{"slug": "gemini-training-opt-out-how-chatgpt-and-claude-compare", "title": "Gemini Training Opt-Out: How ChatGPT and Claude Compare", "summary": "Google announced on November 16, 2026, that Gemini users in the US can keep chat History on while switching off the use of chat content for training, with a global rollout promised over the coming weeks. Google lists chat-content training as on by default for users over 18 and the separate audio, video and screen-sharing control as off by default, while temporary conversations and History-off chats can still be retained for around two months. The comparison notes that OpenAI may retain a temporary chat for up to 30 days for safety, Claude's default Incognito period is 30 days, and Anthropic retains feedback-linked data for five years.", "body_md": "If you want to reopen a Gemini conversation tomorrow without contributing it to future generative AI training, Google is rolling out a way to do both. Leave History on and switch off the use of chat content for training. The distinction matters to anyone who uses an assistant as a working notebook, from a freelancer drafting interview questions to a researcher returning to a long calculation.\n\nGoogle’s US announcement is dated November 16, 2026. It promises a global rollout over the coming weeks, so the controls visible in one account on November 19 may differ from another. OpenAI and Anthropic also separate saved conversations from a personal training preference. Their rules diverge when you start a temporary conversation, send feedback or move a workflow into a business workspace. [Google’s Gemini controls announcement](https://blog.google/innovation-and-ai/products/gemini-app/updated-controls/)\n\n1. 01Gemini can retain a chat you exclude from trainingWith the new controls, keep History on and turn off training on chat content. Live audio, video and screen sharing have a separate setting.\n2. 02Temporary conversations still have a retention periodGoogle specifies around two months under its new rules. OpenAI may retain a temporary chat for up to 30 days for safety. Claude’s default Incognito period is 30 days.\n3. 03Feedback is a separate disclosureGoogle can collect recent conversation context with feedback. OpenAI may use the entire associated conversation. Anthropic retains feedback-linked data for five years.\n4. 04Check the account paying for the workProtected Workspace editions, managed ChatGPT and Claude workspaces, and API projects follow their own terms. The consumer chat switch cannot configure an API application.\n\n## 01 — GeminiKeep the thread and change its *training use*\n\nIn a personal Gemini account with the update, open Settings, then History. Leave History enabled, open the control for chat content including text and uploads, uncheck it and confirm. Google’s help page describes the same category as text and attachments. The media control for audio and Gemini Live video and screen shares is separate. Google lists chat-content training as on by default for users over 18 and the media option as off by default. [Google’s history settings guide](https://support.google.com/gemini/answer/13278892)\n\nIf you want a shorter personal archive as well, set a deletion interval. The default for saved history is 18 months. The other choices are three months, 36 months or no automatic deletion. You can also delete an individual conversation. Choosing three months limits the visible archive. Clearing the training control tells Google how it may use future eligible chats. A person can reasonably choose both.\n\nAn interview outline illustrates the difference. Keep the thread if you will need its edits before the interview. Turn off chat-content training if that is your preference. If the work uses a connector, also identify which service keeps the underlying material. Google says the Gemini setting applies to Gemini history and leaves information saved by other Google services under those services’ controls. Our [guide to Tasks in Gemini](https://www.digitalapplied.com/blog/gemini-spark-free-tasks-in-gemini) covers the connected activities an assistant can carry out.\n\n## 02 — RetentionWhy a hidden chat can *still exist*\n\nGoogle previously described a retention window of up to 72 hours for Temporary Chats. Under the new policy, both a temporary conversation and a future chat made with History off can be kept with the account for around two months. Google says it needs this period to respond, handle feedback and protect users and the service. Temporary chats are excluded from training. For History-off chats, Google explicitly qualifies its exclusion with an exception for feedback. [Google’s previous Temporary Chat announcement](https://blog.google/products-and-platforms/products/gemini/temporary-chats-privacy-controls/) and [current Gemini Apps Privacy Hub](https://support.google.com/gemini/answer/13594961)\n\nA separate limit applies to a subset of material seen by reviewers. Google separates reviewed conversations from the account and may keep them and related data for up to three years, including after the user deletes the history. The three-year figure describes reviewed material. It is neither the life of every Gemini conversation nor a three-year setting you need to change before opting out.