Editor · Writer · Communications director
Paul O’Neil
Editor, writer and communications director. Native British English. Over 15 years in senior communications roles, including seven years in editorial leadership. Hands-on with AI tools that scale good copy without diluting the voice.
Selected work
Seven pieces, each captioned by audience, channel and brief
-
Brand story and destination
The Breguet Classic Tour
WorldTempus, 2018A brand initiative brought to life through place and craft: “these artisans work in tiny workshops hidden away in the lesser-known areas of Paris.”
Read the article The Breguet Classic Tour -
Short form
Your next carbon-fibre watch might be knitted
WorldTempus, 2019An unfamiliar manufacturing process made short, clear and intriguing in under 450 words.
Read the article Your next carbon-fibre watch might be knitted -
Search
The top ten Swiss watch brands
WorldTempusRanked in Google’s top three for its exact title throughout my five years as editor-in-chief.
Read the article The top ten Swiss watch brands -
Partnerships
Brick Parkour Asian Tour launch
FIG, 2023A title sponsor, a venue partner and a six-stop tour in one clear story, written by the project manager who ran the tour.
Read the release Brick Parkour Asian Tour launch -
Executive voice
President’s foreword
FIG Yearbook 2024Written for the FIG President, in his voice.
Read the foreword -
Social and performance
Digital impact
FIG Yearbook 2024The biggest engagement increase of any international sports federation, Tokyo 2020 to Paris 2024. Instagram engagements up 254%.
See the spread -
AI and scale
Gymbot: an editorial agent pipeline
Independent prototype, 2026Tone of voice written as rules, human sign-off on every draft, at least an hour saved per article.
Read the case study
Independent writing
From my own sites: watches, photography and search
-
Data as narrative
Always the bridesmaid
tick-ticker blog, 2026Opens on Raymond Poulidor’s eight Tour de France podiums to tell the story of watch houses that have never won the GPHG’s top prize. Award data turned into a story.
Read the article Always the bridesmaid -
AI in practice
Who will win the GPHG 2026? I asked six AIs
tick-ticker blog, 2026Six AI models asked to pick a winner. The ones that admitted their limits proved the most reliable. A hands-on AI test told with wit, from Paul the octopus onwards.
Read the article Who will win the GPHG 2026? -
Travel and place
Lighthouses of Nova Scotia: a photographer’s journey along Canada’s Atlantic coast
pauloneilphotography.comSix lighthouses, from Peggy’s Cove to Louisbourg, in words and my own photographs: “watching the fiery red ball descend, while the sky shifted through orange, purple, and red.”
Read the article Lighthouses of Nova Scotia -
GEO and SEO
Thinking, fast and cited: what ChatGPT’s reasoning mode means for AI citations and GEO
CiteSite blog, 2026Why a brand can be cited by ChatGPT one minute and invisible the next: its reasoning mode shares only a quarter of its sources with quick answers. Research made readable, opening on a Nobel Prize, a pub quiz and a Semrush study.
Read the article Thinking, fast and cited
Case study · 07
Gymbot: an editorial agent pipeline
Independent prototype, designed and built in my own time, 2026
Gymbot turns a one-line request, such as “2026 World Artistic Championships vault final”, into a fact-checked draft article ready for a human editor.
The problem
Gymnastics results arrive in large volumes, spread over many official documents. Every report has to be accurate, quick and written in a consistent voice.
How it works
- Brief. An orchestrator agent works out what is being asked, and a separate check decides whether the request is about an event or an athlete.
- Research. Agents pull event details, entry lists and results from the federation’s official site and documents, and skip the results step for previews.
- Memory. Verified facts are stored and reused, so nothing is researched twice.
- Writing. A writer agent follows a house style brief: a defined voice, a list of banned stock phrases, British spelling and no claim that isn’t in the research.
- Sub-editing. An editorial QA agent checks for opinion presented as fact, internal contradictions, legal risk and spelling, then approves the draft or returns specific corrections.
- Human sign-off. Every article lands in the CMS as a draft. Nothing is published until a person approves it.
What it shows
Tone of voice written down as rules, the way you would brief a network of copywriters or AI tools. Facts kept separate from voice. Speed and volume without losing editorial judgement over the final copy.
Result
At least an hour saved per article, with every draft reviewed by a human editor. First built on OpenAI’s Agent Builder, then rebuilt on Claude (Anthropic API) with Cloudflare and GitHub Actions.
About
Based near Lausanne, Switzerland
I write, edit and set editorial standards. For five years I ran WorldTempus, a leading international watch-industry platform, publishing more than a thousand articles a year with a network of freelance writers. Before and since, I have led communications for luxury watch brands and Olympic sports federations.
I build my own AI tools, from content pipelines to a GEO audit platform, because the best way to judge AI copy is to know exactly how it is made.
- Head of CommunicationsWorld Gymnastics, 2022–2026
- Editor-in-chiefWorldTempus, 2014–2019
- Managing editorEuropa Star, 2011–2014
- Press officerOMEGA, 2000–2005
- LanguagesEnglish (native), French, German
- EducationUniversity of Bath · University of Salford