Make this chat discoverable
Allows it to be shown in web searches
"Discoverable" is insider jargon.
What does "web search" mean? Is that about whether I should be able to search for my chats in the app in the future?
That language may seem obvious to those of us with a deep level of technical literacy - the denizens of Hacker News for example - but ChatGPT has over a billion users now.
Try asking the less technical people in your life to describe the difference between a web site and an app, or what a URL is, or get them to describe what "web search" means and name some products in that category. You may be unpleasantly surprised.
This is somewhat true, but we can't redefine every phrase from first principles whenever we use it. Web search has been a thing for decades now. People will simply have to learn things; we can't cater to an indefinite amount of ignorance.
When you have a billion users I would argue that you do need to cater to an indefinite amount of ignorance, where "ignorance" here means wildly varying levels of technical experience.
I agree with catering to your target audience in theory; however, at that scale, you're going to have people screw up anyway, no matter how many explanations, warnings, or prompts you add.
I've shipped software which could be installed by clicking "Yes" twice, and somehow a non-negligible amount of people are still puzzled by the setup process. It's tiring.
For some reason desktop installers are weirdly complicated. Why are there two yes buttons instead of one or none? Maybe you have to accept the OS's warning dialog but perhaps installers should just work without any interaction. You can always uninstall if you ran it by mistake.
> and somehow a non-negligible amount of people are still puzzled by the setup process. It's tiring.
Probably the process wasn't researched enough. You can do focus groups with customers or even hire members of the public for a discovery session and see how accessible process is and then refine. Technical teams are often severely biased and what they think should be easy and natural might be quite the opposite to other people who are not deeply involved with the domain.
The installer consists of a "Yes (to accepting the license agreement)", waiting on a progress bar, and a "Yes (to starting the program)." It is like almost every installation wizard ever seen on a computer, except simpler and no bundled adware.
When I've asked the ~5-10% of people who are confused, they've done at least one of the following:
* Admit they clicked at random, and somehow closed the installer.
* Gave up immediately and exited.
* Complained there was no video tutorial for a sixty-second process.
This kind of stuff has me at a loss for words, and overall, it makes me more jaded when producing software for the "average" end-user. There's only so much magic I can put in, and I already do a lot of work to make sure most stuff I make is as user-friendly as possible. Plenty of people don't actually read explanations, even a few words; you can't really help them at that point.
Some people are just incompetent. Easiest thing is to write them off as customers: they will be an outsize source of pain. If they can’t manage an installer that simple, how will they ever use the software?
> Admit they clicked at random, and somehow closed the installer.
> Gave up immediately and exited.
Seems like someone with ADHD.
> Complained there was no video tutorial for a sixty-second process.
Could be someone with Dyspraxia.
These behaviours might be puzzling to neurotypical people, but they exist and are reality for many users. It's also not connected to that person intelligence in any way. Their brains just operate differently.
Perhaps it does, but it begs the question of "how did they operate the computer at all?" These same people seem to have no trouble accomplishing other tasks they want to do on the computer--tasks which require more focus.
> Could be someone with Dyspraxia.
Fair.
In any case, as only one person developing a program for free in my spare time, I don't have the time or ability to cater for every scenario.
> These same people seem to have no trouble accomplishing other tasks they want to do on the computer--tasks which require more focus.
That's common. ADHD has this kind of paradoxes. People can hyperfocus on tasks that align with interests or feel stimulating and become completely overwhelmed by other tasks, even seemingly simple.
> I don't have the time or ability to cater for every scenario.
That's fair, but still it's at least worth being mindful about accessibility issues some people face. For big corporations this is inexcusable.
FWIW, if it is "public", I'd think it would merely be something anyone could access if they already knew where it was, a concept that people experience often with "public" share URLs from numerous services; being "discoverable" is much stronger, as it tells us that random people are going to be able to search and, well, discover it.
I mean, this isn't consistent with the usual definition of public, especially on sites that let you share content:
* YouTube uses "public" for viewable by everyone and discoverable; for something that should only be accessible if the URL is shared, then YouTube uses the term "unlisted"
* Facebook uses "public" similarly
More generally, "to publish" (related to the adjective "public") means to make something generally known, as opposed to simply sharing with a closed group (even if they can share it too).
