Editing AI safety communications – Google newsletter

Summary

I worked as Editor for Google's internal AI safety newsletter. As Editor, I worked with AI Researchers and subject matter experts to write about AI safety and increase AI safety awareness.

Helped increase number of newsletter subscribers from 7,000 to 11,000 (a 57% increase) 📈
Increased engagement – subscribers rated the newsletter as 'Excellent', with an NPS score of 74% 🌟
Improved communications. One AI Safety Researcher said: "[Dom's] clear, approachable, and comprehensive writing and editing lifted the whole newsletter to a new level." ✅

(Note: My official role at Google was UX Writer within Google Play – I worked on the newsletter alongside in my spare time.)

The problem

During 2023–24, AI safety was a relatively new and under-explored topic within Google. As a result, the internal AI safety community group wanted to promote awareness of AI safety more broadly within Google.

AI safety is a complex topic, so we needed to make sure the monthly newsletter was compelling, accessible, and included some technical articles for AI experts.

My approach

1. Reach out to AI safety group

After having read a few editions of the monthly AI safety newsletter, I reached out to the writers and offered my services, focussing on editorial and communications skills.

The group organisers accepted, and added me as Editor on the newsletter project.

2. Deep dive on past editions, and data review

I reviewed previous editions of the newsletter – checking for:

Accessibility – was the content easy to understand for non-experts?
Engagement – how interesting was the content, given the interests of our target audience(s)?
Feedback – what were readers saying about the newsletter?
Industry trends – what was the external media saying about AI safety?
Editorial – formatting, consistency, proofreading
Data – was our subscriber base growing? What were the open rates? The click-through rates?

3. Deliver feedback and suggestions

I shared my insights and suggestions with the AI safety community team.

Audience. The data suggested the audience was mostly people who were already enthusiastic about AI safety. So, if we wanted to increase our reach, we needed to broaden our targeting and make the content engaging for non-experts.
Topics. Our audience were more engaged with some topics than others. E.g. stories about government policies weren't as popular as evergreen AI safety principles. So, we decided to focus on those more popular topics.
Unique stories. Feedback from subscribers indicated that they didn't just want a repeat of AI news stories that they could read in external media – they wanted more of a unique spin from Google staff insiders.
Article length. Articles were often very long – some more than to 5,000 words. So we established recommended range of 1,000–3,000 words.
Format and readability. I suggested we break down longer sentences and paragraphs. To use more bullet points and subheadings. Break down acronyms. And where possible, to use simpler words and more common words.
Images. I suggested we include images in the newsletter, which is an easy way to make content more inviting, especially because AI safety could be a dry topic.

4. Write and edit using our editorial process

I worked with the writers to establish a new editorial process:

Collate ideas for articles over the course of the month, using a centralised document
In an editorial meeting, we agreed on which topics to write about, and assign writers to each topic
The writers sent me their first draft. I revised using the tracked changes function, and left suggestions in the comments. For example, "Is there a way to say this in a clearer way?" or "Would a non-expert know what this means?"
The writer amended their article based on my suggestions, then sent back to me for proofreading
I made the final edits, ensuring the meaning of the content stayed the same

Sometimes I would also work with outside contributors (i.e. people from Google DeepMind) to produce articles using our new editorial process.

Sometimes I would author my own articles. For example, I wrote an article called "The story of Elon Musk's mission to create safe AI".

5. Compile and publish

I would then put the final touches on the newsletter:

Compile all finalised articles into the email template
Write the "TL;DR" summary at the top of each article
Add images, embed YouTube videos, etc. (People liked our inclusion of AI safety memes!)
Add and test hyperlinks, including anchor links
Publish final newsletter article to the site

6. Review stats and feedback, plan next newsletter

2 weeks after the newsletter had been published, we would review how it performed. The newsletter contained a link to a feedback form, which often gave us useful insights and suggestions on how to improve. We also tracked Net Promoter Score over time.

We used these insights to plan and improve the next newsletter.

Outcomes

Organisation outcomes

Over the course of the 7 months I spent as Editor, the number of Google staff subscribed to the newsletter increased 57% from 7,000 to 11,000. Therefore, we firmly reached our goal of pushing the AI safety agenda within Google. 📈
Our subscribers rated the newsletter as 'Excellent', with an NPS score of 74%. 🌟
Significantly increased quality of output. ✅ Some quotes from my colleagues:
- Johannes Gasteiger, AI Safety Researcher (now Alignment Scientist at Anthropic), said: "His clear, approachable, and comprehensive writing and editing lifted the whole newsletter to a new level."
- Simon Möller, Senior Software Engineer (now Workstream Lead at EU AI Office), said: "Dom consistently improved the quality of our writing with his detailed feedback and suggestions".
- Swante Scholtz, Senior Software Engineer, and leader of the Effective Altruism community within Google, said: "He was always very reliable, flexible, and his perspectives made our articles significantly more nuanced and approachable."

Personal outcomes

Learned some technical AI and AI safety topics ⚙️
Deepened my skills in collaborating with subject matter experts, and writing for technical audiences ✏️
Built relationships across the organisation 🤝
Made a difference to AI safety culture within Google and Google DeepMind 💡

Below, a couple of recommendations from my ex colleagues who worked on the project: