Outsourcing Digital Marketing to AI: 60-70% Ready Content

The CMO issued a challenge to build an AI bot capable of performing the marketing team's job, specifically focusing on efficiency and time savings. A key constraint was the CMO's aversion to unoriginal AI content, setting a high bar for quality. The success metric for the project was achieving 80% publish-ready content with minimal human intervention across three core tasks: a product updates blog post, a YouTube script for monthly updates, and a newsletter for tens of thousands of subscribers .

The underlying objective was not to replace human marketers but to demonstrate AI's potential to save significant team time without compromising content quality. This meant the AI needed to produce content that was indistinguishable from human-written material in terms of originality and effectiveness. The experiment aimed to prove that AI could augment, rather than diminish, the output of a high-performing B2B product marketing team.

The first task involved generating a product updates blog post. The process began by downloading 12 months of existing product marketing blog posts and feeding them into a ChatGPT project. Instructions for ChatGPT were initially generated by ChatGPT itself, and after providing context, the system was almost ready to create the first draft .

Initial drafts had simple errors, but refining the instructions with more details on writing style significantly improved the output. Feedback from Andre, a product marketing expert, highlighted areas for improvement: the content was too technical, lacked specific benefits, and needed more examples for complex features. Incorporating Andre's advice, the instructions were updated, leading to a much-improved second draft that was deemed 'beautiful' and 'pretty good' .

Following the blog post success, the next step was generating a YouTube video script. This involved collecting a year's worth of video scripts from product updates and adding them to the ChatGPT project. The instructions were updated to repurpose content from the blog post into a video script format. Initially, this process seemed straightforward, producing a script that sounded similar to existing content .

However, a critical issue arose when attempting to generate a new blog post after updating video script instructions; the system broke, losing previous formatting and placeholders. A breakthrough occurred by downloading and re-uploading the blog post document, then asking ChatGPT to generate the video script from the re-uploaded content, finally achieving success. For the newsletter, a specific template was created, and content from the AI-generated blog was used to populate it, resulting in a 'pretty good' output .

Upon presenting the AI-generated content to the CMO, Tim, his initial reaction to the blog post was that it 'feels AI' due to formatting and awkward phrasing like 'maps cleanly.' He preferred a human-written example, noting the AI's convoluted language. For the video script, Tim initially identified the human-written script quickly but then struggled to differentiate, indicating some improvement in AI output .

Tim articulated that the AI bot lacked genuine product understanding and real-world context of how Hrefs is used. This deficiency led to inconsistent language and an inability to communicate effectively with the target audience, often being either overly complex or unprofessional. The newsletter also received criticism for pitching the video instead of the product, highlighting the AI's lack of strategic intent. While the AI saved significant time (reducing a week's work to 3 minutes), Tim concluded it was not a success, as the content was only 60-70% publish-ready, falling short of the 80% goal .