Sagoræ · Blog

Which AI Gives the Best Feedback on a Novel? ChatGPT, Claude and Gemini for Writers

Published 20 November 2026 · 11 min read · Writing Software

Which AI Gives the Best Feedback on a Novel? ChatGPT, Claude and Gemini for Writers

No AI wins across the board at critiquing a novel. ChatGPT, Claude and Gemini are all good at catching repetition, dialogue that explains too much and nearby continuity slips, and nearly all of them miss your voice, pacing over a hundred pages, and honesty (they flatter). The right pick depends on what you want read (structure, continuity, a scene or the prose) and on two settings most writers never check: what the tool does with your manuscript, and how much text it can take in at once. Your prompt matters more than the model.

Below: a table by type of feedback, what really differs between the three assistants (privacy included, checked on November 20, 2026), a bad prompt next to a good one, a one-evening protocol to run your own comparison, and a feedback prompt you can copy.

What exactly do you want feedback on?

"Read my novel" means nothing to an AI, and not much to a human. There are at least four different kinds of read, and a tool that's great at one can be mediocre at another.

What you want readWhat AI tends to do wellWhat it often missesWhat it needs from you
Structure (the whole book)Summarizing, spotting a sagging middle or a dropped subplotWhether the ending pays off; it almost always likes the endingThe full manuscript at once, or a chapter-by-chapter synopsis
Continuity (names, ages, dates, places)Comparing two passages you show itAnything it can't see; it may invent a detail of your textYour character profiles and timeline next to the text
A scene or chapterExpository dialogue, tension that leaks out, point of view that slipsYour voice: it tends to "smooth out" what makes your style yoursThe previous chapter, and what this scene has to do
The proseRepeated words, grammar, filler adverbsDeliberate rule-breaking, consistent house style (US or UK spelling)A clear instruction: flag, don't rewrite

The table already says one thing: the tool matters less than what you feed it. An AI that only has chapter 14 can't tell you Lucy's eyes were green in chapter 2. It can, however, assure you they were brown, which is worse.

ChatGPT, Claude or Gemini: what actually changes for a novelist?

All three can read a chapter and give a reasoned opinion. The differences that matter to an author aren't the ones on model leaderboards. They're what happens to your text, and how much of it the tool will take.

ChatGPTClaudeGemini
Is your text used to train models?Yes by default on individual plans, unless you opt outOnly if you allow it in your privacy settingsYes while History is on; some chats are read by human reviewers
How to avoid it"Do not train on my content" option, or a Temporary ChatTurn off training use in privacy settingsTurn off History, or use a temporary chat
Stated retentionNot specified on the page we read30 days if you opt out, 5 years if you allow trainingChats reviewed by humans kept up to three years
How much it reads at onceNot checked (help page wouldn't load on the day)Not given as a number on the page we readGoogle advertises up to 1,500 pages of text on Google AI Pro

Sources, checked November 20, 2026: OpenAI's page on how your data is used (updated March 13, 2026), Anthropic's announcement on its consumer terms (August 28, 2025), and the Gemini Apps Privacy Hub (updated November 17, 2026). On price, Claude Pro is listed at $20 a month in the US. For what a full novel actually costs with these tools, see how much it costs to write a novel with AI.

Stop on the training row for a second. An unpublished manuscript is the one thing you fully own before it's out in the world. Before you paste it into a chat window, check that setting once and for all. Five minutes, and it's settled.

So which one gives the best feedback?

Nobody has a serious answer: models change several times a year, and we haven't found a public test run on whole novels. The most concrete recent trial we've read is Amanda Caswell's for Tom's Guide, published August 27, 2026, where she gave the same draft to all three. She rated Claude best at stepping back to see how the whole story worked (it noticed readers never learn what the employees in her story actually do for work), ChatGPT the strongest line editor, especially on pacing, and Gemini the clearest about where to start revising. One draft, one writer. Treat it as a hint, not a verdict, and run your own test (the protocol is below).

What AI catches well in a novel

Credit where it's due: on some jobs, these tools are fast and accurate.

Repetition and tics. The same verb three times on a page, "suddenly" opening every other paragraph, characters who shrug every ten lines. An AI sees them better than you do on your tenth pass.

As-you-know-Bob dialogue. "As you know, Mark, Dad died ten years ago": the line that informs the reader, not the person being spoken to. They flag it nearly every time.

Point-of-view slips. You're in Lucy's head, and suddenly we know what the waiter thinks. That's exactly the kind of thing a tired human reader lets through.

Nearby continuity errors. If both passages are in the text you gave it, an AI will often find the time that doesn't add up or the car that changes color.

What it misses, and why that matters

Honesty. This is the big one. General-purpose assistants are built to be pleasant. Ask "what do you think?" and you'll get three compliments, two suggestions wrapped in cotton wool and an encouraging sign-off. A first reader who loves everything teaches you nothing.