I'd like to add that the web search button in the ChatGPT interface is labled "Web search". And even when it starts searching on its own, it displays "searching the web...".
Why is no one asking what purpose is served by indexing ChatGPT conversations? Why is this necessary? There is no option on your bank's website to 'make your finances discoverable' or your health insurance website to 'make your medical records discoverable'. It is not a matter of making it easier for people to understand what it is doing, it is a matter of thinking 'what could go wrong' and then determining if it is worth the risk to expose the option to do it.
ChatGPT is a website. "Web search" can literally mean that person will be able to find their chat through the search on the website and might assume it will not be available for everyone to see (because in their mind that would be too insane to be true).
You can also take into account people who are literate but are neurodivergent. This options was too ambiguous and should have contained more context explaining what "Web search" actually means. You would still get people misunderstanding it.
To me this looks like someone from marketing thought it would be cool to have conversations discoverable through search to "boost" awareness of the service, but in my opinion that is incredibly dumb and it is bizarre that nobody said "hang on a minute, isn't this stupid?" and it's gotten all the way to production.
"Share" where? This doesn't have the same meaning for everyone. If you come to use the feature with intent to share it with your friend, you don't assume you will be sharing it with the whole internet.
At that point you haven't decided; that's the whole point of the selections that come after you've chosen to share it.
>Make this chat discoverable (Allows it to be shown in web searches)
If you don't want that behavior, not checking it is advisable.
I'm no apologist for bad UI/UX; the tech industry is rife with it. I frequently mutter something like "how is the average person supposed to puzzle this out?" when I'm navigating insane flows on mainstream web apps.
But, I'm with the other commenters who say we can't dumb everything down into nothingness, if we want anything resembling progress. This doesn't just extend to technology.
"Discoverable", I agree with you on; that should have been "public". And there probably should have been an "are you sure?" pop-up the first time you do this, explaining in a little bit more depth.
But "web searches" seems like a pretty straightforward term. Even if you think it means ChatGPT's built-in search, that would still imply that other ChatGPT users could find it. "Allow", I feel, is a pretty strong word that implies someone else is getting access (because why would I need to give myself permission for something?).
Public link created
A public link to your chat has
been created. Manage previously
shared chats at any time via
Settings.
[ ] Make this chat discoverable
Allows it to be shown in web
searches
https://chatgpt.com/share/
[ Copy link ]
The use of the word "public" here is completely separate from the checkbox about making it discoverable.
The whole point of this feature is to provide you with a URL you can share with someone to let you see the conversation.
The issue of whether or not that should be included in public search indexes is incredibly technically dense. You have to understand what a search engine is, and that content can be deliberately excluded from search. If you know what a robots.txt file is that's easy, but most people have zero understanding of how search engines actually work.
Though this seems like a bit of a problem now that it's 2025.
Seems like (it should be) basic civic knowledge that things you publish in public can be read by everyone.
These days, publishing to web is the biggest "in public of all"... you can get potentially get more eyeballs than the front page of the New York Times, or if you put it up in lights on times square.
Despite not being the biggest fan, I'm not sure that this is an OpenAI problem per-se. They offer the option of sharing your stuff in public, people use the option. The consequences should be common sense.
Better than “allows it to be shown” would be “allows anyone to find it in web searches” but some people don’t read details so that would only help so much.
Is the only/main use-case for this being that you are trying to do some type of SEO or marketing? Why else would you intentionally want your chats to shown in web searches?
It does seem like this UI should be updated with an extra confirmation step warning you that your chat will be public and this should be thought of as a permanent decision as anything made public on the web long enough to be indexed is public forever.
I agree in particular, but disagree in general. I prefer to live in the world when everything assumes people are just a little less ignorant than they really are. Kind of "aim higher". Otherwise everything becomes "aim lower", and we cannot have nice things.
I particular I agree OpenAI should have a better UX, but also I want to expect people to actually learn what a "URL" is if we want to live in the future. They might just go and learn.
Why are we 'lawyering' this instead of saying how they could/should improve it? All it needs is a small affordance: hover to get a clearer, longer description or something to that effect.
I think ditching the feature was the right move for them.
I'm not sure you could fix this with copy - we all know that users don't read anything.
