Write for Common Signals
Common Signals publishes evidence on how the public responds to messages about AI. If you have tested a message, studied a reaction or run a campaign, and can show what happened, we would like to publish it.
What is a Common Signals piece?
A Common Signals piece gives someone who explains AI to the public a finding they can use, and shows the evidence behind it.
Our readers are communicators: campaigners, researchers who talk to the press, creators, journalists and press officers. Some work on present-day harms and some on existential risk. A good piece is useful to all of them.
We publish four kinds of piece.
- Reception studies. You analysed how an audience responded to a video, post, interview or news story about AI. Our method and codebooks are public, and you are welcome to use them.
- Message tests. You ran a split test, a survey experiment or a paid campaign with more than one version of a message. Tests where nothing worked are as welcome as wins.
- Evidence from other fields. Climate, public health and science communication have decades of research on fear appeals, messengers and framing. Tell us what one finding says and how far it carries over to AI.
- Case notes. You explained AI to a real audience (a town hall, a classroom, a committee hearing, a comment thread) and can describe what you tried and what came back.
Every claim we publish carries one of three labels, the same ones used in How to communicate about AI safety. Established means a meta-analysis or a finding replicated across fields. Tested means a single experiment or representative survey on AI. Observed means a comment study or a campaign's own report, which we treat as a hypothesis. You don't need to label anything before you pitch. We do that together in the edit.
What don't we publish?
- Arguments for or against a policy. We take positions on what the evidence shows and how strong it is. We take none on what governments or companies should do.
- Pieces about whether AI is dangerous. Our subject is how people respond when someone tells them it is, or tells them it isn't.
- Promotion of a campaign, product or organisation. You can write about your own campaign if you share the numbers, including the ones that disappointed.
- Claims a reader can't check. Every figure needs a source, a country and a date.
- Criticism of named individuals. Decisions, policies and messages are fair subjects. People are off limits.
- Studies of a creator's work that the creator doesn't know about. We ask creators before we publish a study of their content, and we ask contributors to do the same.
Can I use AI tools?
Yes, if you tell us how. We use Claude to code comments in our own studies, check a sample by hand, and say so in every method section. We ask the same of you: say which tool did what, and who checked it. That applies to drafting as well as analysis. Your name goes on the piece, so every sentence in it should be one you would defend to a sceptical reader.
Why write for us?
- An edit. Ben Matthews, who runs Common Signals, edits every piece with its author and checks each figure against its source.
- Readers who do this work. Pieces are published on commonsignals.org and cross-posted to our Substack and LinkedIn newsletter.
- A public home for your data. If your piece rests on a dataset, we can publish it in our data explorer with its codebook, under a CC BY 4.0 licence and credited to you.
- You keep your rights. Republish the piece wherever you like, with a link back to the original.
We can't pay contributors. Common Signals has no funding yet, and you should know that before you write a pitch.
What topics are we looking for?
Last updated 13 November 2026. A good pitch on something missing from this list is welcome too.
Messages on specific issues. Which messages hold up, and with which audiences, on:
- loss of control and catastrophic risk
- regulation and independent oversight
- jobs and the economy
- children and young people
- deepfakes and synthetic media
- AI, war and national security
- a pause, a ban or a treaty
Questions that cut across issues
- Messengers. Who do different audiences trust to talk about AI, and what happens to that trust under scrutiny?
- Fear and what to do about it. What changes when a warning arrives with an action attached, and when it arrives without one?
- News moments. How does public reaction to an AI incident shift in the days after it, and what difference does independent verification make?
- Platform and format. Does the same message land differently as a podcast clip, a reel, a thread and a broadcast interview?
- Audiences we haven't reached. Our own studies so far cover English-language comment threads. Evidence from other languages and countries would fill a real gap.
- Replications and challenges. Our datasets are open. If you recode one and get a different answer, we want to publish that.
How do I pitch?
Email [email protected] with "Pitch" in the subject line. One to three paragraphs is enough, and you don't need a finished draft. Tell us:
- What you found, in one sentence.
- What you measured: how many people or comments, where, and when.
- Which communicator needs this, and what they would do differently after reading it.
- What limits the finding: a small sample, a confound, or your own stake in the result.
- Who you are and how you are connected to the work.
We reply to every pitch, usually within two weeks.
How long should a piece be?
Most pieces run from 800 to 1,500 words. A study with its own data can run longer, with the method and data notes at the foot. Lead with the finding. We use British English and will handle house style in the edit.
What happens after we say yes?
You send a draft and we edit it with you. We check every figure against its source, add the evidence labels and agree a headline. Nothing is published until you have approved the final text. The piece carries your byline and a line stating any affiliation or stake you have in the subject. If a reader finds an error after publication, we correct it with a dated note.