Start writing free

Blog

Which AI writing tools don't train on your manuscript?

Some tools never train on your writing, some do unless you switch them off, and some depend on your plan. We read each vendor's own policy on 26 September 2026.

Published 26 September 2026

Engraving of an author holding the key to a locked manuscript cabinet while an assistant works outside it

The usual fear is easy to picture: a chapter pasted into a chatbot at midnight, absorbed into a model and resurfacing in pieces somewhere else. Some writing tools never train on your text, some do unless you switch it off, and some answer differently depending on your plan or on which AI company sits behind the tool. The answer is rarely on the pricing page. It sits in privacy policies and help-centre articles, where the wording slides between "we", "our models" and "our service providers". Those are three different promises.

We make Chaptora, one of the nine tools below, so weigh our view of it accordingly. We read each vendor's own policy and help pages on 26 September 2026, quoted them, and listed every page under Sources. Where a policy is silent or unclear, we say so. The ranking follows what decides the question: whether training is off by default, whether the promise also covers the AI provider behind the tool, and whether the answer changes with the plan you pay for.

Which AI writing tools train on your manuscript by default?#

Of the nine tools here, Grammarly (for individual users), ChatGPT (for individuals) and Gemini (personal accounts) train by default until you switch it off, and Claude asks you to choose. Chaptora, Type.ai, ProWritingAid, Novelcrafter and Sudowrite state that they do not train on your writing. The bigger difference is a second question: what the AI provider behind each tool may do with your text.

ToolTrains on your content by default?How to opt out
ChaptoraNo, on every plan. AI providers are contractually instructed not to train on your content eitherNothing to switch off. The rule is in the terms and covers every account, free or paid
Type.aiNo, and its FAQ says the same of its third-party AI providersNot needed, according to its FAQ
ProWritingAidNo. "Your texts will not be used to improve the quality of our services"Not needed for ProWritingAid. Whether its LLM providers may train is not stated
NovelcrafterNo, "under any circumstances", for Novelcrafter itselfNot needed for Novelcrafter. With your own AI key, the provider's settings apply
SudowriteNo. "We do not use your writing to train AI models"Not needed for Sudowrite. What its model providers may do is not addressed
GrammarlyYes for individual users. Off for accounts bought through its Sales teamTurn off Product Improvement and Training in account settings
ClaudeConsumer apps: your choice, default not stated. Commercial plans and API: noPrivacy Settings, or use Incognito chats
ChatGPTIndividuals: yes. Business, Enterprise, Edu and API: no by defaultTurn off Improve the model for everyone, or use Temporary chats
Google GeminiPersonal accounts: Keep Activity is on by default for adults. Work and school accounts: not used for training outside your organisation without permissionTurn off Keep Activity

1. Chaptora#

Chaptora does not train on your writing, and the answer does not change with your plan or a setting. Our terms put it in one sentence: "We do not use your content to train models or authorize our AI providers to do so." The same terms list what that covers: the manuscripts, notes, research, prompts and other material you provide. That covers the Sources you upload as well as the chapters you write, which matters if you are ghostwriting a memoir under a confidentiality agreement or your research folder holds a family's private letters.

The second half of that sentence is the part several policies here leave out, including ProWritingAid's and Sudowrite's. Like every cloud AI writing tool, Chaptora sends text to a model provider when you use an AI feature. When you ask the AI Sidebar a question, run Inline AI with Cmd+K / Ctrl+K or give the AI Agent a task, the relevant parts of your content (selected text, chapter context, prompts) go to that provider so it can return a response. Our privacy policy names the provider, Google for Gemini, and states: "We instruct these providers contractually not to use your content to train their models." The output comes back and is stored in your Project.

Because there is no training to opt out of, there is no toggle to hunt for, no difference between the free Octavo plan and the paid plans, and no special private mode you have to remember to open before pasting a chapter.

There is one copy you should know about. We keep a record of each AI request (the prompt and the answer, which can include excerpts, Canon and Sources) with Langfuse in its EU region in Ireland, to find errors and improve the features, never for training. Langfuse is part of ClickHouse, whose US affiliate ClickHouse, Inc. provides technical and support services. Records expire after 90 days, and deleting a chat does not shorten that. Write to info@chaptora.com to object or ask for earlier deletion.