Your voice. An AI reading your prose tends to pull it toward a competent average. Your verbless sentences, your deliberate tense shift, your clipped punctuation: it offers to make them "flow." Accept every suggestion and your chapter 12 will read like everyone else's chapter 12. More on that in how to spot AI writing in fiction.

Long-range pacing. Feeling a middle sag over eighty pages means reading eighty pages and getting bored in the right place. An AI doesn't get bored. It can tell you chapter 9 "slows the action," not that a reader put the book down in chapter 11.

Your actual text. They sometimes make things up: a quote from your chapter that isn't there, a minor character misnamed, a scene summarized slightly wrong. Check every example they quote before acting on it.

House style. Ask for UK spelling and you may get "color" back in paragraph six. Dialogue tags, single or double quotes, the serial comma: if you have a style, state it in the prompt, or the tool will drift toward whatever it sees most often.

Before and after: same chapter, two prompts

The difference between useless feedback and feedback that moves you forward is the prompt. Here's an invented example: a chapter where Lucy meets her brother after ten years apart.

Before. The prompt: "Can you read my chapter and tell me what you think?"

Typical reply: "This chapter is really strong! The reunion is moving and the dialogue feels natural. You might add a bit more description to deepen the atmosphere. Here's a reworked version of the first paragraph…" Then ten rewritten lines you didn't ask for, which are no longer yours.

After. The prompt: "You're a demanding first reader, not a copy editor. Don't rewrite a single sentence. This chapter has to make readers feel Lucy hasn't forgiven him, without her saying so. Tell me: 1) the exact point where you stopped believing it, quoting the sentence; 2) any lines that explain to the reader what the characters already know; 3) one thing that doesn't work and that I probably won't want to hear."

The reply becomes: "1) 'I'm still angry with you, you know': Lucy says what the scene was supposed to make us feel. 2) 'Ever since Dad left in 2014…': Mark knows the date; that line is for the reader. 3) Mark has no reason to show up that day. Without one, the reunion looks arranged by the author."

Three notes, all checkable, not one sentence rewritten. That's the feedback you're after, whatever the tool.

Run your own comparison in one evening

Instead of trusting a ranking, test the tools on your own pages. You need a chapter whose flaws you already know (ideally one a human reader has already commented on) and about ninety minutes.

  1. Pick that chapter and, before anything else, write down the three flaws you know it has.
  2. Give it to each tool in a fresh conversation, with exactly the same prompt (the one at the end of this article).
  3. Fill in the grid below for each tool, without rereading the other replies in between.
CriterionTool 1Tool 2Tool 3
Found my three known flaws (0 to 3)
Found a flaw I hadn't seen (yes / no)
Every quoted example is really in my text (all / not all)
Rewrote sentences even though I said not to (yes / no)
Said at least one thing that stung and was right (yes / no)
Kept my spelling and punctuation style (yes / no)

Reading the grid. First, drop any tool that quotes passages that aren't in your text: you'll never be able to trust it on a whole manuscript. Then the one that finds your known flaws and adds a new one is your best reader for this kind of chapter. If it rewrites despite the instruction, keep it for prose, not for scenes. And if none of them says anything unpleasant, your prompt is too polite: toughen it and run it again.

Quick test: is the feedback you got usable?

Answer yes or no with the AI's reply in front of you.

  1. Does every note point to a specific place in the text (a sentence, a line of dialogue)?
  2. Are the quoted sentences actually in your chapter?
  3. Does the feedback include at least one criticism that stung?
  4. For each note, can you tell what to change without copying a sentence from the AI?
  5. Does it talk about what the scene was meant to do, not just style?
  6. Could you show it to a writer friend without being embarrassed you asked?

Reading the test. Six yeses: keep this feedback and this tool. A no on question 2: throw the whole reply out, part of it is based on a text that doesn't exist. A no on 3: the tool flattered you; rewrite your prompt to demand criticism. A no on 4: it handed you ready-made sentences, and if you paste them in, the chapter is no longer quite yours. A no on 5: you got a proofread, not a critique.

What about tools built for novelists?

A general assistant knows nothing about your novel beyond what you paste into the chat. Every new conversation, you start over: the synopsis, the characters, the timeline. That's where dedicated tools make the bigger difference, more than the model underneath.

Grammar checkers handle the mechanics of English and leave structure alone. Fiction-generation tools are built to produce text, and we cover them in should you use AI to write a novel.

In Sagoræ, the coaches read your chapter where you write it, with your character profiles alongside, and only when you ask. Théo reads your chapter the way an editor would and tells you how it felt, each note pinned to the passage it's about. The Proofreader checks the prose (agreement, punctuation, repetition) and explains the rule; for a plain error, it suggests the fix and you're the one who clicks. The Archivist compares the chapter to your character and place profiles to catch what's drifted. Dédale looks at the frame of the story and asks the questions that get you unstuck. None of them writes a scene for you, and Sagoræ doesn't train any model on your writing.

