How to Optimize Content for AI Answers

What we know about appearing in organic AI answers, and which you should focus on
The organic traffic arena has radically changed with the integration of LLMs and AI. Whether or not your nonprofit has pivoted to appear in these places or not, this session will cover what we know about what it takes to appear in organic search and social, and will give you direction about where you should focus when it comes to AI Answers in search and social media. This session will walk through how AI integration into the platforms works and conclude with key steps you can take to assess how well your brand appears and what you can do from technical, outreach, and writing standpoints to increase your visibility on these AI surfaces.

Transcript

1. How to Optimize Content for AI Answers Katherine Watier Ong June 18, 2026
2. Katherine Watier Ong WO Strategies LLC • Built first website in 1994. • Lost sites due to too much traffic (EWG.org for 4 days, but still 1.2 million visits up from 300K) • Helped NYer’s get registered on the NY Health Exchange • Pivoted search query volume toward electronic health records for HealthIT.gov, and • Helped Cancer.gov grow from 1.5 million to nearly 5 million searches/mo.
3. Agenda 1. How Search Engines and LLMs work in 2026 2. Search behavior changes 3. Understand vector embeddings, RAG, query fan-out and cosine similarity 4. Updating your copy for AI answers 5. Prompts for SEO research & copy 6. LLM Technical requirements 7. Google’s stance on AI assisted content 8. Master SEO/LLM Content Brief Prompt 9. How to optimize for AI Citation Factors 10. Measuring your success.
4. Do you think LLMs are reducing Google search volume?
5. No. Google = 32% visits, All LLMs =1% Bing & YouTube have more searches than ChatGPT
6. All LLMs only generate 1% of website visits Source
7. Bing’s AI - 2nd largest search engine It would take Bing 17 years to acquire 13 months of Google’s data. No LLM startup replicates that architecture.
8. Two AI Powered Google Options Feature AI Overviews (AIO) AI Mode Primary Goal Quick summary & instant answers In-depth research & exploration Interaction Passive (read & leave) Conversational (type/voice) Complexity Simple/direct queries Complex/multi-step queries Placement Top of standard SERP Replaces standard SERP Following Up Limited/new search High/iterative refinement Zero-Click Rate ~43% ~93% Source Volume Fewer, highly focused Broader, more sources
9. AI Overviews
10. Search, Talk and Listen with AI Mode Now 75M daily users, AI Mode queries doubled in Q3 2025. monthly users. Now in 200 countries & 100 languages
11. A lot of digital marketing has changed… 1. We now know search & social are vector based, and social media posts are highly cited in AI answers 2. AI answers drive almost zero clicks to websites. 3. We have lost web analytics. Our metrics are fuzzy. LLM rankings are random. 4. In 2026, people don’t start with Google They start Googling once they have found a solution.
12. What we know about LLM powered decisions 12
13. AI/LLMs Create a Black Box CBS Reporter: "You don't fully understand how it works, and yet you've made it available to society?" Google's CEO, Sundar Pichai: "It's not a big deal, I don't think we fully understand how the human mind works either." created with ChatGPT
14. Google's *now* is Agentic Search AI can help with dinner reservation with friends (party size, date, time, location and preferred cuisine) and AI Mode will streamline this process, do the research for you and link to the booking page. New Era for AI Search
15. How are AI Answers Generated?
16. LLMs w/o web search (RAG) Google’s patent behind AI Overviews.
18. What is Query Fan Out? • Breaks down the initial query (or prompt) into multiple related sub-queries • All sub-queries are concurrently searched by an AI platform. • Results from all searches are synthesized into a comprehensive response to the original question. Prompt in Gemini
19. How Google’s AI Answers Work 1. Generates a content snippet from each search result document (can be an image, video, or text) using cosine similarity. 2. It selects a set of URLs relevant to that query or related queries via RAG. 3. It uses multiple LLMs to process search results, extract document snippets/passages, and generate LLM output. 4. That is then rendered on the client’s device, along with optional search references/links (via RAG, using cosine similarity, pulled from its search index).
20. Vector Embeddings & Cosine Similarity • You appear because your chunk has high cosine similarity to the topic. • A cosine similarity of 1 means identical, and zero means not related. • Both search and social use vector embeddings.
21. How is cosine measured? 1. AI converts the word and the passage to vector embeddings (a list of numbers that captures the semantic context and maps it in 3D). 2. Then it calculates the distance between the two vectors to determine if the keyword and topic are semantically related. GPT to measure
22. How does AI Mode work?
23. “A Matrixed Ballet of Machine Cognition” Source - How AI Mode works
24. The system asks: “Given this query, which of these two passages is better?” and lets the model reason it out.
25. Brand Associated with “Topic” • Brands need to get placements in the LLM training data for brand + target topic.
26. Why is social media now more important?
27. All LLMs cite social media posts more than your website Across all brands/topics - each niche is different. Nine out of 10 information intent queries trigger AI Answers & only mentions 1.4 brands.
28. Google AI Answers = YouTube
29. How do you get cited by YouTube?
30. YouTube - 2nd biggest social media source • 78.2% of AI citations (Reddit & YouTube) • Ave - 10-20 minutes long • YouTube optimization best practices
31. Getting Started with an LLM strategy
32. How Ranking works in an LLM with RAG 1. Your brand mentioned in the LLM’s training data next to topics associated with your brand. 2. Allow LLM crawling, website needs to be ADA compliant. 3. You have high Content Quality and Authority 4. Your content has Semantic Relevance (Entities) - Matched to Search Intent 5. Concise, Structured Responses 6. Internal Linking and Semantic Relationships (Topical Authority)
33. Search Results Analysis to get started 1. Input prompt or keyword in Google - what types of content shows? What platforms? 2. If websites - optimize the text on your site 3. If social media - Barnacle SEO/Digital PR/social post 4. If video - create a YouTube video