Chaptora is run from Vienna, so our privacy policy follows the GDPR and you can complain to the Austrian Data Protection Authority. Some providers, such as Clerk, which handles sign-in, are in the US; those transfers rely on Standard Contractual Clauses and, where applicable, the EU-U.S. Data Privacy Framework. What matters most for a manuscript are these points from our terms and policy:

  • Human access for support requires your explicit permission and is limited to the material needed for that case. Without it, access is limited to complying with a legal obligation or investigating a concrete security or abuse incident.
  • Content stays while your account exists. Delete a Project, a chapter or the account and it leaves our active systems within a reasonable period, and our backups on their rotation.

One limit follows from how Chaptora works. It is a browser-based service, so your manuscript is stored on our infrastructure, not only on your laptop. If your requirement is that the text never leaves your machine, no cloud AI tool meets that bar, us included; you would need a model running locally, such as through Ollama or LM Studio.

2. Type.ai#

Type.ai's FAQ is one of the few that answers both halves: "No, Type does not use your data to train any AI models, and nor do any of the third-party AI providers Type integrates." Its privacy policy names those providers as OpenAI and Anthropic and says data may be transferred to the United States. The policy itself does not repeat the training statement, so the FAQ is where the promise lives.

3. ProWritingAid#

ProWritingAid's privacy policy, updated 4 March 2026, is direct: "When using ProWritingAid, your texts will not be used to improve the quality of our services." It trains its algorithms on "texts hand corrected by professional copyeditors" instead, and its help centre repeats the point: "We never use your data or your writing to train our AI models or algorithms."

It is also brief with your text. Once checking has finished, "all submitted texts and their improvements will be deleted", unless you save a document in its Web Editor, which is then stored encrypted in the cloud. For a line-editing tool that sees a chapter at a time, that is a strong position.

The open question is the providers. Its Responsible AI page says some features "use Large Language Models (LLMs) developed by leading providers in the field of generative AI". Neither that page nor the privacy policy says whether those providers may train on or retain what they receive. Data is stored in the UK and in countries where its providers operate, including the EEA and the US.

4. Novelcrafter#

Novelcrafter's privacy policy, updated 31 August 2026, is emphatic: "We do not, under any circumstances, use the novels or other creative content created by our users to train any artificial intelligence or machine learning models." It adds: "We do not store or log any text content or prompts you send to the AI." Its data sits on AWS in Frankfurt and the US.

The catch is Bring your own Key. On the Hobbyist plan and above, its pricing page lists AI as "Bring your own Key": you connect your own account with a provider such as OpenRouter or OpenAI, and your requests go to that provider under its terms and your settings there. Novelcrafter's own help centre warns that on OpenRouter, "many of the free model providers require you to enable training and/or logging to use their models." Pick a free model and your chapters may be used for training by someone other than Novelcrafter. Pick a provider whose terms you already trust and you control that part yourself. More on the product in Chaptora vs Novelcrafter.

5. Sudowrite#

Sudowrite answers the question in its FAQ: "We are constantly listening to writers' feedback to improve Sudowrite. However, we do not use your writing to train AI models." It also says it claims no rights over anything you put in or anything it generates for you.

The same FAQ says where your text can go: "We use dozens of AI models, including the latest Claude models by Anthropic, multiple open source models, in-house models made expressly for fiction, like Muse, and several models created by OpenAI." What those providers may do with it is not addressed there, and the privacy policies we fetched on sudowrite.com and its documentation site contain no clause on AI training. That is silence, not a yes; ask Sudowrite if it matters to you. See Chaptora vs Sudowrite for the rest.

6. Grammarly#

Grammarly is the clearest case of a default that depends on who pays. Its support article says: "Product Improvement and Training is on for individual users by default." By contrast, "For all Grammarly for Education, Grammarly Business, Pro, and Grammarly Enterprise accounts purchased through our Sales team, Product Improvement and Training is automatically off", while multi-user accounts bought on the website start with it on until an admin changes it. An individual turns it off at account.grammarly.com/security/privacy.

On providers, Grammarly is explicit, which is to its credit: "We do not allow these large language model (LLM) service providers to train their models on user content." Grammarly's privacy policy now lives on superhuman.com (the link on grammarly.com redirects there) and says you decide "whether Superhuman can use your user content to train our AI models" through training controls in account settings.

7. Claude#

Anthropic's consumer apps (Free, Pro and Max) ask you to decide. Its privacy centre says it may use your chats to improve its models if "you choose to allow us to use your chats and coding sessions to improve Claude", if a conversation is flagged for safety review, or if you opt in to a programme such as its Trusted Tester Program. New users "can select your preference in the signup process", and you can change it later in Privacy Settings. We could not find a page that states which option is pre-selected, so we do not claim one.

