Kristin Tynski - Automating Marketing Tasks With AI

Transcript

1. AI Meets Content Marketing: Actionable Automation Strategies (With Code!) Kristin Tynski - SVP Creative @Fractl October 17 - 18 2024 CAM Raleigh 409 W. Martin Street Raleigh, NC 27603
2. I'm Kristin, Co-Founder @Fractl Co-Founder at Frac.tl a leading content marketing and PR agency that leverages data-journalism style content to earn press and links at scale. • I have overseen the creation and promotion of 5,000+ press earning Content Marketing campaigns in the last 15 years. • We leverage LLM’s extensively in many workflows, but always with a human-in-the-loop. • I’ve personally been using and talking about the implications of large language models on our industry since 2018 (ThisMarketingBlogDoesNotExist.com). • About Me
3. 1. AI's Role in Marketing a. From pattern matching to reasoning. b. AI agents as autonomous team members. 2. The Content Paradigm Shift a. Evolving beyond keywords to intent and context. 3. Effective AI Partnership a. Navigating the balance between human and AI collaboration. 4. Actionable Automation Strategies a. AI-driven content creation and SEO. b. Real-world applications with code examples. 5. The Future of Marketing a. Shift focus from tactical execution to strategic orchestration with AI. Today’s Agenda How to think about, plan for, and execute in the new world of AI first marketing.
4. The AI Marketing Revolution Stay ahead or fall behind
5. Unprecedented Growth 100M 1T 12 Years Active users in 2 months for OpenAI. Fastest ever product growth Estimated words generated by LLMs so far Majority of web content will be AI generated
6. Surviving the Commoditization of Content • Could this content have been written easily by GPT4? • Information gain or recapitulation? • End-game of generative AI is near zero cost content creation. • Not 20yrs, not 10yrs, not 5yrs. This is happening now.
7. A CMO in Your Pocket? Time AI Capabilities Human Role Key Challenges Impact 5 years ago Limited, poor quality Expert-dependent Narrow applications Minimal 2 years ago Basic automation, LLMs Heavy oversight Integration Niche efficiency 1 year ago Multi-task agents Pipeline management Scaling Process automation Today Complex reasoning Significant guidance Autonomy vs control Knowledge work shift 5 years ahead Autonomous AI teams Strategic direction Governance, adaptation Industry disruption
8. These Graphs Should Scare You…
9. This one too…
10. So Should This… SITUATIONAL AWARENESS by Leopold Aschenbrenner - ex OpenAI
11. https://services.google.com/fh/files/misc/the_roi_of_generative_ai.pdf
12. https://services.google.com/fh/files/misc/the_roi_of_generative_ai.pdf
13. Dead Internet Theory Coming True
14. Key Trends Driven by AI in Marketing/SEO • Decreased Organic Traffic: LLMs offer direct answers, reducing clicks to sites. • Selective Click-Through: Users click references in AI-generated answers with high intent but potentially lower frequency. • Verticals Impacted Differently: Those relying on "well-known answers" suffer most • Even Less Data Transparency for marketers • Index’s Still Matter: LLMs need access to the current Internet, but the signals that matter most to them will likely shift. • Data-Driven Standard: Content backed by data becomes the norm. • Even 'Winners' Lose Traffic: Overall decrease in organic search traffic. • Verifiable Identity and Closed Gardens: Necessitated by bot and generative AI tsunami • Is a Perfectly Efficient Market the End State of AI+Marketing?
15. To Survive, You Need AI Marketing Superpowers The Power of Agents and Pipelines
16. LLMs + Agentic Frameworks + External Apis = Superpowers • Navigating the latent space • Agentic Prompting • Pipelines and chained prompting (Looping scripts/pipelines) • Leverage external sources of data for new capabilities. • APIs and web scraping are key. • AutoGPT, CrewAI, Open Interpreter and other Agentic Frameworks • Examples and free Google Colab scripts ahead!
17. Skillfully Navigating the Latent Space
19. Improved Reasoning With Agentic Prompting ✨"Instantiate two agents competing to find the real answer and poke holes in the other's answer until they agree, which they are loathe to do. Each agent has unique skills and perspective and thinks about the problem from different vantage points. Alan Turing: Top-down agent Donald Knuth: Bottom-up agent Both agents: Excellent at the ability to think counter actually, think step by step, think from first principles, think laterally, think about second order implications, are highly skilled at simulating in their mental model and thinking critically before answering, having looked at the problem from many directions." ✨
20. Context Cleanup
21. Critiques,Counterfactuals, and Checklists Challenge LLMs’ Thinking for Better Results • “Critique your last answer in-depth” • “Provide counterfactuals and edge cases” • “Use checklists extensively to plan, then follow the checklist” • “Broad to Narrow - Leverage AI’s information synthesis skill” • High-Density information context (documentation, data, hard facts)
22. Advanced Pipelines 1. Multimodal 2. Iterative improvement 3. Agentic 4. Enabled by APIs 5. Multi-step 6. Complex processes comprised of many discrete tasks
23. One Input - Many Outputs Automatically Generate a Summary, Article Outline, Long-form Article, and Tweet Thread from a Youtube URL
