MorseSign inStart freeOpen Morse

What AI chatbot feedback buttons send: thumbs up, thumbs down, and report at 10 apps

The Morse teamPublished 11 min read

AI chatbot feedback buttons send more than a rating. As of November 2026, a thumbs up or thumbs down at ChatGPT, Claude, Gemini, and Qwen Chat sends the conversation behind the reply along with it, Mistral's Vibe sends the exchange you rated, and at ChatGPT, Grok, Qwen Chat, and Vibe your feedback can be used for training even after you've turned training off. Claude keeps a rated conversation for up to 5 years, Gemini keeps reviewed feedback for up to 3 years, and at ChatGPT and Duck.ai the button you press to report a harmful reply is the same thumbs down.

On November 20, 2026, we read the feedback, training, retention, and reporting pages of 10 AI chat apps: ChatGPT, Claude, Gemini, Grok, Microsoft Copilot, Meta AI, Mistral Vibe, Qwen Chat, Duck.ai, and Proton's Lumo. For each, we noted what a rating sends, whether it gets past a training opt-out, how long it's kept, and who may read it. Where a page didn't say, we say so. Microsoft has one of the most careful flows. Copilot collects nothing beyond the thumb until you press submit, and asks your permission for the prompt and chat history.

Morse is a private AI chat app with GPT, Claude, Gemini, DeepSeek, and uncensored models, encrypted so even Morse can't read your chats. Morse doesn't train AI on your chats, and its privacy policy says no AI provider trains on what you send through Morse. Your chat history is stored in your browser, encrypted, not on Morse's servers.

What does a thumbs up or thumbs down send at 10 AI apps?

At most of them, the conversation the rated reply came from. The table sums up each company's own pages, read November 20, 2026, for personal accounts. Every page is named in the Sources list at the end.

AppWhat a rating sendsGets past a training opt-out?How long it's keptWho may read it
ChatGPTOpenAI's model-improvement page says "the entire conversation associated with that feedback"Yes, on Free, Plus, and Pro. No on Business, Enterprise, and EduNot stated on the pages we readNot stated for feedback. Staff and service providers may access content "to improve model performance (unless you have opted out)"
ClaudeAnthropic says "the entire related conversation," with custom styles and preferences, but not raw connector contentYes: material "explicitly reported" through feedback is used after an opt-out (privacy policy)Up to 5 yearsDe-linked from your user ID first. Only designated Trust and Safety staff may access it
GeminiThe "associated conversation," plus files and Connected Apps content. With History off, "the last 24 hours of your chats"Yes: feedback is the exception to Google's training settingUp to 3 years, disconnected from your Google AccountTrained review teams at Google
GrokYour rating of an output with "the thumbs-up/thumbs-down icons" (privacy policy). Whether the chat goes too isn't statedYes: feedback "may be used for training purposes" after an opt-out (FAQ)Not statedNot stated for feedback. In general, a limited number of "authorized personnel" may review conversations
Microsoft CopilotOnly the thumb, until you submit. Then, with your permission, the prompt, the response, and content samples such as your chat historyNot used to train the foundation models behind CopilotNot statedOn work or school accounts, your IT admin, with your name
Meta AIFeedback counts as part of your "Content" in Meta's AI termsNo separate feedback rule on the pages we readNot stated; the terms say "the AIs will sometimes retain and use that information"Review "may be automated or manual (human)"
Mistral VibeYour rating plus "the related input and output"Yes, "whichever subscription plan you adhere to," for a rating with a commentNot stated for feedbackDe-identified first. A report opens the content to Mistral team members
Qwen ChatQwen's terms say it will "retain the corresponding conversation content"Yes: feedback "will still be used for model training"Not statedNot stated. Qwen "may use it in any manner we choose"
Duck.aiA conversation you choose to share with the feedbackDuck.ai doesn't train on chats; DuckDuckGo stores and reviews shared ones "to improve our products"Not statedDuckDuckGo
LumoAnonymized feedback, only on responses from the Apertus 1.5 modelLumo doesn't use chats for training, except feedback you choose to give on Apertus 1.5Not statedProton and "the university research teams behind Apertus"

Five of the 10 attach conversation content to a rating on their own (ChatGPT, Claude, Gemini, Mistral Vibe, and Qwen Chat), two ask first (Copilot and Duck.ai), and Grok's and Meta's pages don't say. Four say outright that a rating gets past a training switch you turned off, and Claude and Gemini reach the same place by another route. Copilot and Lumo are the most careful, in different ways. Microsoft asks before each extra piece, and Proton accepts feedback for only one model and anonymizes it before Proton or the researchers see it.

Does a thumbs up override your training opt-out?

