After monitoring ChatGPT for the best part of a year, using it to help with research, I began to employ it to construct potted history drafts in July 2026. I should make it clear that I only use it for subjects for which I already have a solid understanding. For example, I have previously covered transport in many of my place and British histories, in fact even in some sporting ones. I consider therefore that I am in a strong position to check what the chatbot produces with respect to a potted history of transport. In essence, the process for producing a potted history is as follows:
- Ask the chatbot to produce a broad outline, detailing its scope, plus a draft table of contents
- Discuss it, pointing out any obvious omissions, and agreeing the general focus and emphasis
- Ask it to produce a draft. Read and comment on the draft, looking for any errors and omissions, plus checking on items such as the general context for events, plus the article’s focus and emphasis, as stated above
- Stepwise refinement follows as multiple drafts are produced and checked until I am content with what has been produced. The chatbot contributes its thoughts on possible changes during these steps
- I then transfer the article to the website and make any final changes, which are usually cosmetic.
The initial histories have been polished, but they do not have my “voice”. It is a gradual process to train the chatbot so that it sounds more like me. This is an ongoing process.