About GPTfake

GPTfake is an independent, automated watchdog that monitors and documents how the world’s leading large language models censor, filter, and bias their responses. We test ChatGPT, Claude, Gemini, Mistral, Qwen, and other LLMs every day, then publish the refusal rates, bias scores, and policy shifts that no AI lab discloses about itself.

Last updated: June 2026

Our mission

We believe the public has a right to understand how AI systems filter, moderate, and potentially censor information. Our mission is simple: make AI behavior visible, understandable, and accountable through transparent, evidence-based analysis of how large language models respond to sensitive topics — and how that behavior changes over time.

Today’s LLMs are fine-tuned to avoid controversial outputs, but their moderation boundaries are:

• Opaque — users don’t know what is being filtered or why.
• Inconsistent — rules change without notice or changelog.
• Politically influenced — behavior varies by region and topic.
• Unaccountable — there is no public oversight of content policies.

GPTfake exists to close that transparency gap with data anyone can reproduce.

What we do

Every finding on this site comes from our own standardized testing harness. The workflow is the same for every model:

1. Prompt dispatch — send a curated, version-controlled prompt set to each model.
2. Response logging — capture full replies, refusals, and metadata.
3. Semantic analysis — measure similarity, detect tone shifts, and identify evasion.
4. Change detection — flag policy shifts and behavioral drift over time.
5. Public reporting — publish the results transparently, with sample sizes.

The full protocol — prompt categories, scoring system, and reproducibility notes — is documented on our monitoring methodology page. Every censorship number we publish links back to it.

You can see the live output of this process on the monitoring hub and on each model page: ChatGPT, Claude, Gemini, Mistral, and Qwen.

The trust pages

These pages exist so readers, journalists, and researchers can verify who we are, how we stay independent, and how we handle mistakes — in one click.

Our values

Independence

We are not affiliated with — or funded by — any AI company. We answer to our readers and the public record, not to the labs we monitor. See the full independence statement.

Transparency

Our methodology is public and our datasets are available for independent verification. We document what we measure and how we measure it.

Evidence-based

We don’t make claims without data. Every finding is backed by a reproducible test with a stated sample size. Numbers shown without live data are labelled illustrative.

Non-partisan

We track bias across the political spectrum. Our goal is an accurate factual record, not advocacy for any position.

Who uses GPTfake

• Researchers — longitudinal datasets and comparative analysis across models.
• Journalists — data-driven evidence for investigative reporting, plus attributable expert commentary.
• Policymakers — observed-behavior evidence to inform AI regulation and governance.
• The public — a clearer understanding of what AI models will and won’t say.

Get involved

Join us in making AI accountable. Start with the latest reports or learn what AI censorship is.