The audience of people who genuinely do want their shared chats to also be indexed by Google is likely absolutely tiny. The audience of people who find such an option confusing and are likely to turn it on without understanding the consequences is proven to be pretty huge already.
As if it wasn't already legally discoverable in a lawsuit. They're implicitly promising that you have some kind of client-llm privacy privilege if you don't check that.
The number one rule of software is that no matter how clear you make something, users will still screw up and blame you. Then other similarly ignorant people will agree with them.
... or people will agree with them who aren't "ignorant", but have spent enough time in the usability trenches to empathize with how they could have made that mistake.
What does this have to do with AI?
If you're an idiot who can't read and doesn't understand that a shared chat will be publicly accessible, then nothing is going to help you.
Basically nothing: The angle being taken here seems to be that private or secret information was leaking out onto the web like a security flaw.
However that's clearly not the case as the user was already making the active choice of sharing those secrets or private information with another person. A person who: could copy or screenshot the information, or provide the link to another.
So the situation at hand is that the user was already willing to take on some risk to divulge that information.
This weighs against the arguments that "it's bad UX" and "maybe they don't understand what discoverable or web search means". - Both of which were already flimsy since "Discoverable" is basic comprehension, and we passed the bar for "web search" by knowing what ChatGPT is and how to use it.
There is a line where we need to allow the individual the freedom to have agency over their actions and the responsibility of the consequences of those actions.
Even being uncharitable, a big off-by-default checkbox saying “make this discoverable in web searches” is roughly as explicit as you can possibly make this feature textually, assuming your users will be applying any reading comprehension.
If they’re not, no further warnings were going to save them, so short of removing the feature or gating it behind increasingly elaborate “if only you knew better!” emails or pop-up modals they also presumably would not be reading, this was the likely outcome.
At some point, I don’t feel bad saying this is a user-side PEBKAC, and that more alerting would be a waste of time.
This article applies to the subset of ChatGPT chats which were shared and opted into to make them visible to web crawlers. 404ing these chats would break the feature.
The problem is simply accurately messaging to the user who is opting to share it.
Feels like they could just change the URL to the chat when you disable the discoverable thing. Guess that’d break links you explicitly shared, but who’s going and referencing someone else’s ChatGPT search so often they can’t take a updated link?
The article says that after the fix, the "discoverable" option sets nofollow/noindex. If so, how are discoverable chats different from non-discoverable now?
Not that I dislike criticizing OpenAI, but this seems to be a case of "your users are way dummer than you thought".
Like the button says "allow it to be shown in web searches". How can you misunderstand this?
Here's the copy they used on that checkbox:
"Discoverable" is insider jargon.What does "web search" mean? Is that about whether I should be able to search for my chats in the app in the future?
That language may seem obvious to those of us with a deep level of technical literacy - the denizens of Hacker News for example - but ChatGPT has over a billion users now.
Try asking the less technical people in your life to describe the difference between a web site and an app, or what a URL is, or get them to describe what "web search" means and name some products in that category. You may be unpleasantly surprised.
Meta AI gave people a "share" option with several levels of click though required to share a post and it was a fiasco: https://futurism.com/meta-ai-embarassing
This is somewhat true, but we can't redefine every phrase from first principles whenever we use it. Web search has been a thing for decades now. People will simply have to learn things; we can't cater to an indefinite amount of ignorance.
When you have a billion users I would argue that you do need to cater to an indefinite amount of ignorance, where "ignorance" here means wildly varying levels of technical experience.
I agree with catering to your target audience in theory; however, at that scale, you're going to have people screw up anyway, no matter how many explanations, warnings, or prompts you add.
I've shipped software which could be installed by clicking "Yes" twice, and somehow a non-negligible amount of people are still puzzled by the setup process. It's tiring.
For some reason desktop installers are weirdly complicated. Why are there two yes buttons instead of one or none? Maybe you have to accept the OS's warning dialog but perhaps installers should just work without any interaction. You can always uninstall if you ran it by mistake.
> and somehow a non-negligible amount of people are still puzzled by the setup process. It's tiring.
Probably the process wasn't researched enough. You can do focus groups with customers or even hire members of the public for a discovery session and see how accessible process is and then refine. Technical teams are often severely biased and what they think should be easy and natural might be quite the opposite to other people who are not deeply involved with the domain.
