Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Dan Hinckley - Developing Automated SEO Tools

Dan Hinckley - Developing Automated SEO Tools

Tech SEO Connect

October 23, 2024
Tweet

More Decks by Tech SEO Connect

Other Decks in Marketing & SEO

Transcript

  1. Dan Hinckley Co-founder Go Fish Digital, an Agital Company Developing

    Automated SEO Tools What We Learned About Search In the Age of LLMs October 17 - 18 2024 CAM Raleigh 409 W. Martin Street Raleigh, NC 27603
  2. But… we also have increased access to tools that power

    search engines Google’s Vertex AI
  3. Access to LLMs and related technology at lower costs lay

    the foundation for developing powerful automated SEO tools
  4. Generate text based on patterns in their training data rather

    than performing actual mathematical operations. Bad At Math LLMs sometimes generate content that sounds plausible but is factually incorrect or completely fabricated. Hallucinations LLMs can only retain a limited amount of information in their context window. Interacting with lots of information can be difficult. Limited Memory 01 02 03 But… LLMs have weaknesses
  5. “Don’t use LLMs to help with title tags or meta

    descriptions because they can’t count characters”
  6. Solving LLM’s Math Problem: Let Them Check Their Work Tip:

    Data returned to LLMs from functions work better if they include instructions Example: “Title Tag is too long with 65 characters, please write a shorter one”
  7. 1. Give LLMs the ability to say “I don’t know”

    2. Include the answer to the questions people ask in the prompt
  8. User Prompt: Did the Dallas Cowboys Win the Super Bowl?

    Sent To LLM: Did the Dallas Cowboys win the Super Bowl? Use the data below to answer the question, if there is no answer in the data, reply with I don’t know. [Data: The Dallas Cowboys did not go to the Super Bowl] LLM Response: The Dallas Cowboys did not win the super bowl.
  9. How do we as Technical SEOs include answers to all

    possible questions or queries in the prompt? How do we solve the limited memory problem?
  10. We need to gather information (APIs, scrape), organize it (database),

    and make it easy to retrieve (code) to include in a prompt to LLMs. Act like a Search Engine In other words: we need to crawl data, organize it in a database, and retrieve relevant information based on what a user needs.
  11. Solving LLMs limited memory with vector embeddings Text Embedding Model

    Text as Vector Tech SEO Connect [0.06,-0.03] … [0.05,-0.09]
  12. Vector Embeddings & Similarity Once you have converted text to

    embeddings you can then measure the distance between them to understand how similar they are Storing and retrieving the most similar text to any prompt or query from vector databases allows you to give LLMs just the information they need when they need it.
  13. Vector embeddings & cosine similarity scores allow us to find

    the most similar text and often the answers to questions to share with LLMs
  14. All of these new tools at affordable prices allows us

    to automate the information gathering and content scoring to provide better SEO recommendations
  15. Identify helpful content gaps from competing sites (LLMs) + Tip:

    Use JSON to create comparative tables for LLMs
  16. Reviews content on a page to see if it’s answering

    People Also Ask questions (API & LLMS) +
  17. Create vector embeddings and generate similarity scores for client &

    competing pages (API) + Google’s Vertex AI
  18. Create vector embeddings for every page on a website &

    competing sites to measure topical authority (API) + Google’s Vertex AI
  19. Assess topical authority to identify if a website needs more

    related content (API) + Google’s Vertex AI
  20. At Go Fish Digital, we’ve built Barracuda to automate these

    things. It gathers important page level information and organizes it so that an SEO can easily make recommendations for page improvements. And it gathers all that information in under 4 minutes.
  21. 1. There are more tools available to help us see

    the web the way search engines do. 2. Accessing and using these tools is cheaper and easier than ever. 3. We can use these tools to automate information gathering and scoring to allow us to make recommendations that make an impact.