This brings us full circle then as we’re back to my first point - the only reason anyone would look down on you using AI is because of your corporate culture, which presumably puts more value on how work was created rather how good the end product is. As I said before, the end product is more highly valued at my company than the process used to create it.
Our 7 stage process is Prepare, Research, Compose, Outline, Fill, Review, Critique.
So, for example, last week I did a proposal around tackling TB in a city in the UK.
I started by pulling together a bunch of public health reports I had on file, some strategy documents by local government, some reports that my company had created on similar work and a few other bits and pieces, and put them in the Claude Projects folder.
Then I created a research document to act as a base. I walked the AI through analysing the papers above, asking it to find gaps in the evidence and then do web searches for missing info. I check the content for credibility as I go and manually ask it to remove citations I don’t like, then eventually get it to create what is essentially a fact sheet that is used to support the rest of the process. (Pastebin link to it here) I then use prompts shaped to only use information in this document, which more or less eliminates hallucinations.
Then I told the AI the composition I want, the order of things, etc. Also what not to include.
Then I asked it to create the paper in outline. I check the outline, make changes, then ask it to fill in the narrative, specifying things like tone, grammar, word count and audience.
I did a manual review at this point, reading it in its entirety and making any manual changes I want. I don’t remember how much I changed, but usually 30 or 40 things.
Then I go through a structured critique, asking it analyse its own work from various angles, like factual accuracy, how compelling it seems, whether it uses plain English, whether the proposed work is plausible, and so on. The more important the work, the longer the critique. We all use a snippet manager with a shared set of AI prompts that includes critiques.
We obviously don’t use every single step for shorter pieces of work, but for any long form content like proposals, reports, research, press releases, etc we encourage staff to run through this process.
This is also why we use the phrase Guided Creation. We’re walking the AI through the process and using our expertise at every step to keep the quality high. So its less like asking someone to do your homework, more like telling a team of builders how to build a house for you, section by section. You’re not lifting bricks, you’re coordinating their work.