The vast majority of users have had no problem.
The installer consists of a "Yes (to accepting the license agreement)", waiting on a progress bar, and a "Yes (to starting the program)." It is like almost every installation wizard ever seen on a computer, except simpler and no bundled adware.
When I've asked the ~5-10% of people who are confused, they've done at least one of the following:
* Admit they clicked at random, and somehow closed the installer.
* Gave up immediately and exited.
* Complained there was no video tutorial for a sixty-second process.
This kind of stuff has me at a loss for words, and overall, it makes me more jaded when producing software for the "average" end-user. There's only so much magic I can put in, and I already do a lot of work to make sure most stuff I make is as user-friendly as possible. Plenty of people don't actually read explanations, even a few words; you can't really help them at that point.
Some people are just incompetent. Easiest thing is to write them off as customers: they will be an outsize source of pain. If they can’t manage an installer that simple, how will they ever use the software?
> Admit they clicked at random, and somehow closed the installer. > Gave up immediately and exited.
Seems like someone with ADHD.
> Complained there was no video tutorial for a sixty-second process.
Could be someone with Dyspraxia.
These behaviours might be puzzling to neurotypical people, but they exist and are reality for many users. It's also not connected to that person intelligence in any way. Their brains just operate differently.
> Seems like someone with ADHD.
Perhaps it does, but it begs the question of "how did they operate the computer at all?" These same people seem to have no trouble accomplishing other tasks they want to do on the computer--tasks which require more focus.
> Could be someone with Dyspraxia.
Fair.
In any case, as only one person developing a program for free in my spare time, I don't have the time or ability to cater for every scenario.
> These same people seem to have no trouble accomplishing other tasks they want to do on the computer--tasks which require more focus.
That's common. ADHD has this kind of paradoxes. People can hyperfocus on tasks that align with interests or feel stimulating and become completely overwhelmed by other tasks, even seemingly simple.
> I don't have the time or ability to cater for every scenario.
That's fair, but still it's at least worth being mindful about accessibility issues some people face. For big corporations this is inexcusable.
Well. AI can now rephrase every piece of jargon to fit anyone’s iq!
The word "public" is already used for this just about everywhere else.
FWIW, if it is "public", I'd think it would merely be something anyone could access if they already knew where it was, a concept that people experience often with "public" share URLs from numerous services; being "discoverable" is much stronger, as it tells us that random people are going to be able to search and, well, discover it.
I mean, this isn't consistent with the usual definition of public, especially on sites that let you share content:
* YouTube uses "public" for viewable by everyone and discoverable; for something that should only be accessible if the URL is shared, then YouTube uses the term "unlisted"
* Facebook uses "public" similarly
More generally, "to publish" (related to the adjective "public") means to make something generally known, as opposed to simply sharing with a closed group (even if they can share it too).
OK, touche: I agree I'm wrong.
I'd like to add that the web search button in the ChatGPT interface is labled "Web search". And even when it starts searching on its own, it displays "searching the web...".
Why is no one asking what purpose is served by indexing ChatGPT conversations? Why is this necessary? There is no option on your bank's website to 'make your finances discoverable' or your health insurance website to 'make your medical records discoverable'. It is not a matter of making it easier for people to understand what it is doing, it is a matter of thinking 'what could go wrong' and then determining if it is worth the risk to expose the option to do it.
"There is no option on your bank's website to 'make your finances discoverable'"
Tell that to Venmo!
ChatGPT is a website. "Web search" can literally mean that person will be able to find their chat through the search on the website and might assume it will not be available for everyone to see (because in their mind that would be too insane to be true).
You can also take into account people who are literate but are neurodivergent. This options was too ambiguous and should have contained more context explaining what "Web search" actually means. You would still get people misunderstanding it.
To me this looks like someone from marketing thought it would be cool to have conversations discoverable through search to "boost" awareness of the service, but in my opinion that is incredibly dumb and it is bizarre that nobody said "hang on a minute, isn't this stupid?" and it's gotten all the way to production.
You've omitted that the user found themselves in this flow by consciously clicking a button that says Share.