T Theo · Scene critic Pre-recorded demo

Theo reads chapter 7 of "The Night of the Hourglasses"

You

Here is chapter 7. Iris meets her brother again after six years. It feels flat and I cannot tell why.

Iris pushed open the cafe door. Elias was there, at the back table, exactly as she remembered him. He looked up. She sat down. They talked for two hours about everything that had happened, and when she came out, night had fallen.

Theo

I see why it feels flat. The scene you have been promising for three chapters (the reunion) is summarised in one sentence: "They talked for two hours". The reader is waiting for exactly those two hours, and you take them away.

Three things I notice:

  • The tension is gone by the second line. "Exactly as she remembered him" reassures immediately. What has changed in Elias that Iris did not expect?
  • Nobody wants anything in the scene. What has Iris come for, concretely? An answer, forgiveness, money? Without a goal, the conversation can neither succeed nor fail.
  • The first exchange does not exist. The first sentence Elias says after six years is worth more than a summary of two hours.

You

She has come to ask him to testify at their father's trial. She knows he will refuse.

Theo

Then you already have your scene: a request she does not dare to make, in front of someone who is expecting it. Two questions to rewrite it (I will not do it for you):

  • At what exact moment does Iris say the word "trial"? Everything before that word is tension; everything after it is conflict.
  • What is Elias doing with his hands while she talks? One concrete gesture will do more than "he looked up".

Keep the night falling as she leaves, it is a good closing image. But let it fall on a refusal, not on a summary.

Pre-recorded simulation of an exchange with Theo. The chapter text stays the author's: Theo never rewrites a line of it.

The free plan includes 20 credits at sign-up to try the coaches on a real chapter: enough to add a fourth column to the grid above. If you're still unsure about the whole idea, we had five AI coaches read the same bad chapter.

The feedback prompt to copy

Paste it as is into any assistant, with your chapter below it.

You are a demanding first reader, not a copy editor or a co-writer.

Hard rule: do not rewrite any sentence and do not offer a new version.

Context: [genre] novel, chapter [n] of [total].

This chapter needs to [what the scene must do, in one sentence].

What happened before: [three lines].

House style: [US or UK spelling, single or double quotes].

Quoting the exact sentence from the text each time, tell me:

  1. Where you stopped believing it or stopped caring.

  2. Lines of dialogue that inform the reader rather than the listener.

  3. Passages where the point of view slips.

  4. Repeated words or phrasings.

  5. One criticism I probably won't want to hear.

If you are not sure a sentence is in the text, do not quote it.

Frequently asked questions

What's the best AI to critique a novel? There isn't a single one. For pure grammar, a dedicated checker is more reliable than an assistant. For feedback on a scene, ChatGPT, Claude and Gemini all do a decent job if the prompt forbids rewriting and demands quotes. For the structure of the whole book, you need a tool that takes the full manuscript or already knows your chapters.

Does ChatGPT use my manuscript for training? On individual plans, OpenAI says it may use your content to train its models unless you opt out (the "do not train on my content" option). Temporary Chats aren't used for training. Check that setting before pasting an unpublished chapter. Claude and Gemini each have their own setting, covered in the table above.

Can I give an AI my whole novel? It depends on the tool and the plan. Google advertises up to 1,500 pages of text on Google AI Pro. Even when a tool accepts everything, a whole-book read is shallower than a chapter read: use a chapter-by-chapter synopsis for structure, and go chapter by chapter for everything else.

Can AI replace beta readers? No. It catches surface problems fast, but it doesn't get bored, doesn't get attached and doesn't put the book down. That's exactly what you need to learn from a human reader. Use AI before your beta readers so they read a cleaner draft: here's where to find them and what to ask.

If AI critiques my novel, is it still my book? Yes, as long as it flags and you rewrite. The line is crossed when you paste its sentences in place of yours. More in is using AI to write a novel cheating.

Get chapter feedback from coaches that never rewrite, with your profiles right alongside: try Sagoræ on the free plan, no card needed.

✒️ Read next: Should You Use AI to Write a Novel? What to Hand Over, What to Keep · AI-isms in Writing: 12 Signs of AI-Generated Prose (and How to Clean Up Your Own) · How to Find Beta Readers: Where to Look, What to Ask and How to Read Their Feedback · We Asked Five AI Coaches to Read the Same Bad First Chapter. Here's What Each Said · How Much Does It Cost to Write a Novel With AI? Credits, Subscriptions and Price per Word

Write your novel in a real workshop: a focused editor, characters, story structure and AI coaches that guide you without writing for you. Start for free, no card required.