Yes, at six of the 10: ChatGPT, Grok, Qwen Chat, Mistral Vibe, Claude, and Gemini. The first four say outright in their help pages or terms that a rating can be used for training after you switch training off, and Qwen writes it into the opt-out itself. Mistral's rule covers a rating with a comment and applies "whichever subscription plan you adhere to." Claude gets there through material you report, and Gemini through its training setting. Each company's own wording is quoted on our page about which AI companies train on your data.

Business plans are where the rule flips at OpenAI. Its help page says that on ChatGPT Business, Enterprise, and Edu, "response feedback (submitting thumbs up or thumbs down) does not permit training on the associated conversation," and an admin can switch ratings off entirely under Workspace settings, Allow response feedback. Microsoft's Copilot pages say it doesn't use feedback to train the foundation models behind Copilot.

Does rating one reply send the whole conversation?

At ChatGPT, Claude, Gemini, and Qwen Chat, yes. OpenAI and Anthropic both use the word "entire." Anthropic's help article on model training says a thumbs rating means "we will store the entire related conversation, including any content, custom styles or conversation preferences." Qwen's terms say that if you rate an output, "we will retain the corresponding conversation content and store it as part of the Feedback you provide."

Gemini can send more than the chat you rated. Google's help page on sending feedback says the associated conversation "is added as part of your feedback," along with files and images you uploaded and "your personal content that Gemini got from Connected Apps." When Gemini History is off, the privacy hub says feedback brings in context "including the last 24 hours of your chats," so a rating on one chat can carry the others from that day.

Mistral sends less. Its page on who can view your conversations says it uses your rating and "bits of your conversation," which its feedback page defines as "the related input and output." Microsoft sends nothing extra unless you agree: its Copilot feedback page says "Until you submit your feedback, no data is collected other than your initial thumbs-up or thumbs-down choice." If you then agree to share more, the content samples can include the history of your chat with Copilot, and log files that "might contain the content of your files, your name or email address, your IP address."

How long do AI companies keep your feedback?

Anthropic and Google are the only two of the 10 that publish a number. Anthropic's retention article says that where you've given feedback through the thumbs buttons or a bug report, "we retain data associated with that submission for 5 years." Google's privacy hub says reviewed feedback, associated conversations, and related data are "retained for up to 3 years," and that chats reviewed by human reviewers, with related data like your feedback, "are not deleted when you delete your history."

Anthropic says it doesn't combine feedback "with your other conversations with Claude," so the rated copy is stored apart from your history, and its pages don't say that deleting the chat removes it. OpenAI, xAI, Mistral, Qwen, and Microsoft give no period for feedback on the pages we read. Microsoft does let you see what you've sent: its feedback page points to My Feedback in the Microsoft feedback portal. How long each app keeps ordinary chats is in our guide to whether AI chats are private.

Who reads the chats you rate?

People, at several of them, often with your name removed first. Google's privacy hub lists the first thing that happens to Gemini feedback as "Reviewed by specially trained teams," and says human review is "necessary to help identify, address, and report potential problems raised in feedback." Anthropic's page on protecting Claude users' data says employees can't see your conversations by default unless you share them as feedback or a Usage Policy review requires it, and then "only designated members of our Trust & Safety team may access this data on a need-to-know basis."

Most strip the account before the content is used. Anthropic's model-training article says it de-links feedback from your user ID, "e.g. email address." Mistral's feedback page says it will "separate your feedback from your account using de-identify methods" and use it to "create instructions and datasets." Google's privacy hub says reviewed feedback is kept "disconnected from your Google Account." What you wrote stays after de-linking, so a rated chat that names your employer or your town still describes you. Our explainer on whether AI chats are anonymized covers what that kind of stripping leaves, and who reviews AI chats covers the contractors behind some of it.

Work accounts add a reader. Microsoft's Copilot feedback page says that on a work or school account, "your IT admin can view your feedback, including your name and all content samples that are collected," and can delete it, though Microsoft "might have already seen and started working on your feedback" by then.

Is the report button different from a thumbs down?

At ChatGPT and Duck.ai, it's the same button. OpenAI's help page on reporting content tells you to report a conversation this way: "Tap thumbs down (👎) under a ChatGPT message," then pick "Safety or Legal concern," then "Follow the prompts to submit your feedback." Duck.ai's terms say to report an output that breaks its rules or the law "by clicking the thumbs-down button and submitting feedback with the 'Harmful or illegal' option selected." At ChatGPT, that means a safety report travels down the same path as any rating, and OpenAI's pages don't carve reports out of its rule that feedback can put the entire conversation into training.

Other apps keep reporting separate:

  • Gemini: below a response, More, then Report legal issue, apart from the Good response and Bad response buttons.
  • Copilot: the flag icon below each response, or a long press and Report in the app, according to Microsoft's privacy FAQ for the older Copilot app.
  • Mistral Vibe: reports go to its support team, and Mistral's page on who can view your conversations says doing so grants "a limited number of authorized and competent" team members "the right to access the reported content to improve the Services."

