[BUG] Please stop asking me to provide feedback

Same here. I found this report because I was looking for a way to disable this. This is not an enhancement. This is a bug, because it never used to ask me for feedback and now I suddenly get it multiple days in a row. The bug is the missing option: "-1 Never ask me ever again".

We adopted Claude Code specifically because of Anthropic's commitments that customer data would not be used for training or other purposes. Anthropic's tools are significantly more expensive than competitors, and we justify this premium cost primarily based on your privacy guarantees. The recent addition of mandatory feedback prompts (0-5 satisfaction rating) conflicts with these guarantees and potentially breaches our company policy.

As an administrator, I have no ability to disable these prompts.

This creates a problematic situation where:

1. Our developers are being asked to provide usage data during work
2. We cannot verify what data is collected or how it's used
3. We have no organizational controls to prevent this data collection

Request: Please either:

1. Provide an enterprise/admin setting to disable all feedback prompts globally, or
2. Make this opt-in rather than mandatory

Our organization chose Anthropic's tools over cheaper alternatives based on explicit privacy guarantees. Introducing mandatory telemetry without admin controls undermines the core value proposition we're paying for.

I want to report my issue as well:
I'm experiencing exactly the same problem.

First issue: There is no option to disable this feature.

Second issue: When I press "0" with the intention of ignoring the prompt, it displays a "Thank you for feedback" message. This is ambiguous - it's unclear whether data has been collected and sent without receiving explicit consent.

Third issue: While the console input is meant for interacting with code, if a number is entered in the input data, it uses that input for a different purpose (the feedback system) instead of passing it to Claude.

<img width="427" height="122" alt="Image" src="https://github.com/user-attachments/assets/dc5a8c7c-1585-4ee2-bdc3-6ecd2eda92b4" />

#8062

An additional problem with this feature is that it appears to store its state in my ~/.claude/settings.json in the field feedbackSurveyState which isn't documented in the schema. I've gone ahead and set lastShownTime to 2114380800000, which is in 2037, and hopefully that will disable this feature.

I don't actually want to disable this globally, just on specific (proprietary) repositories, but this is the lever I have access to.

[edit, 2 weeks later] Unfortunately, claude is still asking me for feedback, but at least it isn't changing the timestamp in my config.

Adding my voice to this request. The "How is Claude doing this session?" prompt interrupts my workflow constantly, and there's no documented way to disable it permanently.

My Experience:

• The prompt appears randomly during active coding sessions
• Pressing "0: Dismiss" doesn't prevent it from returning
• It's disruptive when you're in a flow state working on complex tasks

What I'd Like to See:

A simple configuration option in settings.json like:
{
"disableFeedbackPrompts": true
}

Or at minimum, make "Dismiss" actually mean "never ask again" rather than just postponing the prompt.

I actually asked Claude if it agreed with my assessment, and it said: "You're absolutely right!"

I agree with the concerns and suggestions previously mentioned as well.

This issue is causing significant irritation and affecting my work performance by increasing my stress levels.

When I press "0" I receive a message saying, "Thank you for feedback." However, if this is intended to collect feedback, it should include dismissed sessions as well.

I honestly think many users simply do not want to see this "feature".

I did not want to put it this way. However, this basically is a cheap way to try to collect data by waiting your users make a mistake by hitting numbers accidentally.

On the other hand I already have "Help improve Claude" disabled on web ui in the first place. Why do I have to deal with this separately in Claude Code?

I came back just to say this clearly to anyone reading this thread:

This kind of feedback mechanism should not be normalized.

When users pay for a premium product, we’re not signing up to be unpaid data providers. If the company needs structured feedback to improve the product, then they should compensate users or offer clear opt-in — not trick them into providing it by nudging or accidental keypresses.

I’ve already disabled “Help improve Claude” on the web UI, so having to deal with the same prompt inside Claude Code separately is unacceptable. This is effectively forced telemetry without transparent consent or control.

