Which AI is the most private? What each vendor publishes, in a dated table
Every page ranking this question ends with a product name. This one doesn't. If you are choosing an assistant for client material, the useful thing is each vendor's own terms on four points: whether it trains on what you type by default, what the opt-out is called, how long it keeps a chat, and who can open one. Below is that table for ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity and NotebookLM, every quotation taken from the vendor's page read on 6 November 2026, with the work-plan variants kept apart. Equerry's own row states only what its privacy page says.
Short answer: the vendors publish terms, not a ranking, and the terms differ by account more than by brand. The same product can train on your chats on a personal plan and not on a work plan; the same vendor can keep a chat for 30 days under one setting and five years under another. So the honest comparison is a table with the account type in it, and that is what follows. This page gives no verdict on whether any of these tools, Equerry included, is private, safe or confidential enough for your work; that stays with you, your firm's policy and your client's terms.
What each AI publishes about your data: the dated table
Two tables, because the question has two halves. The first is training: whether a personal account's chats go into the vendor's models by default, and what the switch is called. The second is retention and access: how long a chat is kept, who can open it, and what changes on a business or work plan. Every quotation was read at the vendor's own page on 6 November 2026, and a cell that says "not stated" means the pages read that day carry no sentence on it. The vendor's own date, where the page shows one, is given at the first quotation in each row and repeated where a row draws on a second page; the full list is in the sources at the end. OpenAI's help articles show only relative "Updated" stamps, so they carry no date here.
Scroll sideways to see every column.
| Assistant (personal account) | Trains on your inputs by default? In the vendor's words | The opt-out, as the vendor labels it | What the setting does not cover | Temporary or incognito mode, as stated |
|---|---|---|---|---|
| ChatGPT (Free, Plus, Pro) | "When you use our services for individuals (Free, Plus, Pro), we may use your content to train our models." | "To opt out, turn off Improve the model for everyone under Settings > Data controls in ChatGPT, or select Do not train on my content in our Privacy Portal." "Either option is sufficient for ChatGPT conversations and Codex tasks." | "If you choose to provide feedback, the entire conversation associated with that feedback may be used to train our models." "After you opt out, we won't use your new conversations to improve our models." | "Temporary chats are not used to improve OpenAI models while they remain temporary, even if you choose personalization." |
| Claude (Free, Pro, Max) | "We will use your chats and coding sessions (including to improve our models) if:" "You choose to allow us to use your chats and coding sessions to improve Claude." (Privacy Center, 16 March 2026.) The Privacy Policy (effective 10 September 2026) puts it as "We may use your Inputs and Outputs to train and improve Anthropic AI models" unless you opt out in your account settings. | Settings, then Privacy, then the toggle "Help Improve our AI models" (Privacy Center, 3 August 2026). The training article calls the same control "Model Improvement" in your "Privacy Settings". | Feedback (16 March 2026): "we will store the entire related conversation, including any content, custom styles or conversation preferences, in our secured back-end for up to 5 years." Also conversations "flagged for safety review". | "Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings." (16 March 2026.) |
| Gemini (personal Google Account) | Under the Keep Activity setting: "Google uses your activity to provide, develop, and improve its services (including training generative AI models), as well as to protect Google, its users, and the public with the help of human reviewers." (Privacy Hub, last updated October 26, 2026.) | The setting is named Keep Activity. With it off: "Temporary chats and chats you have when Keep Activity is off are retained with your account for 72 hours". | "Your Gemini settings don't control processing of your chats to create anonymized data to improve Google services." Separately: "Your audio and Gemini Live videos and screenshares aren't used to improve Google services by default." | "Temporary chats are not used to train Google's AI models." "You can also use temporary chats, which are not used to improve Google AI with help from human reviewers." |