Expect a person to read a report. At Anthropic, a report is also one of the two cases where your chat is used for model improvement after an opt-out.

How do you rate a reply without handing over the chat?

Rate only chats you'd be comfortable with a reviewer reading, and check what the form attaches before you press submit. A few app-specific steps help:

  • Gemini: choosing a reason, typing a comment, and attaching uploaded files are each optional on the feedback form, so leave the files out. With History off, remember that a rating pulls in the last 24 hours of chats. Our guide to whether Gemini is private covers the History setting.
  • Copilot: decline the extra data when the feedback pane asks, and the rating goes in on its own.
  • Lumo: Proton describes feedback only for responses from Apertus 1.5, and anonymizes it before sharing it.
  • ChatGPT: if you're on a Business, Enterprise, or Edu workspace, a rating doesn't permit training. On a personal plan, it does, whatever your setting says. The rest of the ChatGPT picture is in does ChatGPT train on your data.
  • Claude: leave out names and numbers you wouldn't want kept five years.

OpenAI's consumer data article puts the general rule in one line: "Please do not enter sensitive information that you would not want reviewed or used."

How Morse keeps your chats out of AI training

At Morse, your chat history is stored in your browser, encrypted, not on Morse's servers, and Morse's servers keep no log of your messages. Morse doesn't train AI on your chats, sell your data, or show you ads, and no AI provider trains on what you send through Morse.

You pick a privacy mode for each chat, depending on who you'd rather never see it:

  • E2EE, for a chat you'd never want anyone else to read. Your browser checks the enclave's hardware attestation, then encrypts the message so only that verified enclave can read it. Morse's servers relay text they can't read, and every E2EE reply has a lock you can open to see each check that passed. E2EE chats come with Pro.
  • Private, the default, for open-weight models on hosts that agreed to keep nothing. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it. For web search, use a Private chat.
  • Anonymous, for GPT, Claude, and Gemini without the provider knowing it's you. Morse sends your message under its own account, so the model provider never learns your name, email, account, or IP address. The provider still reads the message to answer it, without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so leave out details that identify you.

Your chats are stored on your devices, so Morse has no copy of them to give authorities, and what Morse could be required to hand over is limited to account records. To keep them if you clear your browser or lose the device, save a backup to a file from Settings > Data.

Send a first message from the Morse home page without an account, or start free. Pro is $19 a month, with unlimited messages on Pro models, your choice of model, and E2EE chats.

Sources

Read November 20, 2026.

Common questions

Does a thumbs down on ChatGPT do anything?

Yes. OpenAI uses it to improve its models, and its help page says the entire conversation tied to that rating may be used for training, even if you turned training off. On ChatGPT Business, Enterprise, and Edu, a rating doesn't permit training on the conversation.

Is AI chatbot feedback anonymous?

Partly, at some apps. Anthropic de-links feedback from your user ID before using it, Mistral separates it from your account, Google keeps reviewed feedback disconnected from your Google Account, and Proton anonymizes Lumo feedback. The conversation itself still goes with it, so a chat that names you can still point to you.

Can I delete feedback I already sent?

Not at Google. It says chats reviewed by human reviewers, and related data like your feedback, are not deleted when you delete your Gemini history, and keeps them up to 3 years. Anthropic keeps feedback 5 years, and stores it apart from your other Claude conversations.

Does reporting a response send it to a person?

Often, yes. Mistral says a report gives a limited number of its team members the right to access the reported content, and Google says Gemini feedback is reviewed by specially trained teams. At ChatGPT and Duck.ai, a report starts from the same thumbs-down button as a rating.

Does Microsoft Copilot send my chat when I click thumbs down?

Not at first. Microsoft says only your thumbs-up or thumbs-down choice is collected until you submit feedback, and it asks permission before adding your prompt, the response, or content samples such as your chat history. It says it doesn't use this feedback to train the foundation models behind Copilot.

Does Grok use my thumbs up for training if I opted out?

It may. xAI's FAQ says that even after you opt out of model training, feedback you choose to give may be used for training purposes, and its privacy policy names rating an output with the thumbs icons as one way you give it.

Which AI chatbots don't use feedback for training?

Microsoft says it doesn't use Copilot feedback to train the foundation models behind Copilot. Duck.ai doesn't train on chats, though DuckDuckGo stores and reviews a conversation you share as feedback to improve its products. Proton uses Lumo feedback only for one model, Apertus 1.5, and anonymizes it first.

Does Morse train AI on my chats?

No. Morse doesn't train AI on your chats, and its privacy policy says no AI provider trains on what you send through Morse. Your chat history is stored in your browser, encrypted, not on Morse's servers.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next