ChatGPT is a cool AI system that can write human-like text, but it can often produce generic and uninteresting blog posts. So we created MERCURY GPT, a Python program that uses OpenAI’s GPT API to simulate an entire writing team. This team includes a freelance copywriter, an experienced editor, an SEO professional, a photo researcher, and a production editor. By chaining together multiple AIs, we can create high-quality blog posts based on a prompt and a target keyword that provide genuine insights and rank on Google.

Instead of just sending a prompt to GPT, we make AIs talk to each other. We have a specific task: write a good blog post that's as good as a human. We conjure up a variety of AIs and chain them together in a consistent, predictable way to solve our specific problem.

We turned to the structure of writing teams in marketing agencies to create our AI team. Generally, website owners decide on a topic and hand it to a freelance writer, who researches and writes a draft blog post. Then, an editor reviews the draft, and sometimes, an SEO person offers suggestions. The freelancer then rewrites the draft based on feedback, and a production editor adds photos, links, and formats the draft for publishing. The website owner makes final edits and publishes the post.

We recreate the entire writing team in AI by wiring together multiple, customized instances of GPT. We can specify what type of editor we want to conjure up depending on the article's topic. For example, we can conjure up an AI editor with extensive experience working for finance when generating a finance-related article. The AI editor looks at the AI copywriter’s draft and provides feedback and advice.

In our internal testing, we did a great job of writing coherent, SEO-optimized blog posts. We reviewed the draft, fact-checked the output, and added our own experiences to make the article compelling and make sense. The article passes an AI check for originality and plagiarism and scores a 38 on Surfer for SEO.

MERCURY GPT is more expensive than ChatGPT, but it's still pretty cheap. It's modular, so we can easily add new "team members" with new roles. Team-based AI is widely applicable and can be used in nearly any industry. Combining AI entities and having them work together to improve their output is a powerful process.

We believe that AI-generated teams can help beyond writing. For example, AI-generated legal teams can structure a legal case, identify documents, draft complaints, and collaborate. AI-generated medical teams can do "rounds" on a patient by leveraging various AIs playing the role of different specialists, nurses, and providers.

The basic idea behind MERCURY GPT is that by having AI entities work together, we can move beyond the arms race of building ever-bigger language models. GPT is impressive, but it needs a little help from its virtual friends.