| Microsoft Copilot (personal Microsoft account) | Updated app ("An updated version of the Microsoft Copilot app for web, desktop, and mobile devices is available as of August 18, 2026."): "Prompts, responses, and your file contents when using the Microsoft Copilot app aren't used to train foundation models." Older app, on a page that opens "The information in this article applies only to the older version of the Microsoft Copilot app.": "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." | Updated app (page undated): no training switch is described, because the page states the default above. Older app (page undated): "You can opt out of use of your conversation activity for model training at any time." | The Microsoft Privacy Statement (Last Updated September 2026) sits above both app pages: "As part of our efforts to improve and develop our products, we may use your data to develop, train, and fine-tune our AI models, including large language models (LLMs)." And: "Microsoft Copilot also uses prompts and related data to provide and improve services, including relevant advertising." | Not described on the Copilot app pages read (read 6 November 2026). |
| Perplexity (Free, Pro, Max) | "For Free, Perplexity Pro and Perplexity Max users, AI Data Retention is enabled by default." (Help Center, Last modified on August 31, 2026.) The Privacy Notice (Last updated: October 23, 2026) lists among its uses "Improve or create services and products, including our AI models." | Account settings, then the Preferences page, then the "AI data retention" toggle: "Toggle off to opt out of AI training data collection". | "Opt-outs only apply to data collected after the opt-out date". "Previously collected training data cannot be deleted or removed". | Privacy Notice, under Your Choices (Last updated: October 23, 2026): "Incognito increases your privacy by not saving your search activity across use sessions." |
| NotebookLM (Google calls it Gemini Notebook; personal account) | "The content in Gemini Notebook will not be used to directly train our foundational AI models, unless you choose to provide feedback." (Page undated.) | None needed on that sentence; feedback is the trigger. Google adds: "Please help us by not including any confidential or sensitive information in your feedback." | Feedback is "Reviewed by specially trained teams." "Reviewed feedback, included content, and related data are retained for up to 3 years, disconnected from your Google Account." | Not described on the page read (read 6 November 2026). |
| Equerry | Its privacy page: "We never train AI models on what you save." Its AI disclosure: every provider except Exa "operates under enterprise terms that prohibit using your data to train any model"; Exa receives "only the search queries and web addresses" and "under its own terms may keep them and use them to train its models." | No training switch, because there is no training on what you save. Web search is a Routine or chat you choose to run. | What you send is processed to make it searchable: "the text of your notes, and your voice memos, photos and PDFs so their text can be read." | None. What you send it is kept until you delete it. |
Training by default and the opt-out, vendor by vendor
Three patterns run through that table, and they matter more than any ranking. First, the default is set per account, and three vendors set it to on: OpenAI "may use your content", Perplexity's "AI Data Retention is enabled by default", and Google's Keep Activity covers "training generative AI models". Anthropic frames it as your choice, and Microsoft states for the updated app that prompts "aren't used to train foundation models", while its general Privacy Statement keeps the broader sentence. Second, feedback bypasses the switch on ChatGPT, Claude and Gemini Notebook: a thumbs up or down sends the whole conversation, in OpenAI's and Anthropic's words, "even if you have opted out" and "for up to 5 years". Third, the opt-out runs forward: OpenAI's "we won't use your new conversations", Perplexity's "Opt-outs only apply to data collected after the opt-out date", Anthropic's five-year de-identified retention of what was allowed before. The time to find the setting is before the paste.
Where the switches sit, as each vendor prints them: ChatGPT, Settings > Data controls > Improve the model for everyone, or the Privacy Portal's Do not train on my content; Claude, Settings, then Privacy, then Help Improve our AI models; Gemini, the Keep Activity setting; Perplexity, Account, then Preferences, then AI data retention. Step diagrams for the ChatGPT and Claude screens are on ChatGPT privacy and Claude data privacy; the Copilot app's Personalization screen is drawn on Copilot privacy.
Retention, business plans and the locked or incognito modes
Scroll sideways to see every column.
