Which AI writing assistants train on your messages? I read the policies

Of the thirteen writing assistants I checked, six use what you type to train or improve their models unless you change a setting or pay for a different plan. Eight say they do not. DeepL Write counts on both sides because its free and paid tiers differ, and Microsoft Copilot counts on both sides because Microsoft’s pages differ by app version. Grammarly, ChatGPT, Google Gemini, QuillBot, the free tier of DeepL Write and the older Copilot app are in the first group. Apple’s Writing Tools, LanguageTool, ProWritingAid, DeepL Write Pro, the newer Copilot app (by Microsoft’s own wording, which covers foundation models), Tonely, ConfiText and Subtext are in the second. Wordtune’s policy names “training” once, as a general purpose in its California notice, and never says whether your text is part of it, so I report only what it does say.

I co-founded Subtext, a tone checker for personal messages, and it is one of the thirteen. I read its privacy policy by the same method as the other twelve and gave it a row in the same tables, but weigh my reading of it with that stake in mind. For every tool except ProWritingAid I read the current privacy policy or terms on 4 October 2026, quoted the clause that matters, and noted the date stamp each policy carries. I opened most of those pages again on 10 October 2026, when I also added ProWritingAid. Where a page would not load, I say so, and three of OpenAI’s four pages would not load on 10 October, so the ChatGPT section stands as I read it on 4 October, apart from the API page, which I reopened that day. Nothing below comes from a news article or a comparison site.

For each tool I asked four things. Does it train on your text by default? If so, where is the switch? What does it keep, and for how long? And does it let its own model suppliers train on your words? Training is not the same as processing your text to answer you, which every tool has to do. It is also not the same as storing it so you can return to a conversation. Policies blur the three, so I kept them apart. A company that trains on your words almost always exempts its business and education customers. And most forbid their own model suppliers from doing the same, so “trains on your messages” usually refers to the company you signed up with.

Grammarly trains by default on individual accounts

Grammarly’s training setting is on by default for Free, Premium and Pro accounts with one user. It is off by default for Enterprise, for Pro and Business bought through its sales team, and for every Education account. That comes from Grammarly’s own support article, Product Improvement and Training Control¹, which describes the setting as using your content to “train our models and improve our product for everyone”. Pro teams with several users, bought on the website, also start with it on. Only admins can change the setting on those accounts. A single user who joins a Business or Education team account will “automatically adopt” the setting its admin chose¹. The article carries no date of its own.

The policy behind it moved. Grammarly’s privacy policy URL now redirects to the Superhuman privacy policy², effective 6 July 2026, after the parent company’s rename. It says “you can decide whether Superhuman can use your user content to train our AI models by adjusting the available training control(s) in your account settings”, and it places content uploaded under an organisational account outside the policy altogether.

The policy’s section for readers in the EU, EEA, UK or Switzerland lists “To develop and improve AI” for “All categories of data”, on the basis of a “legitimate interest” in “developing new products through the use of AI”². The policy text has no regional exclusion from training, so living in Europe does not by itself switch training off. The same section lists a right to object to processing based on legitimate interests, and the policy does not say whether the training control starts on or off there².

Grammarly’s opt-out covers your content, and writing statistics can still be used

Turning off Grammarly’s Product Improvement and Training control stops the use of the content you type, and Grammarly says it may still use some data about how you write. To turn it off on Free or Premium, open account.grammarly.com/security/privacy; on Pro, the path is account.grammarly.com/admin/data_settings¹. Once off, “none of the content a user inputs into an application or website where our product is active will be used for training or product improvement for other users”¹. The same article, for Free, Premium and Pro accounts with one user, continues “We may still collect non-content data, such as writing statistics (total words written, types of suggestions accepted, mistakes made, etc.), and may use that data for product improvement and training”¹.

On suppliers, Grammarly’s Privacy and security FAQs³ state “We do not allow these large language model (LLM) service providers to train their models on user content.” Documents saved in the Grammarly Editor stay until you delete them; there is no day count for text processed elsewhere³. I compared the two products job by job in Subtext vs Grammarly; this page covers only the data.

