Since OpenAI’s introduction of GPTs and rollout of the GPT Store, ChatGPT Plus users have built over 3 million custom GPT AI assistants for various niche uses. A large portion of GPTs on the store fall within the writing category, where GPT Plus subscribers are inundated with a barrage of content writing GPTs touting SEO specialties.
The problem with the current crop of AI SEO tools is that none are catered toward SEO and marketing professionals looking for a GPT to take care of the lighter search engine optimization tasks so they can spend more time doing the heavy lifting.
That’s why my Cypress North colleagues and I built our own GPT called CypresSEO.
CypresSEO was initially built out of a small internal necessity where our digital marketing team had 100+ unique blog post URLs needing optimized title tags and meta descriptions. Given the cumbersome nature of reading through each blog post to understand what it’s about and perform keyword research, the team sought a custom GPT solution that could do that and write optimized SEO metadata as well.
At its core, CypresSEO analyzes content from any website, drafts search engine-optimized title tags and meta descriptions, and then, like a seasoned SEO expert, audits its work to refine drafts to meet the highest standards of SEO.
We knew we needed the GPT to have a knowledge base built on industry best practices for content analysis, keyword research, contextual understanding, and writing high-performing title tags. After the knowledge base was formed via our in-house expertise, PDF documentation from Google, SEMRush, and other sources, we began to train CypresSEO to process the data input and self-optimize its work before outputting final title tag and meta description drafts in a clear and easy-to-export format.
During the development process, we had to contend with ChatGPT-4’s hourly usage limit put in place for Plus subscribers and tune the GPT to leverage short and long-term memory to expand its bandwidth to input and analyze larger batches of URLs and keywords.
Additionally, CypresSEO began presenting signs of fatigue and hallucinations. So to counter that and ensure optimal SEO metadata output, the GPT was trained to assume the role of a “third-party SEO expert auditor” that reviews the initial title tag and meta description drafts, gives itself feedback, and resumes its original role to revise the drafts and output final peer-reviewed versions. To streamline the export of these final drafts, CypresSEO was instructed to output them in a table containing the hyperlinked blog title, final revised title tag, and meta description for each URL input.
Further Improvements and Optimizations
When we felt confident that CypresSEO’s performance could efficiently scale up without infringing on Chat GPT-4’s usage limits, we introduced additional input and analysis capabilities into its framework. Along with the SEO best practices knowledge base, we trained the GPT to input keywords and search volume data we pulled, then scrape the web to perform secondary keyword research to further optimize its initial and final revised drafts. This improvement increased the SEO performance of the title tag and meta descriptions it wrote and strengthened CypresSEO’s contextual awareness for blog content.
For example, in CypresSEO’s first version, we gave URLs to blog posts about current trends in the insurance space and how recent changes in this area can affect a subset of the industry. CypresSEO hit a logic loop where it insisted on writing title tags and meta descriptions about climate change and natural environment impacts. Providing the keyword input processing and research capabilities removed the chances of these logic loops and misaligned contextual understandings due to the GPT having the keywords and internal guidance for establishing touchpoints in its writing stage of the workflow.
GPT Configuration and User Experience
Below is a portion of CypresSEO’s instructions that are found on the back end of ChatGPT’s GPT Builder tool, providing deeper insight into how OpenAI’s GPT Builder can turn conversational language into a functioning AI assistant.
“…It strictly analyzes only the URLs provided by the user and does not seek additional pages on its own for analysis. After reviewing the provided URLs, CypresSEO generates optimized title tags and meta descriptions based on the content and SEO best practices. The final output includes hyperlinked titles of the pages, along with their revised title tags and meta descriptions, presented in a clear and organized table format as per user request. CypresSEO utilizes extensive knowledge from its database to ensure the application of the latest SEO best practices in its analysis and optimizations.”
As previously noted, CypresSEO’s knowledge base currently consists of nine PDFs sourced from Google, SEMRush, and other SEO and digital marketing best practice publishers, along with our 20 years of experience. As the instructions show, CypresSEO utilizes these PDFs as a foundational check for its workflow’s analysis and content production stages. A benefit of OpenAI’s approach to building GPT knowledge bases is that they are fully modular and can be updated with more up-to-date files without hindering the GPT’s instructions and established workflows. This benefit allows GPT builders to re-tool and fine-tune their GPT to new and expanded use cases.
Use Case Example
We provided CypresSEO with URLs to the following Cypress North blog posts:
- Google Ads & Bidding Types: The Ultimate Source for Google Search Bidding
- Google & Apple Business Profiles: A Comprehensive Overview
- Continuous Improvement Techniques for a Growth-Minded Agency
- B2B Marketing With the Rise of Gen Z: Navigating the Digital Native Workforce
- Why Small Creative Agencies Yield Big Results
It performed site content analysis, wrote the initial title tag and meta description drafts, assumed the third-party SEO auditor role to audit the drafts, and directed itself to revise the initial drafts and output the final versions in the following table format:
At Cypress North, we developed CypresSEO to simplify SEO tasks for professionals. It’s grown modestly since our first version was published a month ago, starting as a tool to optimize blog URLs and now aiding in automated SEO metadata generation and revision. While CypresSEO has made strides in improving content relevance, like overcoming logic loops, it is a small step in the vast field of digital marketing.
It’s a tool designed to support, not overshadow, the creativity and marketing strategy that’s crucial in SEO. As we continue refining it, we hope CypresSEO will be a valuable contribution to the SEO community and your daily workflows.