Your choice changes retention. With model improvement on, Anthropic "may retain your data in a de-identified format for up to 5 years in our model training pipelines". Incognito chats are not used "even if you have enabled Model Improvement". Thumbs up or down feedback stores the whole conversation for up to five years and may be used for training.

Commercial products are different. For Claude for Work and the API, Anthropic says: "We will not use your chats or coding sessions to train our models, unless you choose to participate in our Development Partner Program."

8. ChatGPT#

OpenAI's help centre is unambiguous for individuals: "When you use our services for individuals, such as ChatGPT and Codex, we may use your content to train our models." To opt out, turn off Improve the model for everyone under Settings, then Data controls, or choose Do not train on my content in OpenAI's Privacy Portal. After that, OpenAI says it won't use your new conversations. Temporary chats are not used for training while they remain temporary.

Two details catch people. Rating a response can put "the entire conversation associated with that feedback" into training even after you opt out. And the page names only business products as off by default: "By default, we don't use inputs or outputs from ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, or our API to improve our models." A paid Plus subscription is not listed among those exceptions. For the wider question of writing a book in a chatbot, see why not just use ChatGPT, Claude or Gemini.

9. Google Gemini#

For a personal Google Account, the default leans towards training. Google's help centre says: "If you're 18 or over, Keep Activity is on by default." The Gemini Apps Privacy Hub says Google uses this data to "provide, maintain and improve our services", that these uses "extend to the generative AI models", and that "a subset of chats are reviewed by human reviewers", disconnected from your account and kept for up to three years.

Turning off Keep Activity changes the training part for future chats: "If Keep Activity is off and you don't submit feedback, Google also does not use your future chats to improve its AI models." It does not remove people from the loop. The hub adds: "Even if your Keep Activity setting is off or you use temporary chats, Google still uses your chats to respond to you and help protect Google, our users, and the public, including with help from human reviewers." Work and school accounts follow separate terms: "Your content is not human reviewed or otherwise used for Generative AI model training outside your domain without permission."

How can you check any AI writing tool's policy yourself?#

Open the privacy policy and the help centre, search with Cmd+F / Ctrl+F for "train", "improve", "machine learning", "default" and "service providers", and answer six questions.

  1. Who does the promise cover? "We don't train on your data" covers the vendor. Look for a second sentence about the model provider. If it is missing, the provider question is unanswered, not answered.
  2. Is training off by default? "You can choose" and "you can control" tell you a setting exists, not where it starts. Look for the word "default".
  3. Does the plan change it? Business and API terms often differ from personal accounts.
  4. What does "improve our services" include? Sometimes it extends to model training. In ours it means records used to find errors and improve the AI features, never to train models. A good policy says which.
  5. What overrides your setting? Feedback buttons, safety review and beta programmes.
  6. How long is your text kept, and where? Look for a number of days and a list of countries.

If a policy is silent, treat the answer as unknown and email the vendor. A clear written reply is worth filing with your book's paperwork.

Does "no training" mean nobody sees your manuscript?#

No. Training is one question. Human review, retention and sub-processors are separate ones. Gemini involves human reviewers even with Keep Activity off. Chaptora's terms require your explicit permission before anyone reads a manuscript for support. A tool can promise never to train and still keep your text for years, so read the retention clause next.

Frequently asked questions#

Does Chaptora use my manuscript to train AI models?#

No. Chaptora does not use your manuscripts, Sources or prompts to train models, on the free Octavo plan or on paid plans. The relevant parts of your text go to our AI provider to answer your request, under a contractual instruction not to train on it. Records of AI requests expire after 90 days.

Does paying for ChatGPT Plus stop OpenAI training on my chats?#

Not according to OpenAI's help centre. It says ChatGPT Business, Enterprise, Edu and the API are not used for training by default, and treats ChatGPT for individuals as trainable unless you turn off Improve the model for everyone under Settings, then Data controls. Plus is not listed among the exceptions.

Can I use AI on a manuscript without it leaving my computer?#

Only with a model that runs on your own machine, for example through Ollama or LM Studio. Every cloud writing tool, Chaptora included, processes your text on servers when you use AI. What you can choose is a tool that does not train on it, names its providers and says how long it keeps records.

Does opting out cover what I have already written?#

Not necessarily. OpenAI says that after you opt out it will not use your new conversations, and Google describes the effect of turning off Keep Activity on your future chats. Change the setting before you paste in a chapter, not afterwards, and remember that feedback buttons can override it.

If you want a tool where there is nothing to switch off, start on the free Octavo plan or compare the plans. For the features beyond privacy, see our comparison of AI book writing software.

Sources#