Apple Intelligence runs on the device or in Private Cloud Compute, which does not keep your content

Apple Intelligence runs on the device in many cases and on Apple’s servers for heavier requests, and neither path trains on your content by default. Apple’s Apple Intelligence & Privacy⁴ page, dated 14 September 2026, says “In many cases, Apple Intelligence models run entirely on device so that a task can be completed without data leaving your device.” For heavier requests it uses Private Cloud Compute, and “The data being processed in Private Cloud Compute is not stored or made accessible to Apple. It is processed only to fulfill your request, after which point the results are returned securely to your device and are not retained by Private Cloud Compute.” Apple may still record limited information about a request, such as its approximate size, the features used and how long it takes, and says that data does not include the content of your request or the result⁴. In June 2026 Apple said it is collaborating with Google and NVIDIA to extend Private Cloud Compute to Google Cloud systems for demanding tasks, and that it retains complete control of the software⁵. Neither Apple’s page nor its post names Writing Tools, so I cannot say which Writing Tools operations stay on the device and which go to Private Cloud Compute.

Apple does not train on your text by default, and its ChatGPT extension follows your ChatGPT account

On training, Apple’s privacy features page⁶ states “Your private personal data and interactions are never used to train our foundation models unless you explicitly choose to help improve them.” That makes training opt-in at Apple, the reverse of the default at Grammarly, ChatGPT, Gemini and the older Copilot app. Apple’s Intelligence page adds that if you opt in to share Device Analytics, Apple may use privacy-preserving techniques to collect data about aggregated trends, including about the content processed by Apple Intelligence, to improve it⁴.

Writing Tools has one exception, the ChatGPT extension. Apple’s ChatGPT Extension & Privacy⁷ page, also dated 14 September 2026, says that without a ChatGPT account “OpenAI must process your information solely for the purpose of fulfilling your request and not store your information or any responses it provides unless required under applicable laws. OpenAI also must not use your information to improve or train its models.” Sign in to a ChatGPT account and “OpenAI may log your request, attachments, and session history, and use this data to train or improve their models”⁷. My reading of Apple’s iPhone pages found the same gap in what Apple says about Writing Tools.

ChatGPT trains on personal accounts until you turn it off

ChatGPT uses your conversations to train OpenAI’s models on personal accounts unless you turn off one setting; the help page lists Free, Go, Plus and Pro as the plans you manage yourself. OpenAI’s privacy policy⁸, updated 6 February 2026, says “we may use Content you provide us to improve our Services, for example to train the models that power ChatGPT”. The setting is called Improve the model for everyone. OpenAI’s Data controls in ChatGPT⁹ article gives the path, Settings, then Data controls, and says “When Improve the model for everyone is off, your new conversations won’t be used to train OpenAI models.”

Feedback buttons are a second route. “If you choose to provide feedback, such as selecting thumbs up or thumbs down on a response, the entire conversation associated with that feedback may be used to train OpenAI models”⁹, opt-out or not.

Temporary Chat is the other route. The policy says those conversations “will not appear in your history or be used to improve OpenAI’s models”⁸, and OpenAI’s Temporary chat¹⁰ article says a copy may be kept “for up to 30 days for safety purposes”. Deleted regular chats are removed “within 30 days”⁸.

Business accounts and the API follow different rules. “By default, OpenAI does not use content from ChatGPT Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces to train its models”⁹. OpenAI’s API data page¹¹ says “As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)”, with abuse monitoring logs kept “for up to 30 days”.

Google Gemini trains while Keep Activity is on

Gemini uses your chats to train Google’s models whenever the Keep Activity setting is on, which it is unless you change it. Google’s Gemini Apps Privacy Hub¹², last updated 24 September 2026, says “Google uses your activity to provide, develop, and improve its services (including training generative AI models)” and that “A subset of chats are reviewed by human reviewers (including Google’s trained service providers)”.

Gemini’s retention depends on whether a human reviewed the chat. Activity deletes itself after 18 months by default, adjustable to 3 or 36 months or off¹². Reviewed chats sit outside that control. “Chats reviewed by human reviewers (and related data like your language, device type, location info, or feedback) are not deleted when you delete your activity. Instead, they are retained for up to three years.”