If Anthropic wants our feedback, there are fair ways to do it:
– Respect the “Never ask again” option.
– Provide a documented setting to disable prompts.
– Or offer incentives if you’re going to keep asking paying users for free data.

We need control over this feature. Period.

Same here. Using VSCode and the Claude Code CLI on an AWS instance and these nags are extremely distracting and frustrating if you are in the middle of a debug session. "Now look at line number xxx. Hey where did the 1 go? Oh the feedback nag must have consumed it"

Is there any workaround for this? does changing feedbackSurveyState work?

More over I could not get, if that survey takes into account your session as training data or not.

https://www.anthropic.com/legal/consumer-terms

Based on the Terms of Service you've shared, here's what happens with your data when you opt out of training on the Pro Plan:

Section 4 - "Our use of Materials" states:

"We may use Materials to provide, maintain, and improve the Services and to develop other products and services, including training our models, unless you opt out of training through your account settings."

However, there are two exceptions even when you opt out:

"Even if you opt out, we will use Materials for model training when:

1. you provide Feedback to us regarding any Materials, or
2. your Materials are flagged for safety review to improve our ability to detect harmful content, enforce our policies, or advance our safety research."

Summary:

If you're on the Pro Plan and choose "do not improve" (opt out of training):

• ✅ Your regular code and data will NOT be used for training
• ❌ Exception 1: If you give feedback (like thumbs up/down ratings), that conversation will be used for training
• ❌ Exception 2: If your content is flagged for safety review, it may be used for training to improve safety systems

So generally, your code and data won't be used for training if you opt out, but avoid using the feedback buttons if you want to completely minimize training usage.

This is so annoying. I'm sure they will fix it, because it makes no sense.

This is so annoying. I'm sure they will fix it, because it makes no sense.

They're trading a hit on UX for more data to guide product decisions, and (imo correctly) betting that it won't make a material impact to their user base from people churning out due to it. The data is valuable, it's high ROI.

If they dropped a feedback link at the start of a session, they'd probably get orders of magnitude less feedback than this obtrusive method.

Don't get me wrong, I find it quite frustrating and hope they provide a way for us to disable, but it makes sense to me.

> This is so annoying. I'm sure they will fix it, because it makes no sense. They're trading a hit on UX for more data to guide product decisions, and (imo correctly) betting that it won't make a material impact to their user base from people churning out due to it. The data is valuable, it's high ROI. If they dropped a feedback link at the start of a session, they'd probably get orders of magnitude less feedback than this obtrusive method. Don't get me wrong, I find it quite frustrating and hope they provide a way for us to disable, but it makes sense to me.

I wonder if it is really useful, because the first option they have listed is 1 Bad, so that is my default response

I wonder if it is really useful, because the first option they have listed is 1 Bad, so that is my default response

I imagine they can get signal from aggregating data and correlating trends with releases. Or discounting people who have answered bad all along? Statistical analysis is not my forte though.

Provide a way to disable this. I am not paying you 20$/month for you to disturb my workflow as you please.

Anthropic THIS IS AWFUL....we all should give the most negative feedback ON THIS HORRIFIC change.

This can be such a deal breaker. Sometimes you can end up submitting feedback and hence your data by accident. I'm looking for alternatives...

This feedback bug is annoying, distracting, and disruptive!

I spent more time than I should investigating this. Apparently, there is an undocumented env variable in the binary, which should disable the survey: CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY

Searching for it on google also gave this: https://www.threads.com/@boris_cherny/post/DQIGLVXkj1l/claude_code_disable_feedback_survey1-claude-you-can-also-add-that-to-the-env-pro

I've added the following to settings.json:

  "env": {
    "CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY": "1"
  },

Testing it now, but I guess it should work. Would have been nice if that had been added to https://docs.claude.com/en/docs/claude-code/settings#environment-variables

If anyone's interested, it showed up in
strings ~/.local/share/claude/versions/2.0.28 | grep Dismiss | grep Good | grep Bad

Anthropic, please don't add things like this in that fashion. Or at least add a _Never show this again_ option, especially since the functionality for that had already been implemented. It was way too intrusive, and made users unsure about what data was collected. This was particularly concerning given that you recently announced that you've started storing user data for 5 years.

