared the two platforms head to head in AuthorFlows vs Novelcrafter.
See the pattern? The easy, detail-level errors got caught by almost everyone. The timeline slip and the dropped letter beat nearly every dedicated tool, because catching them means holding the whole story in mind at once.
Claude also flagged two things that weren't on my answer key, and it was right both times:
I wrote that story, reread it several times, and missed both. No dedicated tool caught either one. That's the most useful thing this whole test taught me: the errors you don't know about are exactly the ones you need a checker for.
Prices were checked on September 29, 2026, and they change often.
This test pulled apart two jobs that usually get lumped together as "AI feedback."
A continuity checker compares facts. It reads what your story established and flags whatever contradicts it: an eye color, a name, a date, a character who should be dead. Novilot, PlotLens, and a well-prompted chat all do this, with different depth.
A developmental analyzer judges the story. It asks whether a reveal is earned, whether an arc lands, whether a chapter drags. That's where the Story Analyzer did its best work and where the fact checkers had nothing to say.
A manuscript can be perfectly consistent and still flat. It can also be gripping and full of slips. If you're self-editing your novel with AI, run a developmental pass while you draft and a continuity pass before you call revision done.
So which one should you use?
For the wider picture of AI tools built for novelists, see our roundup of the best AI novel writing software for indie authors and our 2026 AI writing tools updates.
Yes, with conditions. My test used 3,463 words, which every tool could read in one pass. At novel length, three things change:
Timeline errors like my "only a day" slip are easier to prevent than to find. A simple day-by-day map of the journey, the kind we describe in story mapping for discovery writers, would have caught it before any tool did.
Continuity across several books is harder than within one. Novelcrafter's Codex and PlotLens's canon are both built to carry facts from book to book. In AuthorFlows, each book lives in its own project, so for a series you set up each new book and re-enter the characters and world details that carry over. It's more manual than a shared codex, and worth knowing before you start book two.
Whatever you use, keep a short fact sheet per character (eyes, age, scars, who knows what and since when) and check each new book against it before you send it to beta readers. If you're still choosing where to plan your books, our breakdown of novel outline software helps you decide.
You don't have to take my word for any of this. It takes about an hour:
It's the fastest way to learn what a tool really does before you trust it with a full draft.
Continuity errors are places where a story contradicts itself: a character's eye color changes, a dead character reappears, a journey takes three days in one chapter and one day in another, or a character knows something before they could have learned it. Readers notice them because they break trust in the story's world.
In my test, a general AI chat (Claude) with a clear prompt caught all 8 planted errors, and Novilot was the best dedicated checker with 6 of 8. The right pick depends on length: a chat is strong for one-off checks, while dedicated checkers keep your story bible and scan long manuscripts between sessions.
Yes. A consistency check doesn't write anything for you; it points out contradictions you then fix yourself. Many writers use AI this way as a fast first reader. Keep the final calls yours, since some "errors" are deliberate choices, like a character's lie.
Several tools can be tried for free. Novilot gives 50 one-time credits, enough for several scans of a short manuscript. PlotLens offers free spot checks, Pantser reads your first 30,000 words free, and any general AI chat can run a check with a good prompt.
A continuity check compares facts and flags contradictions in names, dates, and physical details. A developmental edit judges the story itself: pacing, character arcs, and whether reveals are earned. Most manuscripts need both, usually a developmental pass first and a continuity pass before final revision.
Manage characters, plotlines, and relationships visually. Start using AuthorFlows now!
Get Started Now