Turning Keep Activity off changes both answers, with limits. “Your future chats won’t appear in your Activity, and won’t be used to train our AI models, unless you choose to send Google feedback”, and chats held while the setting is off “are retained with your account for 72 hours”¹². Google also says “Even if your Keep Activity setting is off, Google still uses your chats to respond to you and help protect Google, our users, and the public, including with help from human reviewers”, and that “Your Gemini settings don’t control processing of your chats to create anonymized data to improve Google services”¹². Temporary chats “are not used to train Google’s AI models” and are also kept for 72 hours¹². I compared Gemini and the other tools built into the phone on features in Subtext vs Apple Writing Tools, Grammarly and ChatGPT.

Microsoft Copilot trains on most consumer accounts of the older app

On the older Copilot app, Microsoft uses consumer conversations for AI training unless you fall into an excluded group or opt out. The Privacy FAQ for Microsoft Copilot¹³ carries a banner saying it “applies only to the older version of the Microsoft Copilot app”. It says “Except for certain categories of users or users who have opted out, Microsoft uses data from Bing, MSN, Copilot, and interactions with ads on Microsoft for AI training.” The excluded groups are organisational accounts, people using Copilot inside Microsoft 365 apps on a Personal or Family subscription, people who are not signed in, people under 18, people who opted out, and “Users in Brazil, China (excluding Hong Kong), Israel, Nigeria, South Korea, and Vietnam”¹³.

The umbrella Microsoft Privacy Statement¹⁴, last updated September 2026, says “we may use your data to develop, train, and fine-tune our AI models, including large language models (LLMs).”

On the older app, the switch is under your profile. Microsoft’s Copilot privacy controls¹⁵ page, whose metadata shows an update on 21 August 2026 and which is also limited to the older app, says to select your profile icon, then your name, then Privacy, then “Training on conversation activity”; a second toggle covers voice. On the older app, conversation history is kept for 18 months by default¹³.

Microsoft says prompts in the newer Copilot app are not used to train foundation models

Microsoft’s pages for the newer Copilot app, whose metadata shows updates from 18 August 2026, say it does not use your prompts, responses or file contents to train foundation models, and the older controls page opens with a banner announcing that updated app¹⁵. The activity history page¹⁶, whose metadata shows an update on 18 August 2026, says “Prompts, responses, and your file contents when using the Microsoft Copilot app aren’t used to train foundation models”, and that optional customer feedback is not used to train them either. The privacy controls and choices page¹⁷, whose metadata shows an update on 17 September 2026, lists toggles for saved memories, a shared experience, web search and ads personalization, and none for training. Neither page I read gives a retention period, so the 18 months above is an older app figure. The activity history page¹⁶ sends readers who have not updated to the Privacy FAQ, and the privacy controls and choices page¹⁷ sends them to the older Copilot privacy controls page¹⁵, so Microsoft’s two sets of pages split by app and are not reconciled. The Privacy Statement¹⁴ quoted above still says Microsoft may train its AI models, and the pages I read do not say how that fits with the newer app’s answer. Check which app you have before you rely on either answer.

QuillBot and LanguageTool share a policy and take opposite positions

QuillBot and LanguageTool are both owned by Learneo, Inc. and sit under one privacy policy, and that policy treats them differently. I read the Learneo privacy policy¹⁸, updated 17 August 2026, on LanguageTool’s site because the copy on quillbot.com would not render for me. The general clause is wide. “We may use all of the data we collect (including your personal data) to train our algorithms, AI models and AI products.” The same policy then exempts LanguageTool. “Learneo does not use inputs from any LanguageTool user (Individual or Team Plan) to train its language algorithms or models. This is a default setting in LanguageTool that does not require an opt-out or any action by the user.”

QuillBot gets no such carve-out for individuals. Its help article How does Quillbot use my data?¹⁹ says “Your inputs and usage data are also used for both (1) training to improve the services generally and (2) personalizing the service to you.” Team Plan users are the exception¹⁸.