Edit: Have been running with this fix for 3 days, and it seems to be working. Haven't seen the survey since I added the env variable to settings.json.

Thanks for the feedback! We just reduced frequency ~10x. Please open a new issue if it still feels too frequent -- we want to balance frequency with signal. The survey is a really helpful way for us to make sure we didn't accidentally regress something.

@bcherny

This isn't a frequency problem, it's a consent problem.

Users who want to help already have /feedback and GitHub. The ones being interrupted are the ones who don't want to participate.

If this data is critical for quality, then make it opt-in by default or offer free/discounted access in exchange for the data. What you're doing now is charging users AND extracting data from them mid-workflow.

@bcherny Can you please confirm how these reports are used, most critically the associated session data? If I answer this prompt, is my session used for training?

Data from these surveys is not used for training. See https://docs.claude.com/en/docs/claude-code/data-usage

You can opt out by setting CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1 either in your env or in your settings.json's env field.

Again, you should see these rarely from now on, and they're super helpful as a way for us to make sure people using Claude Code are having a good experience. If you really really don't want to participate, you can opt out with the 1-liner above.

Promises that data won't be used in some way are not enough, I need promises that data will never be collected. I work at a company that deals with extremely sensitive user information that has legal implications if it is leaked. Any data collection for any purpose is a deal breaker. This may stop our entire company from using this product. We were already moved over to bedrock and had guides to disable telemetry but this type of data dark pattern is unacceptable. Every ask has a probability associated with it. Even educated users may accidentally hit a feedback option accidentally. That means the probability of data leakage nears unity over time. Data harvesting should -never- allow for accidental data leakage especially when privacy policies and usage imply that using a platform like bedrock will block exactly this type of harvesting. This has stopped me cold today and I and my company are now wasting time digging into this code repo to trace all telemetry to see if we are still safe using this product. This has had a real impact on long term trust that will take a lot of work to rebuild.

Again, to make it super clear @jmward01:

1. We don't train on survey feedback data. This is documented here: https://docs.claude.com/en/docs/claude-code/data-usage
2. To permanently turn off the feedback survey for yourself, set CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1
3. To permanently turn off the feedback survey for your whole company, set CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1 in the settings.json checked into your codebase, or in your enterprise-managed settings.json (docs: https://code.claude.com/docs/en/settings)
4. You can also opt out of both telemetry + survey by setting CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1, or you can more granularly opt out with DISABLE_ERROR_REPORTING=1, DISABLE_TELEMETRY=1, etc. (also documented in the link above)

Security and privacy are extremely important, and we spend a lot of time getting these right. You have full control over data usage, telemetry, and training, and these are configurable for yourself, for your codebase, and for all your employees. We offer all of these options out of the box, so you can choose the mechanism that makes sense for you.

If there is a setting or control that is missing, please tell us. Similarly, if anything is unclear in the docs, please let us know.

@bcherny It would be amazing if the fact of this being simple product satisfaction were mentioned directly in the UI for this!

I absolutely would've responded many many times instead of dismissing had I known that no conversation data was being sent – I and others around me have definitely fallen into the expectation that this is like other LLM feedback UIs that send the chat!

All of that said, thanks very much for the super clear answer!

---

For the folks coming from HN who might not click through to the link, this is what it says (emphasis mine):

When you see the “How is Claude doing this session?” prompt in Claude Code, responding to this survey (including selecting “Dismiss”), only your numeric rating (1, 2, 3, or dismiss) is recorded. We do not collect or store any conversation transcripts, inputs, outputs, or other session data as part of this survey. Unlike thumbs up/down feedback or /bug reports, this session quality survey is a simple product satisfaction metric. Your responses to this survey do not impact your data training preferences and cannot be used to train our AI models.