\n\nThe feedback boundary is especially easy to miss. With History off, Google says a feedback submission can include the preceding 24 hours of chats, attachments and data from connected apps. Turning chat-content training off still allows safety review and other service improvements. A sanitized reproduction of a problem can give support the relevant detail without including a client’s original attachment.\n\n## 03 — ChatGPTSave regular chats or open a *temporary one*\n\nOpenAI documented the ability to keep history while declining training by May 6, 2026. Its guidance, checked November 19, says data sharing is on by default on personal Free, Plus and Pro workspaces. In Settings, open Data controls and turn off Improve the model for everyone. That signed-in choice follows the account across web and mobile. Normal chats stay available. Archiving a conversation changes its placement and leaves the training preference intact. [OpenAI’s May privacy explanation](https://openai.com/index/how-chatgpt-protects-privacy/) and [current ChatGPT data controls](https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt)\n\nTemporary Chat serves a different purpose. While it remains temporary, it stays out of history, creates no new memories and is excluded from model improvement. OpenAI may keep a copy for up to 30 days for safety. Its help page allows a temporary chat to be saved as an ordinary conversation. A saved session then follows the account’s normal preferences. A personalized temporary chat can also use existing context, so choose the non-personalized option if the task needs a fresh start. [OpenAI’s Temporary Chat FAQ](https://help.openai.com/en/articles/8914046-temporary-chat-faq)\n\nDeleting a regular chat removes it immediately from view and schedules system deletion within 30 days, subject to security, legal and prior de-identification exceptions. Feedback can authorize use of the entire associated conversation even after a personal training opt-out. OpenAI also documents a narrow exception for personal Trusted Access Daybreak sessions, whose content it may use to improve models. Regular personal chats outside those sessions follow the account preference. [OpenAI’s chat retention policy](https://help.openai.com/en/articles/8983778-chat-and-file-retention-in-chatgpt)\n\n## 04 — ClaudeWhere model improvement and *deletion differ*\n\nAnthropic announced a consumer choice on August 28, 2025. New users were asked at signup and existing users were asked to select a preference. In its current help guide the switch is under Settings, then Privacy, as Help Improve our AI models. It covers personal conversations and Claude Code sessions run with the corresponding consumer account. Turning it off removes earlier stored and new sessions from future training runs. Runs already underway and models already trained are unaffected. [Anthropic’s consumer policy announcement](https://www.anthropic.com/news/updates-to-our-consumer-terms) and [model improvement settings](https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings)\n\nWhen someone allows training, Anthropic says it may keep de-identified data from new or resumed chats for up to five years in its training pipelines. That clock is different from deleting a saved conversation. A deletion takes effect immediately in the chat list, with removal from back-end storage within 30 days under the standard process. Feedback-linked data has its own five-year retention period. Safety-flagged content and legal obligations can have separate rules. [Anthropic’s consumer retention policy](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data)\n\nClaude’s Incognito mode leaves a conversation out of history, memory and ordinary model training. The default retention period is 30 days. Anthropic also notes that a longer organization retention setting may apply on Enterprise. That matters when identical-looking conversations take place in a personal account on a phone and a managed workspace at a desk. [Claude’s Incognito guide](https://support.claude.com/en/articles/12260368-use-incognito-chats)\n\n## 05 — DecisionChoose the setting for the *actual task*\n\nConsider a freelance researcher doing three pieces of work. Monday’s outline uses public sources and will be reopened on Friday. The relevant setting is a saved personal chat with training disabled. Tuesday’s one-off question needs no ongoing context. A temporary conversation can keep the sidebar and memory clear, subject to the provider’s retention period. Wednesday’s draft includes an unpublished client brief. Its handling depends on what the client authorized and which workspace is covered by that agreement.