Editorial Standards

Scope

These standards cover every video, clip, Clip Library entry, and web page Distilled AI publishes, including takes a host also posts on their own channel. The AI tools that help us research, script, and edit follow them too.

1. Accuracy and Sources

1.1 Sourcing

• Every factual claim on screen carries a visible source. Full links go in the caption, the comments, or the video’s sources page at distilledai.org/v/<n>.
• We prefer primary sources, such as peer-reviewed studies, official data, and people’s own words, over articles that summarize them.
• A video’s central claim gets a real review of the evidence. We check whether studies agree and how much their results vary.
• A side stat can rest on one recent, representative study, hedged in plain words like “most” or “probably.”
• We never invent, inflate, or guess a number. If we haven’t counted it, we don’t say it.
• Numbers from our AI Discourse Monitor come from platform data and AI-coded analysis, and both contain errors. We say how we got the number and how large the sample is.
• Data on AI use goes stale fast. We date our stats, and we flag an old usage number as a likely undercount.

1.2 Quotes and Clips

• Quotes match the words the person said or wrote.
• A quote or clip keeps the meaning it had in context. We check what came before and after.
• We show when the speaker said it.

1.3 Uncertainty

• We say so when the evidence is mixed or when nobody knows.
• We label forecasts as forecasts and keep them apart from what’s been measured.

2. Fairness

• We take arguments from every side seriously, from people who fear the worst to techno-optimists.
• We state each side’s case in a form its best advocates would accept, and then we test it.
• We don’t start from a conclusion. Any verdict follows from evidence the video shows.
• We critique claims and decisions, never a person’s character.
• Stitches and responses extend or explain the original video. They don’t dunk on its creator.
• We hold powerful companies and their leaders to account, and we stay fair to the people who work inside them.
• We never mock people for using AI.

3. Facts, Opinions, and Tone

• Hosts share their own takes and mark them as takes, with words like “I think” or “my take.”
• A take never replaces evidence. When a host reaches a conclusion, the video shows what it rests on.
• We don’t distort facts to fit a view, and we don’t leave out evidence that cuts the other way.
• We avoid inflammatory or insulting language, and we don’t build conclusions on fear or outrage.
• Humor is welcome, but never at the cost of accuracy.
• We explain things for a general audience and give the background a viewer needs to follow the point.

4. Politics and Policy

• We cover AI laws and policy, and elections where AI is at stake, with the strongest case on each side.
• A video about a specific bill explains what it would do and the best arguments for and against it.
• A host may share a view on AI policy in general, marked as a take and paired with the strongest opposing case.
• We don’t support or oppose any candidate or party.
• We don’t take sides on ballot measures.
• We don’t ask viewers to contact lawmakers or officials about a bill, or to sign petitions urging them to act.
• Verdicts and graphics avoid party colors and partisan framing.

5. Funders and Conflicts of Interest

• We hold our funders to the same standards as everyone else. Our independence policy explains how.
• Any video or web page that covers MIRI, its staff, or its work says that MIRI funds us. Videos say it on screen or in the caption.
• Team members tell the project lead about jobs, funding, or investments that touch a topic they cover. The video discloses any tie a viewer would want to know about.
• We disclose free products, early access, or payment from anyone a video covers.

6. Other People’s Content

• We use footage we don’t own only with the owner’s permission or a good-faith fair-use basis. For us, fair use means we comment on, critique, or explain the clip and use only as much as the point needs.
• Clips from other creators carry their handle on screen.
• Viewer comments appear exactly as posted, with slurs and personal details removed.
• We don’t show private people’s personal details, and we take extra care with minors.
• Guests we film for a video sign an appearance release first.

7. AI in Our Work

• We use AI to help research, transcribe, edit, and make some graphics. A person checks every factual claim and source before a video posts.
• Generative AI appears only in minor background footage, where we’d otherwise use stock footage or nothing. It never carries a video’s point.
• We never use AI to create or alter footage, audio, documents, or data and present the result as real.
• When a video shows synthetic media on purpose, such as a deepfake demo, the video labels it on screen.
• We apply each platform’s AI-content label where its rules require one.

8. Clip Library and Clips Account

• Every clip records the speaker, the date, and the source.
• A clip’s title matches what the speaker actually says in it.
• The library offers a clip for download only when the footage owner has given permission. Otherwise it links to the source at the right timestamp.
• Standalone clips post on a separate clips account, never on the main Distilled AI accounts.
• A response video states when the interview took place and that the speaker wasn’t responding to the video shown.
• Clips featuring MIRI staff disclose MIRI’s support in the caption and follow each platform’s rules for paid and political content.

8.1 How We Choose Clips

1. Public questions come first. The AI Discourse Monitor ranks the questions and claims that come up most in comments on viral AI videos, and the top-ranked ones get clips first.
2. No view count decides which viral claims we answer. We weigh each claim, and the team decides which ones matter most to address. A claim gets a response video when the library holds a clip that answers it directly.
3. For each question, we look for answers from more than one perspective, and we clip every credible one the library holds.
4. We choose the passage that answers the question most directly and completely, with the speaker’s caveats left in.
5. A speaker’s employer or funder plays no part in the choice. Clips of MIRI staff go through the same rule.
6. We publish the list of speakers and interviews in the library, so anyone can check its balance.

9. Corrections Policy

Our corrections policy explains who can ask for a correction, how to report an error, what happens next, and how we fix mistakes. It also holds our public corrections log.