The same policy describes session replay. It says Learneo uses tools from providers outside Learneo “including Fullstory, Inc., Amplitude Inc., and Microsoft” that “may automatically collect” mouse movement and clicks and “Keystrokes (not for sensitive fields like passwords or credit card information)”, to understand user activity, fix performance problems, improve the user experience and prevent fraud and abuse¹⁸. That section does not mention training and does not name LanguageTool or QuillBot, though it sits in the shared policy.

QuillBot stores extension text by default, and only account holders can switch off training

QuillBot’s storage default changed with its privacy policy update of 8 November 2025, for extension users on version 4.37.0 and above²⁰. The company’s own note, About the November 8, 2025 Privacy Policy Update²⁰, says “We previously asked for your consent to store your text inputs … we will no longer require consent, and we’ll be storing text inputs for all extensions users by default, with the ability to opt out.” The two controls that followed, one for storing and one for training, sit in the extension under View All Settings, then Data and Privacy¹⁹. Storing is open to “all users” and training is “available to account-holders”, and a change “applies to your account settings from that point forward”¹⁹. For the desktop app, QuillBot says “In the future we’ll store text inputs from all desktop users” and that it will then notify users and “offer you the same controls”²⁰. QuillBot does “not allow any third-party service provider to use your data to train their AI models”¹⁹. That default matters to anyone choosing where to go after Grammarly.

DeepL Write trains on the free tier and deletes on Pro

DeepL Write’s free version uses your text to train DeepL’s models, and the paid version does not. DeepL’s privacy policy²¹, marked “Last update: October 2026” on 10 October 2026 (it read September 2026 when I first opened it on 4 October), covers the free Translator and Write in one section. “We process the content you upload and their translations or improvements for a limited period of time to train and improve our neural networks and algorithms. This also applies to corrections you make to our translation or improvement suggestions.” It gives no number for that period.

The Pro section, headed DeepL Translator Pro, DeepL API Pro and DeepL Write Pro, reads the opposite way. Texts “will not be permanently stored and will only be kept temporarily to the extent necessary for the production and transmission of the translation or improvement”, and “your texts will not be used to improve the quality of our services”²¹. For Write, the paid version is DeepL Write Pro, which DeepL’s About DeepL Write help page says “is not included in a DeepL Pro plan” and comes in a bundle that combines DeepL Pro and DeepL Write Pro²². The same page says “your texts won’t be stored so long as you are logged into your account”²². For private messages, the free tier is the riskier of the two. DeepL’s own terms for the free services, last updated November 2024, say “you may not use the Free Services for processing content containing confidential or personal data of any kind”, and that such content “is only allowed within a DeepL Pro subscription”²³. The terms name a DeepL Pro subscription, while the help page sells Write Pro as a separate product, so I cannot say whether Write Pro alone counts²²²³.

ProWritingAid says it never uses your writing to train its AI models

ProWritingAid says it does not train its AI models on your writing, and it states that on its help page. The article Does ProWritingAid use AI?²⁴, last updated 12 July 2026, says “ProWritingAid never uses your work or your writing to train our AI models or algorithms.” It says large language models power its assistive reports and generative features, and it names no model supplier or supplier terms.

The privacy policy²⁵, last modified 4 March 2026, says “When using ProWritingAid, your texts will not be used to improve the quality of our services.” On retention it says that for grammar checking “once checking has been completed, all submitted texts and their improvements will be deleted”, and that it does “not retain analysed text after processing except for documents you explicitly save in our Web Editor”²⁵.

The same policy also says “We use texts hand corrected by professional copyeditors to train our algorithms.” That sentence does not say whose texts are meant, so I report it and still count ProWritingAid with the tools that say no, on the strength of the help page and the policy line above. If that distinction matters to you, ask ProWritingAid.

