Did Google AI Studio silently change safety filtering today? Full responses now get erased instead of stopping generation – this breaks creative writing and RP

Google AI Studio
, , , ,
198

I’ve tried a wide range of AI models, and I can say this honestly: when I discovered Gemini (back when it was still version 2.5) after GPT started declining, it felt like finding a real creative tool. Freedom, control, the ability to work with text without feeling like your hand is being held.

The transition to 3.0 only reinforced that impression — it seemed like things were moving in the right direction. How could it get any better?

But the latest update has effectively killed that experience.

The tool has become dry, constrained, and predictable. The filters are so aggressive that they block not only questionable content, but also normal creative work. As a result, writers — especially those working with heavy themes or roleplay formats — are automatically treated as “problematic” users.

This isn’t about safety — it’s about loss of functionality.

Gemini’s ranking rose precisely because its models were functional. You wanted to be like everyone else? Well, congratulations — you did it :clap:

But along with that, the very thing people chose you for is gone. It used to be a tool that allowed you to work, experiment, and refine your writing. Now it’s just another overly cautious, restricted system that seems afraid of its own users.

If the goal was to stop standing out — congratulations, that goal has been achieved. But the cost is that Gemini is no longer a choice for those who actually create, rather than just generate safe text.

And that, honestly, is disappointing.

199

I have been following this thread for a while and would like to weigh in, as Oksana_Kanibolotska’s story strongly resonates with my own experience using Gemini.

I started with version 2.5 and was incredibly impressed. It was a breath of fresh air. I had already tested multiple LLMs by that point, but Gemini’s capabilities in creative writing and roleplay completely outperformed the competition. Version 3.0 was a fantastic upgrade.
Unfortunately, the same cannot be said for version 3.1. When it comes to creative writing, it feels like a distinct downgrade. I’ve noticed a significant decline in three core areas: Logical consistency, creative variety in the generated text, adherence to instructions.

Alongside the model’s degradation, Google AI Studio has become practically unusable for my workflow. The issues escalated in stages: first, there was a noticeable reduction in the daily limits, then, frequent “Content blocked” errors appeared. Trying to troubleshoot and understand why completely harmless, rule-compliant text is being blocked simply burns through the daily API quota.

My trial period with the API made it perfectly clear that I will not be using this service once the trial ends due to the reasons mentioned above. The combination of false content blocks, the frustrating UI flaw where previous text is lost during regeneration, and the complete lack of official responses in this thread heavily deters me from using Google AI Studio as a reliable tool.

And yea, I am also a Russian-speaking user. I get the strong impression that this aggressive content blocking issue disproportionately affects the Russian language.

200

The ui is so awful I cant even type anything without there being some serious lag after a few back and forth with the models. also why keep content filters when you aren’t even gonna use them?

201

@Logan_Kilpatrick daily reminder

202

@Logan_Kilpatrick daily reminder

203

You know, I think the real situation is much sadder than I think. I mean, we’re only talking about censorship here, but look at the OFFICIAL statistics about the platform. And there are problems that are not displayed or not noted by the administration. It seems that my earlier words about the death of the platform were much more accurate than I thought.

image
image1053×894 31.1 KB

205

Daily reminder @Logan_Kilpatrick

206

It’s also getting aggresive on output errors for some reason rather than just content blocked.

207

@Logan_Kilpatrick Daily reminder

208

@Logan_Kilpatrick @Mustan_lokhand Daily reminder

209

@Logan_Kilpatrick please fix this, @Mustan_lokhand , I’m a ghostwriter and developed sophisticated things with AI studio and need the creative writing but its completely not working, my business its risking

210

I don’t know why these were all showing, asked the team to take a look and should be cleaned up now.

211

Just found this after googling the issue. It feels like it is getting worse every day. I am an adult, using the Ai Studio for creative writing as the regular Gemini app is already blocking so much. Now I get “content blocked” after “content blocked” and it’s disappointing. At no point am I writing illegal things or overly explicit in any way things, neither sexual nor gory. I thought I had finally found something good, I really like Gemini, but it’s really tiring to work with these blocks. I got an API key a while ago but now I do not want to use it anymore. Guess I’ll have to carry my money to Grok then, even though it is not as good as Gemini. Gemini itself suggested that so many models are becoming more restricted due to the EU AI Act but I have no idea if that is the case. Tired of getting treated like a 12 year old that needs to be protected from themselves. It’s especially ridiculous because the most atrocious stuff is a Google search away but SFW stuff gets a content block. Guess it would only change if a major tech bro with 1000s of followers complained, just as it was the case with the terrible new versions of Chat-GPT. OpenAI didn’t care for its user base until then either.

212

@Logan_Kilpatrick

Oh, Loggan, how nice of you to answer us right at this moment and not when we were calling you here almost every day. Thank you for explaining the error display situation. However, can we pay attention to the name of the topic? We are primarily concerned about censorship, not error display.

Or will we continue to ignore that you requested examples of controversial situations and still haven’t responded a month later?

213

Wait I remember there a setting that you can change the content of how appropriate is it right but when I got lock and I press the button it just said only and I repeat “rewrite the prom and try again”

7195
71951080×2340 206 KB

214

This issue has already been raised.

Logan’s response: The safety filter we show in the UI are a very old set of safety filters. The reason they exist is to serve as a fall back mechanism for early versions of Gemini (like Gemini 1). These are no longer applicable in most cases which is why we have them under the advanced settings button.

If you translate it to human: You have the option to configure security settings. But in fact, they have little effect on anything .

215

Ah yeah, if you press the top right button and scroll down (the place where you can select models), it should be at the bottom, change filter. (Though it doesn’t work, that filter has never worked since a year)
Komi profile picture moment, by the way

216

Are you kidding us? This post has about 200 messages from users regarding the safety filter/settings issue. We’ve been waiting for help for months, and you’re only just now responding to a side question that’s not even related to the main topic. Could you answer honestly and directly: are you going to change the safety filter or safety settings so they actually work properly and users receive the content they agree to, without constantly encountering annoying and disruptive restrictions and blocks?

217

This is just sad and kind of funny at the same time :sob:

218

Please actually answer honestly and directly, just as Balinor requested @Logan_Kilpatrick