24. Automatic High Quality Long-Form Content
25. Source of Truth + Automatic Semantic SEO + Iterative Outlines = Content Worth Ranking
26. Automatically Find Keyword Gems
27. Automatic Onsite SEO and UI/UX Analysis
28. Build Oracles with Simple RAG Pipelines and Finetuned Models
29. Automated Short Form Virality Understanding
30. Automated Reddit Assistant
31. Automatic FAQ’s
32. LLM SEO - Are You in the Training Data?
33. Semantically Complete Sitemaps with LLMs and Graphs
34. Automatic Cluster Evaluations • Applications: ◦ Competitive analysis ◦ Keyword research ◦ Trend research ◦ Social media research ◦ Data story creation ◦ Trend analysis • Steps: ◦ Scrape/collate Serps, Keywords, or Content ◦ Get embeddings (OpenAI or Huggingface) ◦ Do clustering ◦ Use LLM to evaluate clusters
35. Automatic Persona Generation • Infer personas based on: ◦ Keyword ◦ SERPs ◦ Metadata • What to infer: ◦ Demographics ◦ Psychographics ◦ Affinities ◦ Purchase intent ◦ SWOT
36. Automatic Trend Analysis • Scrape posts on reddit by keyword: ◦ Sort by top ◦ 1day, 1week, 1month, 1year, all time • What to infer: ◦ Primary topics and hierarchies ◦ Emotional activators ◦ Entity summaries ◦ Abstractive summarization ◦ Notable trends
37. Corpus Generation and Analysis • Generating massive research datasets in short periods of time with LLMs enables data journalism never before possible: • ◦ Previously prohibitively time consuming ◦ Previously too difficult for one human to grok all together ◦ New avenues for unique content creation
38. 1. Crawl 2. Scrape 3. LLM content analysis 4. Anchor text recommendation Web Crawling with Content Analysis and Anchor Text Recommendations
39. Understanding Brand Visibility in an AI First Search Environment • AI search is even more black-box than Google • Can it be probed? • Optimizing for future visibility in new foundational models by understanding training datasets
40. 1. Scrapes a given URL 2. Analyzes the content with an LLM 3. Extracts relevant data 4. Generates JSON-LD schema 5. Outputs the enhanced HTML Automatic Schema Improvements with LLMs
41. Code Interpreter, Canvas and Code Executing Agents • A data scientist polymath assistant for everyone. • In-context editing • Code execution and self healing
42. On Your Radar • Agentic Frameworks (autogen, crewai, dozens of others • Openai Realtime Voice API • Kling and Text to Video AI/APIs • Dead Internet of Image Results already happening • Improving tooling and user friendly agentic AI tools • Openai Strawberry and improved reasoning • Larger and larger context windows • Suno/Udio and high quality music generation • Open Source AI’s catching up • Domain Specific Languages (DSPs) • AI Breakthroughs (next page)
43. AI Technology Model/Tech Disruption Link Notes Video Generation MoStGAN-V Blockbuster films for all MoStGAN-V Transforms entertainment & marketing Music Generation DiffuseRoll Instant hit songs DiffuseRoll Democratizes music production Image Generation DALL-E 3 Custom visuals for anyone DALL-E 3 Changes marketing & design Protein Prediction AlphaFold-latest Faster drug discovery AlphaFold-latest Speeds up biotech innovation Reinforcement Learning Scalable Multi-Agent Complex tasks automated Scalable Multi-Agent Redefines robotics & logistics Liquid Neural Networks Loihi-2 Real-time AI adaptability Loihi-2 Key for autonomous systems Mixture of Experts OLMoE Cost-effective large-scale AI OLMoE Reduces infrastructure needs Fast Training ZeRO-Infinity Faster innovation cycles ZeRO-Infinity Speeds up model training Multi-Token Prediction Multi-Token Efficiency Faster, human-like AI responses Multi-Token Efficiency Enhances digital assistants Probabilistic Relational Models PRMs AI decision-making under uncertainty PRMs Important for finance & healthcare Binary Networks XNOR-Net AI on everyday devices XNOR-Net Enables mobile/edge AI Dataset Quality Improvement MS-COCO Cleanup Reduced AI bias MS-COCO Cleanup Vital for trustworthy AI Quantum Machine Learning Quantum ML Quantum speedup in AI Quantum ML Revolutionizes optimization Thermodynamic Computing Thermodynamic Computing Energy-efficient AI Thermodynamic Computing Enables sustainable AI
44. • Full repo of AI Automation Scripts • https://github.com/openai/openai-cookbook • Intent, Personas, and Buyer Needs by Keyword • Automatic Article Outline Generation by Analyzing The Article Text of Top Ranking Pages for a Given Keyword • Automated Subreddit and Post Title Recommendations Based on Any Article • Free Script - Automatic Question Expander (Free Alternative to AnswerThePublic Powered by GPT3) zz • Automatically Generate a Summary, Article Outline, Long-form Article, and Tweet Thread from a Youtube URL ZZ • Automated Newsjacking Content Ideation • Automatic Brand or Entity News Media Monitoring and Analysis • Free Script - Web Crawling with Content Analysis and Anchor Text Recommendations • Automatic Content Gap Report and Analysis With Clustering and Cluster Descriptions (Public) • Automatic Schema Improvements with GPT4 (Public) • Automatic Long-Form Article Generation With Semantic SEO Driven Outlines and Iteration • Exploring the Future of Research and Long-Form Content: Harnessing the Power of Super Large Context Window LLMs with Python and Chained Prompting! • Understanding Your Brand's Visibility in the Outputs of Large Language Models - A Start to LLMSO (Large Language Model Search Optimization) • Automatic Long-Form Article Generation With Semantic SEO Driven Outlines and Iteration - Now With Images! (Public) 25+ AI Automation Scripts
45. Thank you! Kristin@frac.tl https://www.linkedin.com/in/kristintynski if you have more questions.
46. And Thank You Robots! Kristin@frac.tl https://www.linkedin.com/in/kristintynski if you have more questions. Ya know, just in case…