| Assistant | How long a chat is kept, as stated (personal account) | Who can open a chat, as stated | The business or work variant, in the vendor's words |
|---|---|---|---|
| ChatGPT | "Regular and archived chats remain saved to your account until you delete them or an applicable workspace retention policy removes them." On deletion, "OpenAI schedules it for permanent deletion from its systems within 30 days, unless the chat was already de-identified and disassociated from your account or OpenAI must retain it longer for security or legal obligations." Temporary chats: "OpenAI may keep a copy for up to 30 days for safety purposes." | "A limited number of authorized OpenAI personnel, as well as trusted service providers that are subject to confidentiality and security obligations, may access user content only as needed for these reasons: (1) investigating abuse or a security incident; (2) to provide support to you if you reach out to us with questions about your account, or to troubleshoot issues affecting the product experience; (3) to handle legal matters; or (4) to improve model performance (unless you have opted out)." And: "Please do not enter sensitive information that you would not want reviewed or used." | Enterprise privacy page (Updated: January 8, 2026): "We do not train our models on your data by default". Business: workspace admins "can view, access, export, and delete end user conversations in the workspace"; OpenAI's access "is limited to (1) authorized employees that require access for engineering support, investigating potential platform abuse, and legal compliance and (2) specialized third-party contractors who are bound by confidentiality and security obligations, solely to review for abuse and misuse." Enterprise: "Your workspace admins control how long your data is retained. Any deleted conversations are removed from our systems within 30 days, unless we are legally required to retain them." |
| Claude | Training allowed: "we may retain your data in a de-identified format for up to 5 years in our model training pipelines" (Privacy Center, 1 July 2026). Training declined, from Anthropic's 28 August 2025 post: "If you do not choose to provide your data for model training, you'll continue with our existing 30-day data retention period." A deleted chat: "Removed from your chat history immediately" and "Deleted from our back-end storage systems within 30 days." Safety-flagged chats: "We retain inputs and outputs for up to 2 years". | The Privacy Center retention article and the Privacy Policy read for this page (read 6 November 2026) carry no sentence naming who at Anthropic may open a chat; flagged conversations are the stated exception to the training setting. The sharing and access side is read in full on is Claude confidential? | Commercial article (18 August 2026): "By default, we will not use your inputs or outputs from our commercial products" to train, naming Claude for Work, the Anthropic API and Claude Gov; the exceptions are "If you explicitly report feedback or bugs to us" or if you "otherwise choose to allow us to use your data". |
| Gemini | Privacy Hub (last updated October 26, 2026): "Your activity is auto-deleted after 18 months." "You can change this to 3 or 36 months, or choose not to auto-delete." With Keep Activity off, chats are "retained with your account for 72 hours". "Reviewed chats are retained for up to three years." | Privacy Hub (October 26, 2026) and Privacy Notice (October 13, 2026): "Chats are also reviewed to keep Gemini safe and secure for everyone". "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." | "If you have a work or school Google Account, your use of Gemini Apps may be subject to different data handling terms." The consumer Hub says no more; the terms sit with your organisation's Workspace agreement. |
| Microsoft Copilot | Personal account, updated app: the activity-history page (undated) describes viewing, deleting and exporting your history and states no fixed period (read 6 November 2026). Older app, from the FAQ that applies "only to the older version": "By default, we store conversation activity for 18 months." Work account: under enterprise data protection (shown "Last updated on 11/04/2026") Copilot "applies your retention policies", so the period is your organisation's. | Personal account: no administrator; the record is yours in the app and on the Microsoft privacy dashboard. Work account, from "Data, Privacy, and Security for Microsoft Copilot" (shown "Last updated on 2026-11-03"): "To view and manage this stored data, admins can use Content search or Microsoft Purview." On human review: "While abuse monitoring, which includes human review of content, is available in Azure OpenAI, Microsoft Copilot services have opted out of it." | Enterprise data protection (shown "Last updated on 11/04/2026"): "Consistent with our other Copilot offers, the prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation models." "We won't use your data except as you instruct." Web queries are the stated exception: they go to Bing under "different data-handling practices". From the Microsoft Copilot privacy page (2026-11-03): "Your users can delete their Copilot activity history, which includes their prompts and the responses Copilot returns, by going to the My Account portal." |
| Perplexity | Privacy Notice (Last updated: October 23, 2026): "We keep your personal data for only as long as necessary to fulfil the purposes in this Notice unless a longer retention period is required or permitted by law." Self-service: "You can export or delete your data yourself from your account settings." | The Privacy Notice (October 23, 2026) states "We share your data with service providers." and lists law-enforcement and legal disclosures; it carries no sentence on human review of queries (read 6 November 2026). | The Privacy Notice "does not apply to data collected through Perplexity's Enterprise and API offerings". Help Center: "Perplexity Enterprise data is never used for AI training purposes." For Enterprise Pro and Max, "Uploaded files are retained for only 7 days", and organisations with enough seats "can set custom data retention periods and force deletion at any time". |
