Best practices for utilizing generative AI for SEO — Kristin M Tynski // Fractl

Kristin Tynski, Co-Founder and SVP Creative at Fractl, talks about generative AI for SEO and PR. As we explore the possibilities of large language models such as ChatGPT, we're discovering various potential use cases beyond the obvious ones like content generation. While many people are already discussing this application, there are exciting opportunities to enhance the results from these models in a variety of ways. Today, Kristin discusses the best practices for utilizing generative AI for content.
About the speaker

Kristin Tynski

Fractl

 - Fractl

Kristin is Co-Founder and SVP Creative at Fractl

Show Notes

  • 02:35
    Potential use cases of generative AI
    Use cases of generative AI like ChatGPT include prompt chaining and utilizing external datasets programmatically through Python and APIs to generate more refined results. These methods can significantly improve the quality of results and make content marketing, SEO, and PR more efficient.
  • 07:25
    Tips for improving AI prompting and results
    To get the best results from generative AI like GPT, it's important to be very specific in your prompts and define the AI's role clearly. This means being specific about what you want the AI to do, how you want it to do it, and what format you want the output to be in.
  • 10:35
    Comparison between ChatGPT and GPT 3 API
    The recently released ChatGPT API has a different format from the GPT-3 API, allowing for a more specific and interactive chat format. With ChatGPT, you define a system message to tell the model who to be, and then provide additional parameters for user and system messages.
  • 12:28
    The flexibility of changing parameters within ChatGPT and GPT 3
    Overall, both models are comparable and will allow you to change parameters such as temperature, length of return result, etc. However, the parameters can only be changed through the playground on Openai.com and not through the chat interface on ChatGPT.
  • 15:31
    Sequencing prompts in ChatGPT and GPT 3
    When sequencing prompts in ChatGPT or GPT-3, it's best to take it one step at a time and refine the prompts as needed. It's important to ensure that the output format is in the desired format as you will need to parse the results between each stage.
  • 17:14
    Effective prompt chaining with ChatGPT
    To effectively use prompt chaining with ChatGPT, start by asking ChatGPT questions to fully understand your goal and refine your prompts. Continually ask for ChatGPT's input and feedback to ensure you're not missing anything and take advantage of its breadth of knowledge.
  • 19:54
    Using prompt chaining for ideation and content creation
    Prompt chaining for ideation involves defining a topic, subtopics, and relevant data sets to generate a list of data-driven ideas that match the criteria. This process allows for more information and original ideas by refining and iterating based on defined datasets and subtopics.
  • 21:56
    Using a multi step setup for organized content planning
    You can use a multi-step process starting with a main topic and then breaking it down into subtopics and subtopics of those subtopics. Next, generate titles from the subtopics, outlines from the titles, sections from the outlines, and finally, combine everything to create a complete blog post.
  • 22:35
    Embedding based clustering using GPT 3 cluster descriptions
    This process uses AI tools like embeddings and clustering algorithms to group keywords into semantically related clusters. GPT-3 is then used to automatically label each cluster with a natural language description, making it easier and faster to define the taxonomy of the clusters.
  • 25:55
    Using GPT 3 for automatic content ideas based on search intent
    GPT-3 can infer search intent for a given keyword, which is useful for content planning and gap analysis. It can be combined with API calls to tools like Ahrefs or SerpAPI for further refinement and analysis, including data visualization and competition comparison.
  • 28:24
    Using APIs to enhance content creation and analysis
    To gather data and insights for content creation, its recommended to use a mix of SEO tools and APIs, such as Ahrefs, SerpAPI, Reddit, and Semrush. Additionally, ChatGPT can provide recommendations on other relevant APIs based on your specific needs.
  • 31:22
    Data enrichment with large language models
    Large language models like GPT-3 can be used to analyze customer data and infer demographic and psychographic information. While bias in the model must be taken into consideration, you can use the information to develop personas, infer age, and other attributes.
  • 32:56
    Helping AI models become more proficient and creative
    A continual refining process is necessary which involves asking for feedback and suggestions at every stage of the conversation. This includes asking the AI model if it has enough information and if it can do better, and then using that feedback to make adjustments and improvements.

Quotes

  • "I'm not by any means a great Python programmer. But with ChatGPT's help, I'm probably five times as fast as I would be if I were just trying to do it on my own." -Kristin Tynski, Co-Founder, Fractl

  • "I think we're going to see a lot of businesses crop up that are based on leveraging prompt chains to do discrete types of tasks." -Kristin Tynski, Co-Founder, Fractl

  • "Be specific and define the role of AI clearly, including its style, voice, format, and every aspect, not just the data or response desired." -Kristin Tynski, Co-Founder, Fractl

  • "Think about it as guiding the model through an almost infinitely expansive, latent space of possible ways to answer a question; the more generic your question, the more generic the answer will be." -Kristin Tynski, Co-Founder, Fractl

  • "Part of becoming good at prompting is realizing all the places you can continually ask for help from ChatGPT." -Kristin Tynski, Co-Founder, Fractl

  • "You're not necessarily going to know what you don't know, but ChatGPT will know what you don't know, at least to some extent." -Kristin Tynski, Co-Founder, Fractl

  • "The reality is that you can use ChatGPT at each stage to help you understand what the chain of prompts might look like, and how to improve it each time." -Kristin Tynski, Co-Founder, Fractl

  • "One interesting thing to me that GPT-3 can do is that it can make inferences. It does a really good job of doing things like inferring search intent for a given keyword, or creating personas." -Kristin Tynski, Co-Founder, Fractl

  • "One of the coolest things is you can ask ChatGPT to give you ideas about what APIs might exist out there to help you with your particular use case." -Kristin Tynski, Co-Founder, Fractl

  • "Data enrichment is a really interesting use case for these large language model tools." -Kristin Tynski, Co-Founder, Fractl

About the speaker

Kristin Tynski

Fractl

 - Fractl

Kristin is Co-Founder and SVP Creative at Fractl

Up Next: