Using genAI tools to Produce Original Content

This post lays out a methodology and step-by-step instructions for producing original content, for example, white papers, marketing collateral, SEO content, press releases, website copy, and training material.

This methodology works equally for content targeted at expert audiences and lay audiences. In fact, content produced using this methodology can easily be repurposed for different audiences by the genAI tools.

Methodology

This is done with only two free AI tools;

1.     NotebookLM

2.     ChatGPT 4o

We all know ChatGPT can ingest anything it is given and rewrite it to achieve different objectives such as creating a summary, or changing the style of writing. However, in order to leverage these capabilities of ChatGPT to produce high-quality, truthful, original content it needs to receive a high-quality original source document. Creating a high-quality original source document is both time consuming and difficult.

This is where NotebookLM comes to the rescue. By using both NotebookLM and ChatGPT in a coordinated workflow, you open up the possibility of creating lots of high-quality, original content, super-fast, and for free. The following methodology shows how it is done.

The process is as follows:

·      Select a topic

·      Decide on the main point or thesis of the final document

·      Identify several reference documents. These can be news articles, internal documents, original drafts, bullet point outlines, or anything that provides background information, data, and context for your topic

·      Upload these documents to NotebookLM

o   NotebookLM connects directly to Google Docs or will accept PDFs

o   NotebookLM indexes these documents and uses them as a RAG database

o   When queried, NotebookLM extracts information from these documents and submits questions to Gemini 1.5 Pro to generated augmented answers

·      Select a story outline. You already have a topic and a main point or thesis. Now decide how you want to tell the story. Use any standard outline, for example:

o   Situation-complication-solution

o   Three main points

·      [PRO-TIP ADVICE] Ask NotebookLM a  series of questions in a hypothesis-driven manner. Follow your chosen outline structure and essentially lead-the-witness. Ask questions so that the answers naturally fill in the content for your sections. This is the key skill involved. In order to get a good end-result, you need an a priori idea of what you want in the end. Typically, you’ll want to assert an insight (or hypothesis), and then lay out the logical argument to defend or prove that insight

·      The output NotebookLM generates for each hypothesis-driven question is called a “note”. Notes are typically formatted as bullet-point outlines with citations.

·      Save each note as it is generated (this is important because unlike ChatGPT, NotebookLM does not automatically save every chat)

·      Merge the notes into a single Word (or equivalent) document simply with copy-paste

·      Edit the Word doc to remove artifacts such as the interactions with the Notebook chatbot and the paraphrasing of questions. In other words, spend 5 or 10 minutes cleaning it up. The end result is the substantive parts of the answers in bullet-point form

·      [PRO-TIP ADVICE] Insert section headings to delineate the groups of bullet points that pertain to each of the questions you posed to NotebookLM. The intention of this is to make the structure and story-flow obvious to ChatGPT. When looking at the answers, feel free to reorganize if you think you can improve on your original outline, but this is not necessary

·      Beyond these steps, there is no need to edit the bullet point answers from NotebookLM, although you are welcome to if you feel so inclined

·      Save this document in a format to upload to ChatGPT. PDF works as do others

·      You should feel confident that ChatGPT will produce an accurate and truthful document because it is working from a bullet-point outline that is based on and referenced to the original source documents uploaded to NotebookLM

·      Give ChatGPT specific instructions on what you want it to produce. The following prompts are illustrative but have also been tested and work well:

o   Using the attached bullet-point outline, you will create a white paper that is, structured, well-written, cohesive, polished, and logically organized. You may draw on other work we have done together on this subject that you can find in our other chats. The tone of this white paper should target a sophisticated audience with domain knowledge of the subject matter. Use examples where possible

o   Create a new version of this white paper. This time, you should address a lay audience. Do not truncate any of the information. All the content is still relevant and important. However, you should spin this for a less knowledgeable audience this time. If you are able to introduce metaphors and analogies to make points more clearly, you should do this liberally. Use examples where possible. Finally, you should give the paper an optimistic and opportunistic tone

·      The end-products from ChatGPT will likely be close to your desired result(s). It behooves you to spend 30 minutes or so editing the final output. ChatGPT tends to use some clunky phrases and has a few tell-tale signs of its fingerprints. But these are easy to spot and fix

Critique of the Methodology

This method and step-by-step guide are particularly powerful in a couple of situations:

1.     Time Constraints – This method is incredibly fast. With no real editing, you can go from ideation to end result in as little as 15 minutes. In applications like SEO, this approach may make the most sense because it yields acceptable quality with excellent productivity. For more important pieces, the entire process can be accomplished in under two hours with an excellent end product(s).

2.     Voluminous Reference Material – Sometimes the sheer volume of background material is too daunting to even begin the task of producing a piece of content. For example, if the background material consists of peer-reviewed scientific research papers (of the type published on Nature or NIH), these reports can be 70 pages each, full of data and analyses. If you’re staring at twenty of these papers, you could be looking at a month of work just to digest and understand them. By contrast, if you upload these same reports to NotebookLM, they will be indexed and ready to query within seconds. You can start creating your “notes” immediately.

This methodology has been tested against both these conditions and works equally well for each.

If you have questions or want to know more, please contact us.