How I Built WritingGPT, a Fully Automated AI Writing Team

WritingGPT is an AutoGPT that "simulates an entire writing team, crafting high-quality blog posts [with] only a prompt and a target keyword... articles that provide genuine insights — and that rank on Google."

"AutoGPTs are AIs talking to AIs... the output of one AI system as the input for another... chain together multiple AIs, enhancing their collective capabilities.

Writing "teams have a fairly consistent structure:

• owner of a niche site will decide on a topic...
• freelance writer... research the topic and write a draft blog post...
• editor ... review the draft and make suggestions... Sometimes, a dedicated SEO person will weigh in with suggestions
• freelancer rewrite their draft ...
• production editor will review the draft, insert photos, add links, format & upload...
• owner ... make their own edits, ensure that the piece is in their voice ... work in personal experiences ... hit publish."

This his recreated in silico. Two inputs:

• "An idea for a blog post... as simple as a title ... or [also including] notes about what the post should include
• target keyword or phrase for SEO...

GPT-4’s roles allow me to create a series of instances of GPT-4, each ... a member of a traditional writing team. GPT-4 can perform all the roles well — it just needs to be told which role to play using different system messages."

1. hand prompt to GPT-4-freelance copywriter, which "writes a first draft of the target article."
2. copywriter sends the draft to GPT-4-editor... "I can specify what type of editor I want to conjure up ... [eg] with extensive experience working for tech publications... [it] provides a bulleted list of pieces of feedback and advice."
3. then to GPT-4-SEO, who has “10+ years of experience” and "provides a list of pieces of advice for the AI copywriter based on their draft and also the target keyword specified."
4. Back to GPT-4-freelance copywriter, which "takes this feedback ... and rewrites", then sends draft to
5. GPT-4-photo, which suggests photos* and sends the draft to
6. GPT-4-production editor, "who formats the draft using HTML, adds links to the websites of any businesses or locations mentioned in the article, and then spits out the final version."

(*) "GPT-4 can’t produce photos (yet), so ... I instructed... AI photo researcher to give me a photographic brief just as a real photo researcher/editor would... I could potentially integrate with DALL-E API to add custom illustrations."

The author provides an example: how he got from a basic prompt to this article, which "passes an AI check from Originality ... as 99% original... 0% plagiarized."

For SEO it scored an OK 40 ("average competing article is around 52") as it's short (~500 words): "For more complex topics, WritingGPT writes longer articles". Nevertheless, "Within six hours ... it had ranked #2 on Google for the target term"

Limitations:

• the chaining of AIs creates more hallucinations.. despite including guardrails (eg "telling WritingGPT not to perform tasks for which it has inadequate information")
• relatively expensive: using GPT-4 to "constantly handing long-ish drafts and chunks of text back and forth... 1,000 words ... cost about $1... also about 10 minutes"

Tips:

• provide detail... eg "notes from a visit to a local restaurant... helps it produce truly interesting and unique content... less AI-sounding"

Future: as it's modular, add in other team members - eg

• "AI fact checker...
• AI critic, who would debate the points ... helping copywriter to make the article better ...
• once GPT-4’s Browse function gets added to the API... AI research assistant ... to feed WritingGPT up-to-the-minute data.

Final notes:

• it's an AutoGPT, but less inscrutable: "leverages the power of having multiples AIs talk to each other, but also ensures that their work followers a clear operating procedure and thus yields useful, consistent output."
• the appoach "avoids the arms race of building ever-bigger language models. WritingGPT is smarter, not bigger"