Wordtune’s policy does not say whether your text trains its models

Wordtune’s privacy policy, from its owner AI21 Labs, does not say in plain words whether your text trains its models, and I am not going to decide the question for it. The Wordtune privacy policy²⁶, last updated 3 August 2025, says it uses data “to improve, enhance and develop the Services, AI21’s products and Services and to conduct research”, and may “aggregate or de-identify your Personal Information and further use it … to analyze and improve our Website, as well as our products and Services (including through the use of artificial intelligence)”. The only opt-out on data use it describes is by email, to stop future aggregation. Retention is “until we proactively delete it or you send a valid deletion request”²⁶. AI21’s own corporate privacy policy²⁷, last updated 2 August 2026, says “unless otherwise agreed in writing or required to provide the Services, we do not use your inputs or outputs to train AI21’s foundational or generally available AI models”, and it does not mention Wordtune. The table below marks it as not specific, because I will not guess either way. A missing answer is not a yes and it is not a no, so if training on your sentences matters to you, ask AI21 directly.

Tonely says its analysis stays on the device, and its store labels differ

Tonely answers the training question with a flat no, and says it does so by keeping text on the phone. Tonely’s privacy policy²⁸, effective 18 July 2025, says it “does not collect, store or transmit” your messages, keystrokes or analysis results, and that “All message analysis takes place entirely on your device. Tonely does not receive, store or use your communications to train or improve its AI models.” I cannot test the claim that analysis happens on the device from the outside; I am reporting what the policy states. Its store labels, which I opened on 10 October 2026, differ. The App Store page²⁹ shows “Data Not Collected”, a label the developer supplies, while the Google Play listing³⁰ lists an email address, used for account management, as collected and says “Data isn’t encrypted”. The policy says that at sign-in it may collect an email address and name if you share them through Apple, Google or Microsoft²⁸, so Apple’s label and the policy do not line up on that point.

ConfiText says it does not store drafts and relies on OpenAI’s terms against training

ConfiText also answers the training question with a flat no, by passing text through a supplier that may not train on it. ConfiText’s privacy policy³¹, effective 10 February 2026, takes the supplier route. “We do not log, save, or store your text drafts on our servers after the response is sent back to your device”, and “We utilize the OpenAI API strictly under terms that input data is not used by OpenAI to train their models and is not retained beyond the required time to generate a response.” The second half is a little stronger than OpenAI’s own API page, which allows abuse monitoring logs for up to 30 days¹¹. Its App Store label, which I opened on 10 October 2026, lists no User Content, lists Usage Data as used to track you, and lists Contact Info, Identifiers, Usage Data and Diagnostics as linked to you³². I read the same ConfiText policy for the piece on reading chat screenshots.

What Subtext does with your messages

Subtext does not use anything you send it to train an AI model, and its policy says the same about the companies that process your text for it. Subtext’s privacy policy³³, last updated 26 July 2026, reads “we do not use what you submit to train AI models.” It names Anthropic’s Claude as the model that reads and rewrites your text, Deepgram for turning a voice note into text and OpenAI only for the feature that reads text aloud, and says of those three “None of them may train on your content. Their commercial API terms rule it out.” Google Firebase, in the United States, holds accounts, conversations and settings³³. The policy also names Mixpanel, Adjust and RevenueCat and lists what each receives, which is product events, device and campaign identifiers, and user ID, email and subscription status³³. When web search is on, search terms drawn from your request go through Anthropic to third-party search providers, and you can switch that off in settings³³.

A voice note you dictate goes to Deepgram. Subtext sets Deepgram’s opt-out from using audio to improve its models, so Deepgram keeps the audio only as long as transcription takes³³. The policy says Anthropic deletes inputs and outputs within 30 days, and may keep a request its safety systems flag for up to 2 years and the related safety scores for up to 7 years. Subtext has no zero retention arrangement with Anthropic, Deepgram or OpenAI³³.

On its own servers, Subtext deletes a conversation five days after you last used it, or 90 days if you pin it, and deleting one yourself erases it immediately³³. A file still in processing is deleted within 24 hours at most, and account deletion completes within 14 days³³.

