The suspicion behind every well-crafted document
The phrase you often hear when someone proposes a beautiful and excellently formatted document is “but you made it with AI!”.
By now, in almost all areas related to conceptual work, the suspicion arises that what we produce is on the steroids of some latest-generation LLM, like Claude, Gemini, or ChatGPT. Indeed, if a simple prompt is enough to produce a complete and well-crafted document, where does my merit lie?
It's a thorny issue, troubling both the workers who use these tools and the people who receive them, whether they're evaluating or using them. It's also a nontrivial issue from the perspective of an employee's duties, who may run the risk of not performing their job ethically.
After all, these tools exist and it would be a total waste not to leverage these revolutionary technologies.
Where is the value of those who use AI?
Can we therefore outline some guidelines that suggest a use of AI capable of bringing tangible added value for those who use it?
In reality, there are many possibilities. First, a well-crafted prompt is already a first value that can be introduced. Not everyone knows how to do this effectively.
But more can be done in this area. There's the concept of "skills," which are essentially a very detailed way of setting context in Markdown (.md) format. You can also start from existing skills and customize them.
Having skills rich in detail and with the right calls to action allows you to add that stylistic and content-rich touch that can make the final document recognizable and distinctive of the person who writes it, exactly as we might write it with our intentions as "writers."
Review, expertise and control
Furthermore, among the many other possible approaches, there's always the option of reviewing the document produced by the prompt. It's always helpful to read the document from top to bottom to check for any errors, especially logical ones, if not factual ones.
This requires specific knowledge of the field the document deals with, and this also enhances the skills of the person editing it.
If the document produced is very long and complex, the cognitive load could be unsustainable and would ultimately negate the advantage of using AI.
Building the text by sections
An alternative strategy could be to build it in sections or sift through it piece by piece. The individual piece can be read quickly, possibly corrected by hand, or, in some cases, the AI can be asked to rewrite it from scratch with specific instructions indicated in the prompt.
With this bottom-up approach, we avoid having ChatGPT write sections we wouldn't have included. It's always awkward, after the fact, to edit a document by cutting out entire sections, potentially compromising its overall coherence.
The human touch remains decisive
Finally, there's always the option of writing some relevant sections directly by hand, without having them generated by AI, which can at most correct spelling or suggest a translation, if necessary.
Alternatively, the editor could write a brief, or dictate a voice message, and with the magic wand of these tools ask for it to be expanded to the length appropriate for the development of the text.
In conclusion, rather than giving up on using these tools, a good idea might be to include as many customizations as possible, so that you can finally make your interlocutor say "great work, it's clear you had a hand in it!"