\n\nIf the assistant makes a factual error in the client brief, rating the response raises a fourth question. The provider may receive the surrounding conversation, even when the mistake fits into one sentence. Re-create the failure with an invented company name and public numbers if you want to report it. A team answering similar questions externally can use our [guide to customer questions about AI content privacy](https://www.digitalapplied.com/blog/customer-questions-ai-content-privacy) to explain its own practice.\n\n## 06 — Work accountsFollow the terms of the *workspace*\n\nFor Google Workspace, the relevant question is whether the organization’s license includes Gemini as a core service with enterprise-grade data protections. Google says chats and uploaded files on the protected service receive no human review or use to improve generative AI models. With Gemini as an additional service, different terms apply and chats may be reviewed and used to improve Google products. Having a company email address alone does not resolve which applies. [Google’s Workspace edition comparison](https://support.google.com/gemini/answer/14620100)\n\nOpenAI excludes ChatGPT Business, Enterprise and Edu inputs and outputs from training by default, as it does for its API platform. Anthropic likewise excludes Claude for Work, including Team and Enterprise, and the Anthropic API by default. Anthropic allows exceptions for material explicitly reported as feedback or otherwise shared for training. A team owner reviewing a feedback policy should check who can submit a rating and what that rating transmits. [OpenAI’s business data commitments](https://openai.com/business-data/) and [Anthropic’s commercial training policy](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training)\n\nA purchasing requirement such as “no model training” leaves several legitimate decisions open. An organization may need a shared archive for a project, a 30-day deletion schedule, or a contractual right to a different retention policy. Write the required period and the applicable product next to the training commitment. Our [comparison of agent-platform retention rules](https://www.digitalapplied.com/blog/ai-agent-platform-data-retention-eligibility-census) shows why eligibility can change across features.\n\n## 07 — Developer accessCheck where the *request runs*\n\nA team that turns a chat prototype into an application crosses another boundary. The Gemini API terms generally let Google use submissions and generated responses from unpaid services to improve products and machine-learning technology. The paid-service data-use terms exclude prompts and responses from product improvement, while permitting limited logging for policy enforcement and legal purposes. Identify the Google Cloud project and billing status for the request. A personal Gemini subscription or preference alone cannot establish that status. [Gemini API Additional Terms of Service](https://ai.google.dev/gemini-api/terms)\n\nThe same terms make a geographic distinction. For a developer in the European Economic Area, Switzerland or the United Kingdom, the paid-service data-use provisions also apply to Google AI Studio and unpaid Gemini API quota. Separately, Google permits only paid services for API clients made available to users in those regions. A developer testing in London and a company releasing an app to London customers have two related but distinct questions to resolve.\n\nFor OpenAI and Anthropic, default API training exclusions still leave endpoint and model retention rules to check. An organization requiring zero data retention needs the appropriate agreement and a compatible deployment. See our [analysis of Anthropic’s covered-model retention policy](https://www.digitalapplied.com/blog/fable-5-30-day-data-retention-zdr-enterprise-2026) for an example. Our [AI transformation team](https://www.digitalapplied.com/services/ai-transformation) helps map these requirements to the actual application and workspace.\n\n### Make the promise specific\n\nFor a personal drafting session, identify whether you need the chat later, set the training preference and choose how long the archive should last. For a client workflow, record the actual provider, account, product and retention rule. That gives the person handling next week’s document a precise answer even when a vendor offers several products under the same name.", "url": "https://wpnews.pro/news/gemini-training-opt-out-how-chatgpt-and-claude-compare", "canonical_source": "https://www.digitalapplied.com/blog/gemini-training-opt-out-vs-chatgpt-claude", "published_at": "2026-11-19 00:00:00+00:00", "updated_at": "2026-11-19 09:45:50.892644+00:00", "lang": "en", "topics": ["ai-policy", "generative-ai", "ai-products"], "entities": ["Google", "Gemini", "OpenAI", "ChatGPT", "Anthropic", "Claude"], "also_reported_by": [], "takeaway": null, "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/gemini-training-opt-out-how-chatgpt-and-claude-compare", "markdown": "https://wpnews.pro/news/gemini-training-opt-out-how-chatgpt-and-claude-compare.md", "text": "https://wpnews.pro/news/gemini-training-opt-out-how-chatgpt-and-claude-compare.txt", "jsonld": "https://wpnews.pro/news/gemini-training-opt-out-how-chatgpt-and-claude-compare.jsonld"}}