Subtext sends your text off your phone for processing

Subtext is not a keyboard, so it sees only the draft, screenshot or voice note you hand it, and that text does leave your phone for the providers its privacy policy names and Subtext’s own servers. Any tool that processes text on a server works this way. I found no independent audit of how Subtext handles messages, so what I describe here is what the policy says it does. If “never leaves the device” is your test, only Tonely, by its own account, states that its analysis stays on the phone, and Subtext does not pass. If “nobody trains on it and the app deletes it from its own servers five days after I last used it unless I pin it” is your test, Subtext passes, within the provider limits above. Where each tool fits is in what the best AI writing assistant for text messages is. The tone checker for text messages page shows what the app does instead.

Six assistants train on your text by default, with how to opt out and what is kept

Every entry restates a fact from the sections above or the policy it cites. “Default” means a personal account before you touch any setting.

Six more assistants say they do not train on your text, and Wordtune does not say

This table gives the same columns for the other seven. DeepL Write Pro and the newer Copilot app, which also say no (for the newer app, by Microsoft’s wording on foundation models), sit in the first table next to the tier and the app version that train.

Which AI writing assistant is the most private?

Private can mean two things here. Either the text never leaves the device, or a server does not train on it. Only Tonely, by its own account, states that its analysis stays on the phone. Apple does not say whether Writing Tools run on the device or in Private Cloud Compute, and the ChatGPT extension sends requests to ChatGPT once you enable it⁷. For a tool that runs on a server and does not train on your text, the group is LanguageTool, ProWritingAid, DeepL Write Pro, the newer Copilot app, ConfiText and Subtext, plus Grammarly, ChatGPT, Gemini and QuillBot once you have changed the setting. The newer Copilot app is in the group by Microsoft’s own wording, which covers foundation models.

On retention, Gemini’s three years for reviewed chats is the longest fixed period stated for the content of a conversation. Anthropic’s safety scores, which are not conversation content, can be kept for up to 7 years according to Subtext’s policy³³. Among the tools in the tables that keep a conversation by default so you can return to it, Subtext’s five days on its own servers is the shortest default period, though Anthropic deletes inputs and outputs within 30 days and may keep a flagged request for up to 2 years³³.

Turning off training covers what you write next and leaves some uses in place

Turning off training changes what happens to what you write next, and at several tools it leaves human review, anonymised or aggregate use and safety processing in place. ChatGPT says your new conversations will not train its models once Improve the model for everyone is off⁹, and Gemini says future chats will not train its models unless you choose to send Google feedback¹². Google says that even with Keep Activity off it still uses your chats to respond to you and help protect Google, its users and the public, including with help from human reviewers¹². Gemini’s settings “don’t control processing of your chats to create anonymized data”¹², Grammarly may still use writing statistics after you opt out¹, Apple may collect aggregated trends about processed content if you share Device Analytics⁴, and ProWritingAid says texts corrected by copyeditors train its algorithms²⁵. OpenAI’s Temporary chat article says a copy may be kept for up to 30 days for safety purposes¹⁰. The Learneo session replay section, which applies to LanguageTool too, does not mention training¹⁸.

Whatever you pick, search its policy for “train”, “improve”, “retain” and “opt”, read the date on the policy page, and do the same for this article.

Checked 4 October 2026, and again on 10 October 2026 where the Sources say so.

Subtext, the one tool here I co-founded, takes a draft, a screenshot of the conversation or a voice note, sends it to the providers its privacy policy names, uses none of it to train an AI model, and deletes the conversation from its own servers five days after you last used it, or 90 days if you pin it. Try Subtext in your browserGet the appTry Subtext in your browser

Sources

Numbered in order of first appearance. Each entry gives the date or dates I opened the page: 4 and 5 October 2026 for the first read, and 10 October 2026 for pages I reopened or added that day, including Apple’s Intelligence page, the Microsoft Privacy Statement, DeepL’s privacy policy and Subtext’s privacy policy. Three of OpenAI’s four pages would not load on 10 October, so they carry 4 October only. Where a page shows its own date, the entry gives it. Microsoft’s four Copilot support pages show no date to a reader, so for those the entry gives the update date in the page metadata, which for the privacy controls page (21 August 2026) is later than its ms.date (18 August 2026). For DeepL’s help page the date is the article’s edited field in the Help Center data, which a reader does not see.

