How To Build a Cyborg Content Strategy: Research & Planning

Our first installment gave an overview of what the cyborg approach to content development was all about. In this next part, our journey into this methodology will focus on the planning phase — the first step in the process. This is the most critical phase to ensure quality content outputs.

Without proper cyborg content strategy research and planning, your end product runs the risk of lacking consistency in voice, tone, factual accuracy, and formatting.

By understanding how to properly develop a Human + AI planning model, you build a strong foundation for your long-term Cyborg content system. There are several steps you will want to prioritize during the planning phase, which I break down in this series so you can build a cyborg content strategy.

Research in the Cyborg Content Strategy

The most important lesson to learn early when working with AI models is that if you rely only on their training knowledge you are unlikely to get high-quality responses.

In other words: garbage in, garbage out.

The AI models don’t understand the difference between factual information and disinformation. They do not prioritize based on expertise. This makes the creation of quality research to guide the model imperative.

So, our goal here in the research phase is to build a document that can then be utilized to generate an outline within our brief. There are three key areas to look at during the research process:

Tools

Your research will not get far without the proper tools, and the right tools will ultimately guide you in how to format your prompts to elicit the best possible output from the AI model.

Utilizing APIs will be the best bet for pulling relevant data to guide briefing and content planning, no matter the topic or vertical.

Data-Rich APIs

Data-centric content works well with the current LLM models on the market. By tapping into data-rich APIs you can give your content accuracy and authority.

ScraperAPI

If you are a developer, then ScraperAPI is a great solution for data and website scraping. If you aren’t a developer, though, don’t worry — the DataPipeline tools make it easy to schedule and download crawls without code.

Agenty

Agenty is, in my opinion, a better selection for no-code scraping. It offers great tools in its system, but it also has a Google Chrome plugin that can make free scraping with parsed results into JSON a breeze.

Anthropic

Claude 3 models, such as what Anthropic uses, are currently my go-to for creating research documents. These allow for 200,000 token context windows in the prompting, which means you can put about 150,000 words of information in your prompt.

This keeps you from having to multi-shot prompt your research document. The results are on par with GPT4 models in terms of ability to follow directions and the quality of output.

Custom GPT

If your resource material for your research is something you have internally or in document formats, csv, or PDF then building a custom GPT is a great way to generate quality research documents.

When building and using a custom GPT bot for research, like in OpenAI’s ChatGPT, you can simply upload the files you need to generate your research documents into the GPT’s knowledge base.

Brief.news

A note of transparency: I am currently heading up growth for the company behind this product, but that’s because I truly love it.

Brief clusters thousands of news stories per day into singular topics then summarizes them and delivers the key points you need to know. If you are generating content for blogs or news sections, this is an amazing utility.

Not only does it offer the key points, but you can view each source’s summary, clearly see each source URL, and navigate to the source for more information.

Prompts

Below is my current prompting approach for generating solid research documents from Anthropic Claude 3, which you can use with any model.

Human Interaction

The most important aspect of a cyborg content approach is that humans should be guiding the research data collection process. Letting a system run on its own is bound to retrieve bad data.

As the system retrieves the data, humans need to interact with the resulting research pages to add a quality assurance level. LLM models will hallucinate and go off task, no matter how solid the prompting or data is.

Rogue information in the data can also lead to issues. This is why human interaction should point the research (and creation) process in the right direction and ensure it delivers the desired results.

Developing Your Style Guides

Style guides are a key component of transferring knowledge about the company, voice, and purpose of the content to a writer. It becomes critical when working with generative AI. That being said, you will need to take a closer look at some core elements.

Head to the CopyPress Knowledge Base for more on how to craft your cyborg content strategy.