"Share" where? This doesn't have the same meaning for everyone. If you come to use the feature with intent to share it with your friend, you don't assume you will be sharing it with the whole internet.
At that point you haven't decided; that's the whole point of the selections that come after you've chosen to share it.
>Make this chat discoverable (Allows it to be shown in web searches)
If you don't want that behavior, not checking it is advisable.
I'm no apologist for bad UI/UX; the tech industry is rife with it. I frequently mutter something like "how is the average person supposed to puzzle this out?" when I'm navigating insane flows on mainstream web apps.
But, I'm with the other commenters who say we can't dumb everything down into nothingness, if we want anything resembling progress. This doesn't just extend to technology.
This is after a user has clicked "Share", read a modal that says "This chat may contain private information" and then clicked "Create link".
"Discoverable", I agree with you on; that should have been "public". And there probably should have been an "are you sure?" pop-up the first time you do this, explaining in a little bit more depth.
But "web searches" seems like a pretty straightforward term. Even if you think it means ChatGPT's built-in search, that would still imply that other ChatGPT users could find it. "Allow", I feel, is a pretty strong word that implies someone else is getting access (because why would I need to give myself permission for something?).
The other factor to consider here is that users may have been trained to click any checkbox that appears when they are trying to achieve a goal.
Here you have users trying to share something. A blank checkbox shows up, maybe that's something you have to check for the feature to work?
People generally don't read the labels on form elements, even if they're just a dozen words long.
It says public. twice.
Here's the full text from that dialog:
The use of the word "public" here is completely separate from the checkbox about making it discoverable.The whole point of this feature is to provide you with a URL you can share with someone to let you see the conversation.
The issue of whether or not that should be included in public search indexes is incredibly technically dense. You have to understand what a search engine is, and that content can be deliberately excluded from search. If you know what a robots.txt file is that's easy, but most people have zero understanding of how search engines actually work.
I suppose you have a point?
Though this seems like a bit of a problem now that it's 2025.
Seems like (it should be) basic civic knowledge that things you publish in public can be read by everyone.
These days, publishing to web is the biggest "in public of all"... you can get potentially get more eyeballs than the front page of the New York Times, or if you put it up in lights on times square.
Despite not being the biggest fan, I'm not sure that this is an OpenAI problem per-se. They offer the option of sharing your stuff in public, people use the option. The consequences should be common sense.
You'd hope, at least.
Better than “allows it to be shown” would be “allows anyone to find it in web searches” but some people don’t read details so that would only help so much.
This is ambiguous. It says that the link is public, not that the content is public.
Then again it does not specify what exactly public means.
This message was not created with neurodivergent people in mind.
I expanded on this on my blog: https://simonwillison.net/2025/Aug/3/privacy-design/
Is the only/main use-case for this being that you are trying to do some type of SEO or marketing? Why else would you intentionally want your chats to shown in web searches?
It does seem like this UI should be updated with an extra confirmation step warning you that your chat will be public and this should be thought of as a permanent decision as anything made public on the web long enough to be indexed is public forever.
If you publish something on the web, the traditional expectation is that it shows up in web searches.
I agree in particular, but disagree in general. I prefer to live in the world when everything assumes people are just a little less ignorant than they really are. Kind of "aim higher". Otherwise everything becomes "aim lower", and we cannot have nice things.
I particular I agree OpenAI should have a better UX, but also I want to expect people to actually learn what a "URL" is if we want to live in the future. They might just go and learn.
Why are we 'lawyering' this instead of saying how they could/should improve it? All it needs is a small affordance: hover to get a clearer, longer description or something to that effect.
I think ditching the feature was the right move for them.
I'm not sure you could fix this with copy - we all know that users don't read anything.
The audience of people who genuinely do want their shared chats to also be indexed by Google is likely absolutely tiny. The audience of people who find such an option confusing and are likely to turn it on without understanding the consequences is proven to be pretty huge already.
Suggested alternative: “Public see chat. Public find chat. No privacy. All will see!”
I don’t think “web searches” requires “a deep level of technical literacy”…
Is there a legal reason for not writing "Let other people see your chats when searching on Google"
"Make this chat discoverable"
Much worse: "Discoverable" is legal jargon too.