| NotebookLM (Gemini Notebook) | Feedback and its attached content: "retained for up to 3 years, disconnected from your Google Account." No other period is stated on the privacy page read (undated; read 6 November 2026). | Feedback only: "Reviewed by specially trained teams." | Qualifying Workspace and Education editions: "Your uploads, queries and the model's responses in Gemini Notebook will not be reviewed by human reviewers even when you provide thumbs up or down feedback, and will not be used to train AI models." Which editions qualify is a licence table, walked through on is NotebookLM private? |
| Equerry | Privacy page (effective October 18, 2026): notes, voice memos, photos, PDFs and your chats live "in your iCloud account, on Apple's infrastructure"; our servers keep "the text, as the searchable excerpts and facts". A forwarded or CC'd email: "our email provider keeps each message for 30 days, and we keep its text as searchable excerpts plus a short pointer to it. Your own copy stays in your own mailbox." On account deletion "Your data is deleted from our live database straight away" and backups within 40 days. | The providers named on the privacy page: "Each provider other than Meta and Exa processes your data solely to deliver its function to us"; Meta receives advertising-measurement reports and Exa only search queries and web addresses, each "under its own terms". "Help messages are never sent to AI providers": a person at Equerry reads those. Settings → Your memory "lists every fact Equerry has extracted from your content", and you can delete any fact or any note. | One plan, one set of terms. There is no inbox connection and no organisation tenant: it reads only what you send it. |
The work-plan column is where the question changes shape rather than gets answered. Four vendors state a work-plan training default of off: OpenAI, Anthropic, Microsoft and Perplexity, in the words quoted above; Google's consumer Hub says only that a work or school account "may be subject to different data handling terms." Two of them also describe narrower access on the work plan, OpenAI and Microsoft, and in the same move your employer gains the record. OpenAI's Business admins "can view, access, export, and delete end user conversations"; Microsoft's admins "can use Content search or Microsoft Purview"; Google's and Perplexity's enterprise terms are a contract your organisation holds. So "which is most private at work" splits into two questions with two owners: what the vendor does, from its page, and what your firm does, from its policy. Two guides read the second half plan by plan: is ChatGPT confidential? and is Claude confidential?
What the privacy rankings on page 1 say, and why this page ranks nothing
If you searched this question you have already seen the rankings. Three of them, dated as their pages show them on 6 November 2026, so you can weigh them as what they are, third-party opinion:
- Incogni, "Gen AI and LLM Data Privacy Ranking 2026." A data-removal company's study; the page shows no publication date. Its own description: "Incogni's researchers analyzed 13 AI platforms," scored on "11 criteria, divided into three categories," and "The data was collected from June 15 to July 6." It is the source most of the smaller ranking pages repackage.
- Captain Compliance, "2026 AI platforms Privacy Rankings." A compliance-software vendor's page, shown dated Jun 29, 2025; its text names Incogni's 2026 ranking while its link points to Incogni's 2025 edition. The first organic result for "most private ai" on the day this was written.
- Tom's Guide, "I compared the privacy of ChatGPT, Gemini, Claude and Perplexity" (7 December 2025). One writer's reading of the four chatbots' privacy practices at that date, with a pick at the end.
Each is a snapshot of terms that have moved since: Microsoft's page says an updated Copilot app "is available as of August 18, 2026" and states a different training sentence for it, Perplexity's notice is dated 23 October 2026, Google's Hub 26 October 2026. A ranking freezes the terms at its own date and then adds a weighting the reader didn't choose. This page skips the weighting. The cells above are the terms, with the dates, and the weighting is yours, because only you know whether your worry is training, retention, a reviewer, or an administrator.
Best AI for privacy: the reader who wants one answer
The phrasing "best AI for privacy" wants a product name, and the reason this page withholds one is in the tables: the terms are set by the account you sign in with, not by the brand. A personal ChatGPT account and a ChatGPT Enterprise seat read differently on every column. With Help Improve our AI models on, Anthropic "may retain your data in a de-identified format for up to 5 years in our model training pipelines", and "This retention period only applies to new or resumed chats after the model training setting has been enabled"; with it off, by Anthropic's post, you "continue with our existing 30-day data retention period." Copilot in the updated app and Copilot in the older app carry opposite training defaults. If you want one answer anyway, the honest version is a procedure: find the training switch, read the retention row for your account type, and treat feedback buttons as a send-to-vendor action. For a reader comparing memory rather than privacy, the companion page is which AI has the best memory?
Private AI assistant: a different search, a different shelf
"Private AI assistant" turns up a different set of pages from the ones above: lists of assistant apps that schedule, draft and act for you, where "private" means the product's own claim about itself rather than a term you can read. That is a different reader, shopping for an assistant rather than auditing a chatbot, and the vendor terms above don't settle it, because most of those apps run on the model providers above and inherit their API terms. The column that matters there is the one this page can't fill for them: what the app itself keeps, and what it reads of yours without being asked. For the assistant Equerry makes, that column is below, in its own privacy page's words.
