AIdas
Every answer from an LLM is an echo. We help you hear whose.
More and more people ask an AI model instead of searching. The model picks the sources, and its answer carries someone's version of events. AIdas checks whose: which outlets a model cites, whose framing it repeats, and how the answer changes with the wording, the language, or the month you ask it.
We work with newsrooms, funders, schools, and researchers to make that visible and act on it.
Who uses AIdas, and why
A funder mapping influence in a region
You back independent media in a country and want to know who really shapes what people learn there, now that many of them start with a chatbot. We test the questions local audiences actually ask and show which outlets the models lean on, where those outlets are based, and whether state or propaganda sources are gaining ground. You see where your grantees show up, where they are invisible, and where the gaps are.
A newsroom reading its competitive environment
You want to know whether AI answers about your beat or your market cite you, your rivals, or no local source at all. We run the queries your readers would run and show who the models treat as the authority. It is a read on your standing in the channel that increasingly sits between you and your audience.
A teacher building AI media literacy
You want students to see for themselves that a confident answer is not a neutral one. In a class session we ask the same contested question several ways and watch the sources and the framing move. Students leave able to spot it on their own, and you keep a simple method you can run again each term.
A newsroom manager raising the team's guard
Your journalists use these tools every day and you want them alert to the bias built in. We run a hands-on session on your own topics, in your languages, so the team sees where a model mirrors a leading question, where it pushes back, and where it quietly favours one side. Awareness they carry into their reporting, writing and prompting.
What we offer
1. Workshop: test it together
For newsrooms, schools, libraries, and civil-society teams.
A hands-on session on your own topics and languages. Your team brings a question, we run it across several models side by side, and you read the bias, the sourcing, and whether the model holds a line or just agrees with whoever asked.
Your team leaves with increased AI literacy and vigilance, a method they can repeat and a first baseline to measure against.
2. Monitor: watch it over time
For funders, research teams, and organizations tracking a topic or a market.
We run a fixed set of prompts across many models on a schedule and put the results in a dashboard you can come back to: who gets cited, how balanced the sourcing is, when models refuse to answer, and how all of it drifts from month to month.
One run is a great positioning snapshot. If you run it repeatedly and you'll catch change early.
3. Study: map the whole ecosystem
For public-service media, media funders and investors, and researchers.
We map who dominates what AI says about a place, topic, or community: the outlets cited most, where they are based, and whether state-controlled or propaganda sources are creeping in.
You get a clear picture of the information base the models draw on, and where independent voices are missing, with recommendations you can use.
Where this comes from
AIdas grew out of commissioned research, in collaboration with Factnameh, and with support from ASL19 and the Open Technology Fund, and was first presented at GlobalFact in Vilnius and SRCCON in Minneapolis.
The method is already published:
- Meet AIdas: our prototype AI bias training tool
- AIdas: a visualization of how AI models answer Iran
- The new gatekeepers: auditing how AI models assess authority (Iran)
- How prompt framing steers models toward different source libraries
- DeepSeek's double life (China)
- Censorship as a coordinate system, not a blocklist
Work with us
Tell us the context you care about, a topic, a country, a community, or an audience, and we will scope the right format: a workshop, a monitoring dashboard, or a full ecosystem study.