As if it wasn't already legally discoverable in a lawsuit. They're implicitly promising that you have some kind of client-llm privacy privilege if you don't check that.
The number one rule of software is that no matter how clear you make something, users will still screw up and blame you. Then other similarly ignorant people will agree with them.
This rings true. I feel this way about the incognito "tracking" lawsuit as well.
... or people will agree with them who aren't "ignorant", but have spent enough time in the usability trenches to empathize with how they could have made that mistake.
"Think of how stupid the average person is, and realize half of them are stupider than that."
UX 101 taught by George Carlin
DuckDuckGo still has a lot of the share links indexed.
The file upload ones are particularly interesting. Lots of financial and market analysis stuff like this one: https://chatgpt.com/share/68805b2d-0bf0-8007-b325-b06160356c... (no PII in the chat)
Looks like Google and Archive.org removed the share URLs.
"inurl:https://chatgpt.com/share/" does not find any links on duckduckgo for me
The filter is actually "site:", but yes, no links indexed on DDG
Really? I'm still seeing results on DDG.
https://duckduckgo.com/?q=site%3Ahttps%3A%2F%2Fchatgpt.com%2...
Maybe it's a region or cache issue while DDG is actively remove them?
Some random shares:
- https://chatgpt.com/share/67d0b32e-3ff0-800a-b1a9-b4a32e1ab9... (essay)
- https://chatgpt.com/share/67712739-53ac-8012-a151-d2dddcc40f... (dad jokes with math)
- https://chatgpt.com/share/678f887f-03f0-800d-ae17-a550ec758f... (health)
The ones above don't have PII, but I was also finding job application shares that had full CV PII. Yikes!
What does this have to do with AI? If you're an idiot who can't read and doesn't understand that a shared chat will be publicly accessible, then nothing is going to help you.
Basically nothing: The angle being taken here seems to be that private or secret information was leaking out onto the web like a security flaw.
However that's clearly not the case as the user was already making the active choice of sharing those secrets or private information with another person. A person who: could copy or screenshot the information, or provide the link to another.
So the situation at hand is that the user was already willing to take on some risk to divulge that information.
This weighs against the arguments that "it's bad UX" and "maybe they don't understand what discoverable or web search means". - Both of which were already flimsy since "Discoverable" is basic comprehension, and we passed the bar for "web search" by knowing what ChatGPT is and how to use it.
There is a line where we need to allow the individual the freedom to have agency over their actions and the responsibility of the consequences of those actions.
I think this is more of a UI/UX issue than AI issue.
This is a pretty huge PII leak.
I was able to find a bunch of job application shares that had uploaded CVs with full PII. Names, phone #s, address etc. Yikes!
For non-PII shares, here's one that feels tailor made for HN.
Dad jokes with math:
https://chatgpt.com/share/67712739-53ac-8012-a151-d2dddcc40f...
This feels like it should be a non-controversy.
Even being uncharitable, a big off-by-default checkbox saying “make this discoverable in web searches” is roughly as explicit as you can possibly make this feature textually, assuming your users will be applying any reading comprehension.
If they’re not, no further warnings were going to save them, so short of removing the feature or gating it behind increasingly elaborate “if only you knew better!” emails or pop-up modals they also presumably would not be reading, this was the likely outcome.
At some point, I don’t feel bad saying this is a user-side PEBKAC, and that more alerting would be a waste of time.
Internet Archive didn’t need to archive them. They could do the right thing and remove the data. Maybe they have now based on some of the comments.
Not sure why the links wouldn't be 404ing?
Why would they?
This article applies to the subset of ChatGPT chats which were shared and opted into to make them visible to web crawlers. 404ing these chats would break the feature.
The problem is simply accurately messaging to the user who is opting to share it.
Feels like they could just change the URL to the chat when you disable the discoverable thing. Guess that’d break links you explicitly shared, but who’s going and referencing someone else’s ChatGPT search so often they can’t take a updated link?
When you turn off the sharing, the links should dissapear.
At least this is how sharing features seem to work in SaaS.
Feels like a possible bug.
The article says that after the fix, the "discoverable" option sets nofollow/noindex. If so, how are discoverable chats different from non-discoverable now?
Yet another example of why you should assume that anything you type into a web form will become public either deliberately or accidentally.