Dan Hinckley - Developing Automated SEO Tools

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. SEO landscape began to change with the release of ChatGPT
3. What’s changed for SEOs?
4. Search Engines started their shift to Answer Engines
5. But… we also have increased access to tools that power search engines Google’s Vertex AI
6. Costs have decreased by up to 99.9% Before After
7. The best way to learn about these tools is to build with them.
9. Access to LLMs and related technology at lower costs lay the foundation for developing powerful automated SEO tools
10. 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
11. What if we could solve those problems, at least from an SEO automation perspective?
12. “Don’t use LLMs to help with title tags or meta descriptions because they can’t count characters”
13. 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”
14. Giving LLMs the ability to check their work reduced error rates from 60% to 1%
15. “LLMs can’t be trusted, they hallucinate and make things up”
16. 1. Give LLMs the ability to say “I don’t know” 2. Include the answer to the questions people ask in the prompt
17. 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.
18. 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?
19. 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.
20. Getting Answers For SEO Google’s Vertex AI
21. Solving LLMs limited memory with vector embeddings Text Embedding Model Text as Vector Tech SEO Connect [0.06,-0.03] … [0.05,-0.09]
22. 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.
23. Vector embeddings & cosine similarity scores allow us to find the most similar text and often the answers to questions to share with LLMs
24. Does Google use similarity scores to rewrite meta descriptions?
25. What if we could score content the way search engines do?
26. All of these new tools at affordable prices allows us to automate the information gathering and content scoring to provide better SEO recommendations
27. What are some examples of ways to automate SEO?
28. Reviews SERPs to identify search intent & content freshness (API + LLM) +
29. Identify helpful content gaps from competing sites (LLMs) + Tip: Use JSON to create comparative tables for LLMs
30. Discover information gain opportunities by identifying unique facts on competing websites (LLMS) +
31. Reviews content on a page to see if it’s answering People Also Ask questions (API & LLMS) +
32. Create vector embeddings and generate similarity scores for client & competing pages (API) + Google’s Vertex AI
33. Create vector embeddings for every page on a website & competing sites to measure topical authority (API) + Google’s Vertex AI
34. Highlight image alt tag and file names to allow SEOs to make quick recommendations
35. Entity extraction with Google NLP to identify gaps with competitors (API) + Google’s NL API
36. Assess topical authority to identify if a website needs more related content (API) + Google’s Vertex AI
37. Identify internal linking opportunities (API) + Google’s Vertex AI +
38. And so much more!
39. 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.
40. In conclusion
41. 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.
42. Thank you!