How Our Custom GPT Accelerated Our Content Flywheel

Submitted by jordan on Tue, 07/30/2024 - 17:00

The evolving landscape of crafting and publishing engaging and valuable content has only intensified over the past decade. Just when a stellar piece gains traction, a shift in perspectives, market conditions, or global events can change everything. For a company like ours, where competence is paramount, the level of mastery required to support our clients effectively is a methodical achievement, not the product of sudden innovation.

The Evolution of Our Content Development Process

Traditional Workflow: Slower and More Fragmented

Initial Meetings: Our team met semi-monthly to brainstorm ideas that emerged from client interactions, industry announcements, or relevant news.

Idea Consolidation: Notes were scribbled on paper, discussed via video, and shared across collaborative spaces like Zoom, Slack, and Google Drive.

Content Development: What felt timely one day could take weeks to develop into a comprehensive, well-presented piece.

Reflect on your current content development process and identify areas where delays commonly occur.

Integrating LLMs: A Testing Phase

Exploring Capabilities: Recently, we began incorporating large language models (LLMs) into our writing process to test their capabilities.

Early Results: Initial use was sporadic and experimental, aiming to understand how AI could assist in various stages of content creation.

Experiment with AI tools in your content creation process to identify potential efficiency gains.

Transforming Content Creation with Custom GPT

Building the Custom GPT

Training on Existing Content: Last month, we created a Custom GPT using OpenAI’s ChatGPT platform. We trained it on our previous blog posts and crafted a YAML prompt to align it with our audience's expectations.

Streamlined Workflow: This Custom GPT now handles everything from generating outlines to producing full articles and gathering supporting documents.

Consider developing a Custom GPT tailored to your specific content needs and audience.

New Workflow: Efficient and Cohesive

Weekly Meetings: We now meet weekly for 30-60 minutes with a skilled writer who manages the Custom GPT and oversees all projects.

Role Transformation: Our writer has transitioned into a manager, mechanic, and editor, ensuring the Custom GPT produces high-quality content.

Rapid Execution: The time to completion for many articles has reduced from weeks to days, thanks to the streamlined process facilitated by the Custom GPT.

Evaluate the roles within your content team to maximize efficiency and output quality.

Benefits of Using Custom GPT in Content Creation

Increased Efficiency and Output

Faster Turnaround: The Custom GPT allows us to produce content much faster, reducing the time required from weeks to days.

Consistent Quality: By leveraging AI, we maintain high quality in our content while freeing up human resources for more critical tasks.

Assess the potential time savings and quality improvements AI could bring to your content creation process.

Continuous Improvement and Adaptation

Ongoing Training: The Custom GPT continues to learn and adapt based on new content and feedback, ensuring it stays relevant and effective.

Agility: Our content creation process is now more agile, capable of quickly adapting to new trends and changes in the market.

Implement a feedback loop to continually improve your AI tools and processes.

Challenges and Solutions

Verifying Information

Manual Verification: Despite the efficiency gains, it is critical to verify every quote, statistic, and concept to ensure accuracy.

Addressing Issues: Occasionally, the GPT needs reminders to provide URLs as text rather than links, requiring manual adjustments.

Establish a verification process to ensure the accuracy and reliability of AI-generated content.

Building a Foundation

Regular Posting: Our consistent posting schedule provided a solid reference for training the LLM, highlighting the importance of having a substantial content base.

Gradual Integration: The transition to using a Custom GPT was gradual, allowing us to refine our process and prompting steps over time.

Build a robust content library to train your AI tools effectively.

Conclusion: Embracing AI for Content Creation

Accelerated Output: The integration of our Custom GPT has significantly accelerated our content flywheel, allowing us to produce high-quality content more efficiently.

Focused on Quality: While AI handles much of the heavy lifting, our focus remains on ensuring the quality and relevance of our content.

Invitation to Explore: If you’re curious about incorporating a Custom GPT into your organization, we’re excited to discuss how it can enhance your content creation process.

Reach out to us to learn more about integrating Custom GPT into your content strategy for accelerated and high-quality output.

(Embed Typeform for free Consultation)

[Initial Notes] How our Custom GPT accelerated our content flywheel

The seeming shell game of crafting and publishing engaging and valuable content has only expanded in the last decade. Just when a great piece lands and gains traction, a sea-change takes a swipe at the world and shifts perspectives, expectations, market conditions, and eventually reality.

Staying ahead is important yet, for a company like ours, where competence is rightfully expected, the level of mastery required to effectively support our clients is a methodical accomplishment, not achieved with the immediacy of radical innovation.

If you follow our blog you’ll notice we’re picking back up the pace on publishing, due in large part to a careful and considered incorporation of AI. Initially, our workflow looked much like a traditional content development approach.

How We Did It:

Our team would meet semi-monthly and hash out ideas that bubbled up from clients, became apparent from industry announcements, or were undeniably relevant news to our client base. We’d bang out notes on paper, via video, and push them together across collaborative spaces (zoom, loom, slack, drive, etc.)

What felt timely one day might take weeks to pull into a thorough piece with depth and thoughtful presentation.

Recently, we began working with LLM’s in the writing process as much as possible, simply to test their capability.

Then, last month, we created a Custom GPT using OpenAI’s ChatGPT platform. We “trained” it on our previous blog posts, and crafted a YAML prompt to help it understand our audience.

How We Do It Now:

We meet for 30-60 minutes weekly with a skilled writer who manages the Custom GPT and pushes all the projects to completion.

“I’m much more of a manager, mechanic, and editor than I was in earlier years developing content,” he shared in a recent meeting.

Once we land on a story concept we use the LLM as follows:

Request a thorough and comprehensive outline of the topic being sure to touch on the elements that will resonate with our audience.
We used to send the outline for approval and upon feedback make any adjustments.
With those notes incorporated into the outline, we’d ask the ChatGPT to follow our YAML and craft a full article.
Last, we’d ask for support documents from the web, with their links as text (to verify) and where they’d fit into the article.

Now, we’re able to accomplish all those steps through a single Custom GPT. It takes our concept and pushes it into a full article along with links to supporting quotes and statistics.

Our writer reviews the final product, deleting extemporaneous phrases and verifying linked materials.

Our time to completion has dropped on many articles to one to two days, from one to two weeks. There are still some which require deeper research or some tweaking of the prompt to get it right but overall the time required of a writer has been reduced dramatically.

This didn’t happen overnight.

We were posting articles regularly which created a useful reference for the LLM to train on. We used off-the-shelf LLM capacity to refine our YAML and prompting steps. And once we’d completed ten respectable articles, we then went about creating the Custom GPT.

It still needs to be reminded (sometimes) to return URLs as text rather than hot links because occasionally the links it returns are just blue text, not actually linking to anywhere.

It’s also critical to verify every quote, statistic, and concept in the final articles because our readers are highly knowledgeable and will call us out when they smell BS.

So, yes, we get pieces done faster. And, our work now is more focused on the quality of output.

If you’re curious about how you’d potentially use a Custom GPT in your organization, we’re excited to talk about that with you.