How I write.

How an article is written

I write most of my articles with the help of Claude, an AI model made by Anthropic. I’d rather be transparent about how I use it.

Each article starts with me. I choose the subject and then begin to plan my argument. The AI then asks me questions about the piece, one at a time: where it should open, what the strongest objection is, what the thrust of the argument is. My answers, typed or dictated, become the framework for the article.

The AI also helps me with the research, usually by checking the figures I use against the original sources. The figures and research findings I cite link to their sources, so you can check them too.

Next, it suggests a first draft from my plan. I rewrite it, often heavily, and add what I know from working in schools. Articles typically go through several rounds of toing and froing with the AI before I’m happy with them.

When an article is ready, it is formatted for the site. Nothing goes live until I have read the final version and agreed to put it out. So the ideas and judgements in every article are mine, and so is responsibility for any mistakes. If you spot one, please let me know.

Information and work for schools

I never put information about clients, staff or pupils into the AI when writing articles.

I draft reports and other work for schools myself, and use AI only for light editing, proofreading and formatting. When I do, I use an account that does not use what I put in to train AI models. Azimuth Partners Ltd is registered with the Information Commissioner’s Office.

Why I work this way

What matters is where the thinking comes from. Asking AI to write a whole piece for you is very different from planning what you want to say, setting out your thinking, doing the research and then having AI help you produce a first draft that you edit and rework. There are many virtuous uses of AI in writing, and this is the one I try to practise.

There are also times when I need to grapple with the shape of the argument and find the right form of words unaided, so that I better understand what I am trying to say and where my reasoning leads. For those, I use my trusty notebook.

AI detectors cannot show who did the thinking, which is why I explain it here.

This is the balance I suggest to schools. Use technology for a clear purpose, keep a person in charge of the judgement, and make sure that whoever puts their name to it could talk you through it.