What Equerry publishes about its own handling
Equerry is a personal assistant who helps you get things done, and the shape of its answer to this page's question is the shape of what reaches it. Nothing is read that you didn't send: an email thread you CC or forward to your own Equerry address; a PDF, an image, a link or text shared from your iPhone's share sheet; a note you type; a voice note you record in Equerry; and, only if you allow it, your calendar, read-only. No inbox is connected, so nothing arrives that you didn't put on the CC line or share yourself, and your own copy stays in your mailbox. The two rows above are everything its privacy page says on training and retention, quoted; here is what the terms look like in use, on this page's question.
| What you send | Your Equerry address, copied on the Marwick renewal thread so each reply arrives as it comes; the client's data-handling addendum, shared as a PDF from your iPhone; a voice note you record in Equerry on the walk back, listing the two clauses their counsel flagged. |
|---|---|
| What you ask | "What have you kept from the Marwick thread and the addendum, where did each line come from, and what does the addendum say about third-party services?" |
| What you get back | The addendum's third-party-services clause quoted back with its section number, the two clauses from your voice note beside it, and the thread's latest position on each. When an answer draws on a note you saved, it carries a source chip: tap it to see the passage Equerry used and open the original. For the thread: What Equerry remembered from it is listed in Settings → Your memory, each line marked as from that email, and you can delete any line. |
Replies that keep the Equerry address on the thread keep arriving, so the client's Thursday reply is there before Friday's call; anyone on the thread can see the address on the CC line, so forward instead where that would be awkward. To make what you send searchable, Equerry processes it, partly through the outside providers named on its AI disclosure, which says what each receives and states that "We never train AI models on what you save." There are no ads. It sends nothing, books nothing and buys nothing on your behalf. And it makes no claim about privilege, any professional rule, or whether what you send it is confidential: read its privacy policy the way this page reads the vendors', against your firm's policy and your client's terms. What differs is the unit of the decision, one thread or one file you chose, and a list you can read down and delete from.
Sources, all read 2026-11-06. OpenAI Help Center, as rendered pages (each shows a relative "Updated" stamp, not a date): How your data is used to improve model performance (article 5722486: the training sentences, the two opt-outs, feedback, temporary chats, the business default); Chat and file retention in ChatGPT (8983778: retention and deletion); How OpenAI handles data in consumer services (7039943: authorized human access, the sensitive-information request). OpenAI: Enterprise privacy at OpenAI (Updated: January 8, 2026: the Business and Enterprise access and retention sentences). Anthropic Privacy Center: Is my data used for model training? (consumer, 16 March 2026: the "if" conditions, feedback, Incognito); Is my data used for model training? (commercial, 18 August 2026); How long do you store my data? (1 July 2026: the five-year, 30-day deletion and two-year sentences); How do I change my model improvement privacy settings? (3 August 2026: the toggle label). Anthropic: Privacy Policy (effective 10 September 2026); Updates to Consumer Terms and Privacy Policy (28 August 2025: the 30-day sentence). Google: Gemini Apps Privacy Hub (Hub and privacy questions last updated October 26, 2026; Privacy Notice October 13, 2026: Keep Activity, 72 hours, 18 months, human reviewers, three years, temporary chats, work or school accounts); Gemini Notebook privacy and terms (undated: training, feedback, three years, Workspace editions). Microsoft Learn: Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat (shown "Last updated on 11/04/2026"); Data, Privacy, and Security for Microsoft Copilot (shown "Last updated on 2026-11-03": Content search and Purview, My Account deletion, abuse monitoring). Microsoft Support: Microsoft Copilot app: your activity history (undated; the 18 August 2026 updated-app notice and the training sentence; applies "only when you sign in with a Microsoft account"); Privacy FAQ for Microsoft Copilot (undated; opens "The information in this article applies only to the older version of the Microsoft Copilot app.": the older-app training, opt-out and 18-month sentences). Microsoft: Microsoft Privacy Statement (Last Updated September 2026). Perplexity: Perplexity Privacy Notice (Last updated: October 23, 2026: scope, uses, retention, service providers, and the Incognito sentence under Your Choices); Data Collection at Perplexity (Last modified on August 31, 2026). Rankings named above: Incogni (undated page); Captain Compliance (shown dated Jun 29, 2025; links Incogni's 2025 edition); Tom's Guide (7 December 2025). Equerry's statements come from its own privacy policy (effective October 18, 2026) and AI disclosure (effective October 7, 2026). Every vendor changes these pages often; where a vendor's current page disagrees with this one, the vendor's page is right. This page is general information, not legal or compliance advice, and it ranks nothing.