It's real simple my friends. Add don't ask again, so users can express their preferences.

I am not great at digging into minified js, but poking through the code I see that the survey -looks- like it sends two messages, one immediately and one with the result. Again, I could be reading the minimized js wrong, but both have message ids attached (not messages). so it looks like you are transmitting even when the user doesn't respond and it looks like you are sending more than just a numeric rating and it looks like you are sending non-essential traffic even when on bedrock/vertex ai. This is different than what is documented in both the the 'Default behaviors by API provider' and 'Session quality sureveys' sections of https://code.claude.com/docs/en/data-usage :

"When you see the “How is Claude doing this session?” prompt in Claude Code, responding to this survey (including selecting “Dismiss”), only your numeric rating (1, 2, 3, or dismiss) is recorded." and

"by default, we disable all non-essential traffic (including error reporting, telemetry, and bug reporting functionality) when using Bedrock or Vertex."

After spending the last few hours digging through the code and dumping packets with claude sessions going it does appear that most telemetry is off when bedrock/vertex ai is set but the feedback survey, which is definitely non-essential traffic, clearly is still active. This should:

• be opt-in, not opt out: '/bug' '/feedback (doesn't exist but could fill this role)' are opt-in but should still warn what will be sent.
• your docs shouldn't claim all non-essental traffic is blocked by default on bedrock and vertex ai
• your docs shouldn't claim that only a numeric rating is sent and that it is only sent when the user responds

(again, assuming I dug through the code correctly. Please comment and clarify since I am not the expert here)

If I used the anthropic endpoints I wouldn't have the guarantees that bedrock and vertex ai make about access to messages. It is why we aren't going direct to anthropic in fact. As is I am -exceptionally- suspicious about why you are sending a message id as part of the payload if it isn't actually used/needed/accessible. Again, please clarify if that isn't being sent and I am wrong about that. This is just two hours of digging so I can easily be wrong about the 'appeared' and 'responded' event_type packets.

In general Claude Code appears to be more open about its collection and use as well as more able to shut down telemetry but the discrepancies here give me pause.

Hmm @jmward01 we don't log if you're on 3p or have opted out. Any of these env vars will opt you out:

• CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY
• CLAUDE_CODE_USE_BEDROCK
• CLAUDE_CODE_USE_VERTEX
• DISABLE_TELEMETRY
• CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC

@bcherny

This shouldn’t be opt-out. It should be off by default. Asking the user to manually disable feedback prompts is backwards.

CLAUDE_CODE_USE_BEDROCK
CLAUDE_CODE_USE_VERTEX
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC
seem to still get the survey. Does it not log with those set but still ask you?

Yeah, it asks but doesn't log. CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY doesn't ask or log. Fix incoming.

I appreciate the incoming fix for the feedback ask. Can you explain what the message_id is used for considering the docs say only a numeric rating is sent back? Is that also being removed or are the docs being updated?

Data from these surveys is not used for training. See https://docs.claude.com/en/docs/claude-code/data-usage You can opt out by setting CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1 either in your env or in your settings.json's env field. Again, you should see these rarely from now on, and they're super helpful as a way for us to make sure people using Claude Code are having a good experience. If you really really don't want to participate, you can opt out with the 1-liner above.

Thanks for finally telling everyone how to turn this garbage off, but your response is completely tone-deaf and it would do you well to go actually read what people are saying instead of vomiting your corporate justifications at us.

No one that found their way into this thread is going to buy your BS excuses and nauseating corporate justifications for violating your own promises.

I have to admit, the responses here appear to be a bit lacking. I'm glad corporate users will be better protected if the survey bug is fixed, but I believe all users are owed an explanation for the message_id being passed automatically and on response. If I am wrong about what the code is doing please point it out but it appears that the code is at odds with the documentation language as I pointed out several times now. If anyone that knows is still monitoring this bug thread please actually consider removing the message_id or updating the docs to explain that it is being sent and what it is used for. This is definitely more than 'just a simple numeric value' being sent back.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.