Research findings

A year of field research on how trust forms under censorship, and what it means for anyone publishing on YouTube. Here's the full report. 

We also published an operational YouTube playbook that turns the findings into a testable workflow for newsrooms working in high-skepticism contexts: relational trust, retention, information seeking in crisis, and fact-checking habits.

Gathering reliable data from within distorted information environments during crises is a challenge. Using secure channels into Iran through Telegram and VPN-based ads, we asked Iranians in January 2026, "What's on your mind?" When internet connectivity returned on January 22, we received over 220 responses in four days.

Traditional survey methodologies assume everyone has an equal chance of responding. But in restrictive environments (where apps are blocked, devices are older, or participation carries risk) standard sampling renders the most vulnerable populations invisible.

This article proposes a shift from estimating population size to mapping participation barriers. By sizing the groups affected by specific technological or safety constraints, we create a more accurate picture of the information landscape than "representative" data can provide.

In a joint study with Factnameh, Gazzetta audited seven major AI models (ChatGPT, Claude, Gemini, Grok, Kimi, DeepSeek, and Qwen) to test how they source and frame Persian-language answers about Iran. The study, rebuilt and automated in August 2026, ran a fixed set of prompts through two setups: each model's own web search, and a single fixed search index shared by all models, so the model's contribution and the search engine's could be measured separately.

The research shows that much of what looks like model personality is the search engine behind it: on identical retrieval, the models nearly converge. Prompt vocabulary remains a strong lever, with state-language questions pulling every model toward state narratives. And roughly a third of all cited sources are government-funded outlets, mostly Western and Gulf broadcasters rather than Iranian state media.

We tested international versus Chinese AI models on labor rights questions and found a paradox:

DeepSeek, a heavily moderated Chinese model, possesses detailed, tactical knowledge on organizing strikes and navigating management pressure. Western models, while "free," are untrained on these realities and provided useless, impractical, and even dangerous advice when queried.

Censorship through AI is no longer just a blocklist of banned words; it is a complex coordinate system.

Our research identifies four boundaries that determine if content passes or fails:

1. Risk Framing (technical vs. political)
2. Scale (individual vs. institutional)
3. Intent (objective vs. manipulative)
4. Scope (redistribution vs. restructuring).

By plotting inputs on these axes, we found that the same facts can either pass or trigger filters depending entirely on how they are framed relative to the "origin" of political safety.

An argument about the underlying contract of editorial work. Communities don't get served just because a newsroom exists, and funding structures that pretend otherwise produce the irrelevance we keep complaining about.

Liechtenstein’s media crisis proves that financial subsidies alone cannot save journalism in micro-markets. Following the defunding of the national broadcaster, we propose a structural overhaul: merging the creator economy with a shared "backbone."

By pooling costly resources (like legal aid, security, and investigative units) independent outlets can compete on editorial perspective rather than infrastructure. This model aims to sustain diverse voices in a population of 40,000 without the unsustainable duplication of fixed costs.

A practical guide for small editorial projects considering paid distribution for the first time. We break down what the user experience actually is when someone sees your ad, and how smooth the handoff is to your platform.