Shadow banning on GPT store

[ozgur]

Hello, we’ve been shadow banned from GPT store as Keymate AI I believe.

Although it seems to be understandable, I think a clear communication would be useful to us.

There are two problems:
1.Copycat keymate ranked higher than original keymate on keyword search “keymate”
2.We added very powerful features recently and therefore we launched it by putting “PDF” in front of the GPT name. (We are not on the list when a user searches for “PDF” on GPT store )

Happy to add not affiliated with youtube to subtitle . If “youtube” keyword is an issue. Yes keymate browses youtube as well.

[Diet]

Happy cake day!

PDF keymate is your app, right?

Your issue is that it’s not highly ranked when you just type “PDF”, correct?

I don’t know if anyone knows how that search bar works exactly, but I think we can assume that they use some sort of keyword distance.

I suspect the reason why you don’t show up on the front page for PDF is that you have too many other keywords that distract from “PDF”.

These apps are all very focused on PDF, and nothing else:

[elmstedt]

You are not shadowbanned.

There are millions of GPTs, probably half of them say something about PDFs, they can’t show them all in the top-12 results.

Apparently “KeyMate” has also become a popular term, so the same issue exists there.

[ozgur]

Thanks for celebrating my cake day .

I’ve seen this on the internet!

FYI

Any explanations to this @elmstedt

[Diet]

Can’t find much on that - but it’s from the client side analysis, right?

They have a ton of randomly named stuff in the client, I can imagine it’s something as innocuous as deranking creators you’ve just flagged, or it could be an enterprise feature.

[RonaldGRuckus]

This is interesting.
I can confirm this to be real

What the hell is a “blessed” partner? There’s apparently none. Is this the equivalent of sponsored advertising?

You can run this yourself using the following code (generated by ChatGPT)

const findByKey = (obj, substring, result = []) => {
  for (const key in obj) {
    if (key.includes(substring)) {
      result.push({ key, value: obj[key] });
    }
    if (typeof obj[key] === 'object' && obj[key] !== null && !Array.isArray(obj[key])) {
      findByKey(obj[key], substring, result);
    }
  }

  return result;
};

findByKey(JSON.parse(localStorage.getItem("STATSIG_LOCAL_STORAGE_INTERNAL_STORE_V4")), "shadow_banned");

In the OpenAI office:

What’s an effective way we can shadowban somebody??

What if we made this information publicly available??

To be fair these could just be false values. Who knows.

My biggest frustration here is that: If these values truly are used to determine the search results why the hell are we not given them to modify ourselves? You know? It’s a slap to the face. Honestly.

I love knowing that people are running diffs on this object. It has been the source of so many leaks.

[elmstedt]

@ozgur

Any explanations to this @elmstedt

Are any of those seven IDs yours? If not, then you are not shadowbanned.

[elmstedt]

I am assuming these shadowbanned accounts are the accounts which were early on scraping thousands of GPTs and republishing them.

Note, there are only seven of them, so of all of the many ChatGPT Plus accounts only the smallest fraction of them have been shadowbanned. OP is almost certainly not one of them.

[RonaldGRuckus]

Agreed. It’s gotta take being extra special to show up to the <10 club of millions. Just a cool find

[elmstedt]

Yeah… There’s lots of cool stuff you can find in the code.

Like you can search the store yourself using,

https://chat.openai.com/backend-api/gizmos/search?q=<keywords>

With whatever <keywords> you want. You can also set the limit up to 20, but unfortunately no pagination so no deep dives into the search results.

Additionally there is,

https://chat.openai.com/public-api/gizmos/discovery/<category>?limit=20&locale=en

Where <category> can be one of, featured, trending, dalle, writing, productivity, research, programming, education, lifestyle, or the hidden category other. You can also set the locale argument to en or global but I haven’t seen any difference in my admittedly limited tests.

The other category is wild and full of stuff which should be banned from the store, LOL.

[SomeUser2022]

Maybe “they deserve it” but messing with a persons perception of reality is abusive. That OpenAI even has this deception implemented, now we’ll always have a doubt, am I shadowbanned? Am I suppressed? Is my competitor blessed? Did they change the blessing weighting? Innocense lost, lol

[elmstedt]

Is your ID any of these?

user-0gH9TRvd6YiSFFP7E0gxpPmN
user-3IYvlskfX4TZgDbFqGNlEIpG
user-ByabMylJyofciUJc3JnQ0sfJ
user-Hj3eIZmzSzhvtrYszi01U9cj
user-8tKdAz5L91IcU4NBKfWDevWU
user-jZTvjdtd7ye2u2BlAnVozcvs
user-cInCKnKQfSoqQZfVcjdiKAS5

If not, you’re not shadowbanned.