Frequently asked questions
Which AI is the most private?
No vendor page answers that, and this page doesn't either; it sets each vendor's published terms side by side with the date they were read. On 6 November 2026 the training defaults for a personal account read: OpenAI, "When you use our services for individuals (Free, Plus, Pro), we may use your content to train our models"; Anthropic, chats are used "if" "You choose to allow us to use your chats and coding sessions to improve Claude"; Google, "Google uses your activity to provide, develop, and improve its services (including training generative AI models)"; Microsoft, for the updated Copilot app, "Prompts, responses, and your file contents when using the Microsoft Copilot app aren't used to train foundation models"; Perplexity, "For Free, Perplexity Pro and Perplexity Max users, AI Data Retention is enabled by default." Retention, human review and work plans are separate rows in the tables above, and each vendor's own page is the authority if it has changed since.
Which AI does not train on your data?
By default, on the pages read 6 November 2026, the work and business plans are the ones that say so: OpenAI, "By default, we don't use inputs or outputs from ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, or our API to improve our models"; Anthropic, "By default, we will not use your inputs or outputs from our commercial products" to train; Microsoft, under enterprise data protection, "the prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation models"; Perplexity, "Perplexity Enterprise data is never used for AI training purposes"; Google, for qualifying Workspace editions of Gemini Notebook, uploads, queries and responses "will not be used to train AI models." On personal accounts it is a setting (ChatGPT, Claude, Perplexity), a default Microsoft states for the updated Copilot app, or activity Google uses unless Keep Activity is off. Feedback is the common exception: a thumbs up or down can send the whole conversation on ChatGPT, Claude and Gemini Notebook whatever the setting says.
Is Claude more private than ChatGPT?
This page won't rule on it; here are the two sets of terms, read 6 November 2026. Training: OpenAI "may use your content to train our models" on Free, Plus and Pro unless you turn off Improve the model for everyone; Anthropic uses consumer chats "if" "You choose to allow us to use your chats and coding sessions to improve Claude", the Help Improve our AI models toggle under Settings, then Privacy. Retention: OpenAI keeps saved chats "until you delete them" and then "schedules it for permanent deletion from its systems within 30 days"; Anthropic "may retain your data in a de-identified format for up to 5 years" when training is allowed, and a deleted chat is "Deleted from our back-end storage systems within 30 days." Access: OpenAI says "A limited number of authorized OpenAI personnel, as well as trusted service providers that are subject to confidentiality and security obligations, may access user content only as needed" for four listed reasons; Anthropic's pages read for this comparison carry no equivalent sentence. Which set fits your work is for you and your firm's policy. The longer reads are /guides/chatgpt-privacy and /guides/claude-data-privacy.
What is the best AI for privacy at work?
Not a question any vendor answers, and not one this page answers. What the vendors publish for work accounts, read 6 November 2026: ChatGPT Business admins "can view, access, export, and delete end user conversations in the workspace", and on Enterprise "Your workspace admins control how long your data is retained"; Claude for Work runs on the commercial default, training off unless you report feedback or opt in; Microsoft Copilot at work logs prompts and responses where "admins can use Content search or Microsoft Purview" to view them, and "Microsoft Copilot services have opted out of" the human-review abuse monitoring available in Azure OpenAI; Perplexity Enterprise Pro and Max say "Uploaded files are retained for only 7 days"; Gemini on a work or school account "may be subject to different data handling terms." In every case the work plan moves part of the question from the vendor to your employer, whose policy and your client's terms decide what goes in.
Does Gemini train on your conversations?
Google's Gemini Apps Privacy Hub, read 6 November 2026 (last updated October 26, 2026), says of the Keep Activity setting: "Google uses your activity to provide, develop, and improve its services (including training generative AI models), as well as to protect Google, its users, and the public with the help of human reviewers." With the setting off, "Temporary chats and chats you have when Keep Activity is off are retained with your account for 72 hours", and "Temporary chats are not used to train Google's AI models." Two sentences sit outside the setting: "Your Gemini settings don't control processing of your chats to create anonymized data to improve Google services", and "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." Reviewed chats "are retained for up to three years." Work or school accounts are covered by different terms Google does not spell out on that page.
Last updated 2026-11-06.
All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.