1. Grammarly (Superhuman) Support. Product Improvement and Training Control. . Checked 4 and 10 October 2026.
2. Superhuman. Privacy Policy, effective 6 July 2026. . Checked 4 and 10 October 2026.
3. Grammarly (Superhuman) Support. Privacy and security FAQs. . Checked 4 and 10 October 2026.
4. Apple. Apple Intelligence & Privacy, dated 14 September 2026. . Checked 4 and 10 October 2026.
5. Apple Security Research. Expanding Private Cloud Compute, 8 June 2026. . Checked 5 October 2026.
6. Apple. Privacy features. . Checked 4 and 10 October 2026.
7. Apple. ChatGPT Extension & Privacy, dated 14 September 2026. . Checked 4 October 2026.
8. OpenAI. Privacy policy, updated 6 February 2026. . Checked 4 October 2026.
9. OpenAI Help Center. Data controls in ChatGPT. . Checked 4 October 2026.
10. OpenAI Help Center. Temporary chat in ChatGPT. . Checked 4 October 2026.
11. OpenAI. Your data (API documentation). . Checked 4 and 10 October 2026.
12. Google. Gemini Apps Privacy Hub, last updated 24 September 2026. . Checked 4 and 10 October 2026.
13. Microsoft Support. Privacy FAQ for Microsoft Copilot, page metadata updated 31 August 2026. . Checked 4 and 10 October 2026.
14. Microsoft. Microsoft Privacy Statement, last updated September 2026. . Checked 4 and 10 October 2026.
15. Microsoft Support. Microsoft Copilot privacy controls, page metadata updated 21 August 2026. . Checked 4 and 10 October 2026.
16. Microsoft Support. Microsoft Copilot for individuals: your activity history, page metadata updated 18 August 2026. . Checked 10 October 2026.
17. Microsoft Support. Microsoft Copilot for individuals: your privacy controls and choices, page metadata updated 17 September 2026. . Checked 10 October 2026.
18. Learneo, Inc. Privacy Policy, updated 17 August 2026, as published on LanguageTool. . Checked 4 and 10 October 2026.
19. Quillbot Help Center. How does Quillbot use my data? . Checked 4 and 10 October 2026.
20. Quillbot Help Center. About the November 8, 2025 Privacy Policy Update. . Checked 4 and 10 October 2026.
21. DeepL. Privacy Policy, last update October 2026 (September 2026 when first read on 4 October). . Checked 4 and 10 October 2026.
22. DeepL Help Center. About DeepL Write, article text last edited 4 December 2025. . Checked 10 October 2026.
23. DeepL. DeepL Free Services - Terms of Use, last update November 2024. . Checked 10 October 2026.
24. ProWritingAid Help Center. Does ProWritingAid use AI?, last updated 12 July 2026. . Checked 10 October 2026.
25. ProWritingAid (Orpheus Technology). Privacy Policy, last modified 4 March 2026. . Checked 10 October 2026.
26. AI21 Labs. Wordtune Privacy Policy, last updated 3 August 2025. . Checked 4 and 10 October 2026.
27. AI21 Labs. Privacy Policy, last updated 2 August 2026. . Checked 10 October 2026.
28. Tonely AI Ltd. Privacy Policy, effective 18 July 2025. . Checked 4 and 10 October 2026.
29. Apple App Store. Tonely AI App, version 1.1.2, shown as updated three days before I read the page on 10 October 2026, App Privacy label. . Checked 10 October 2026.
30. Google Play. Tonely, listing updated 6 October 2026, Data safety section. . Checked 10 October 2026.
31. ConfiText. Privacy Policy, effective 10 February 2026. . Checked 4 and 10 October 2026.
32. Apple App Store. ConfiText: AI Texting Assistant, version 2.2.16, App Privacy label. . Checked 10 October 2026.
33. Subtext. Terms, privacy and your account. Privacy policy last updated 26 July 2026. . Checked 4 and 10 October 2026.