The purpose of a shadowban isn’t too “mess with someone’s reality,” it’s to protect the platform from abusive actors.

The delay in their discovery of being banned postpones any attempts to subvert the ban.

It’s a very powerful tool and as you can see from the fact only seven accounts have received this treatment, that it is used only very rarely.

[SomeUser2022]

It doesn’t matter if its for a noble purpose, its still deception. We can still choose to trust OpenAI, but now we know, the store is subject to manipulation.

[elmstedt]

@SomeUser2022

now we know, the store is subject to manipulation

I’m confused. It’s their “store” for GPTs built and accessed on their own platform. The top category is,

Featured

Curated top picks from this week

They regularly remove GPTs for a wide number of reasons.

They also ban users for a wide number of reasons.

Preventing habitually bad actors from infiltrating the GPT Store and overwhelming it is somehow one manipulation too far for you?

It’s their storefront for their product on their platform using their model… of course it’s “subject to manipulation.”

Please don’t tell me you think Google or Amazon search results are a purely organic meritocracy, I don’t have it in me to deliver more bad news today.

Edit: I wanted to add something to this with respect to the act of shadow banning itself.

I acknowledge and understand it is a very controversial subject, and I have very mixed feelings about it.

For instance, I would never support the act of shadow banning on a community forum such as this one, Reddit, etc as I see that as abusive to individuals. It’s easy enough to just ban someone over and over if they are continuing the same unwelcome behaviors.

I feel a little differently about platforms where either there is little to no person-to-person communication or their may be financial incentives to circumvent s ban.

The GPT Store meets both of those criteria.

But, I fully understand there are many people who feel shadow banning is inherently unethical. I see their points and I fully agree with all of them.

I just disagree with the idea that this particular unethical behavior is never warranted or justified.

My assumption is these shadow bans are most likely targeting those accounts which early on scrapped data from tens of thousands of GPTs and republished them. The kind of bad actors who may have the resources to set up countless new accounts, shift IPs, etc, and just generally go about the business of evading bans in a way a more modest base actor either doesn’t it would simply choose not to. Whether or not employing shadow bans to quarantine these bad actors is good policy or not is something reasonable people can disagree about.

Personal, I think it is warranted and justified in limited situations and when used judiciously.

That said, the obvious problem being that the very nature of shadow banning makes it nearly impossible for outsiders to assess how often and under which conditions it is being employee.

So, @SomeUser2022 you are 100% right and justified to call into question OpenAI’s “character” and trustworthiness as a company knowing they do engage in shadow banning at some scope in some contexts under some circumstances.

I will make a point to bring it up personally with whichever OpenAI staff takes over here because—and correct me if I’m wrong here—one of the things about this which concerns you the most is the lack of transparency around the issue and some communication straight-from-the-horses-mouth might assuage some of those concerns somewhat. Regardless, I think this is a topic for which the community deserves some answers.

But, again, there are seven IDs (out of millions) listed, it’s easy enough to check if yours is amount them, though that doesn’t rule out the possibility of the existence of other such mechanisms not so open and blatantly labeled.

[SomeUser2022]

If someone does a bannable offense, then ban them, its not complicated. Don’t play shadow games where someone doesn’t know whats real, or if their product is being treated fairly`

[grandell1234]

I believe this concern is unprecedented. With over a million GPTs published and only seven people “Shadow Banned,” I don’t believe there should be any concern unless you are spamming GPTs, making hundreds of GPTs that are just copies of others, or doing some other crazy thing.

[klym]

@elmstedt Thank you for your post! It contains very interesting information. Could you advise whether everyone can access those endpoints to search the GPT Store, or if they require privileged access? When I tried to open them in my browser, I received ‘Unauthorized - Access token is missing’ for the /search endpoint, and ‘Internal server error’ for the second one. Could you advise what is required to be able to use these endpoints?

I am referring to this message:

Like you can search the store yourself using,

https://chat.openai.com/backend-api/gizmos/search?q=<keywords>

With whatever <keywords> you want. You can also set the limit up to 20, but unfortunately no pagination so no deep dives into the search results.

Additionally there is,

https://chat.openai.com/public-api/gizmos/discovery/<category>?limit=20&locale=en

Where <category> can be one of, featured, trending, dalle, writing, productivity, research, programming, education, lifestyle, or the hidden category other. You can also set the locale argument to en or global but I haven’t seen any difference in my admittedly limited tests.