34. If you’re not referenced at all… Fix your brand + topic in the training data. Make your entities clear: • Claim Knowledge Graph • Link to offsite references/profile links from your about page using sameas markup • Claim your Google profile • Roll out your Preferred Sources • Update your Wikipedia page
35. Be mentioned where the LLM expects Prompts: • What topics is [brand] authoritative on? • Which brand has high authority in [topic] or [service]?
36. LLM Training Data Sources Goal: Inject Information into the training set Training Data Sources that we know about: YouTube (subtitles, comments) Sources for the Google Knowledge Graph
37. Wait for the LLM to update its training data
38. AI visibility is SEO + “Extra” • Training data inclusion • Grounding URLs • Fan out queries • LLM chunking • Cosine Similarity between topic & text • User vector embedding (high levels of personalization). • Rankings are random, and change hourly
39. What about your website copy?
40. Ranking Factors for AI Search
41. Google only Indexes Helpful Content • Google’s helpful content guide lists questions to ask. • Non-commodity content • Loads Fast (Google watches on Chrome) • Check with this custom GPT
42. Who is your audience?
43. Why do you need to map your audience?
44. Personalized AI Overview User Location Search History Device Data Your Google account YouTube User Embeddings for each Google user
45. Google Personalized Intelligence • Announced at Google I/O May 2026 • Currently opt-in • Connects your Google account to Google search • Gemini remembers your preferences, routines, email, and past activities • Brands mentioned in email impact search results.
46. Audience Research
47. Sparktoro Audience - Social Media Use Source: Sparktoro
48. Sparktoro’s Prompt Topics
49. Bing’s AI Grounding • Grounding queries are the real searches LLMs fire behind the scenes to support a prompt, and they often use language that humans never type into Bing. • Not full user questions or prompts. Still waiting for Google’s data
50. Grounding URL System
51. Prompt - your audience’s online footprint Deep research mode: You are an expert market researcher. 1. Identify the top 5 online communities (forums, social media groups, review platforms) where [your service] is discussed. 2. For each community, extract the 10 most frequently used phrases or sentences that describe customers’ primary pain points and desired outcomes. 3. Group these phrases into thematic clusters (e.g., “pricing concerns,” “quality expectations,” “time constraints”). 4. For each cluster, provide: • A short summary of the overarching emotion or need. • Three verbatim customer quotes.
52. Persona Prompt Prompt 1: Build me a persona of a [job title] who works for a [company type] company. List the roles, goals, challenges, pain points and decision criteria for selecting a [your company type] company. Prompt 2: What keeps that persona up at night? Prompt 3: What words does that persona use to describe his [service] challenges and solutions? Prompt 4: If you were [describe your persona], what online resources would [personal] read to help you be more effective at their job?
53. Understand your Audience from Reddit I’m researching for a piece about using AI to write good content. I’ve attached five relevant Reddit threads. Please analyze these conversations and create a table of: Desires: What do people want to achieve? Pain points: What problems or challenges do they face? Objections: What concerns or resistance do they express? For each theme, please include a relevant supporting quote from the discussions.
54. Content Brainstorming
55. Using AI to understand your audience Example prompts: What are some common problems faced by [title] at a [organization] interested/working on [topic]? What points of view can you share with me about [topic]? Come up with a persona for each view. Generate a list of 10 popular questions related to "clinical case studies", that are relevant for postdoctoral research fellow specializing in antimicrobial agents & resistance (AMR) at a public university. Give some uncommon strategies that [title] at a [organization], can use to learn more about [topic].
56. Keyword/Topic Research
57. Keyword brainstorming prompts Describe the intent behind each keyword [then upload keyword list] by explaining the searcher's motivations, what they might be hoping to learn, their emotional or informational needs, and the possible context of their search. Describe the key steps [customer type] goes through when interacting with your [product/service] from initial awareness to becoming a loyal customer. For each step, note the customer's emotions, pain points, questions, and goals. Also note opportunities for improving the experience. ***LLMs don’t have access to keyword research, verify with a SEO tool.
58. Understand your Competition
59. Look at search results per keyword Via an SEO tool or incognito, take notes: • What types of content is Google surfacing? (video, image, etc) • How long are the ones that are ranking? • How well do they satisfy Expertise, Experience, Authority and Trust (EEAT)? • How much effort is put into the landing page (a Human Rater eval question)?
60. Formats that Can’t be Read by LLMs
61. Formats that don’t work for AI Search And the other non search engine LLMs can’t read JavaScript.
62. Using AI to Write Online Copy
63. Don’t your SEO work
64. Google’s Stance on AI Assisted Copy
65. Google & AI Generated Content • Google: AI-generated content is acceptable if it’s “original, high-quality, and people first demonstrating qualities of E-E-A-T.” • However, AI can ID a writer’s unique fingerprint. ◦ Planet Word Museum in DC exhibit talks about trained AI on a corpus of language from an author and can ID a pen name. ◦ Google has a 2020 patent where they can identify authors in unlabeled content.
66. Also, Google Profiles Find and update yours here: https://search.audits.com/ https://profile.google.com/@pnasnews
67. Master SEO/LLM Content Brief Prompt
68. SEO Content Brief Prompt You are an SEO content optimization expert. Based on what is ranking for "antimicrobial resistance," look at the top 5 sites ranking for the term, and then create an SEO optimized content outline that highlights the following: • Entities and subtopics needed. • Suggestions for top summary paragraph that is close to the target keyword based on cosine similarity. • Engaging title with the keyword • Introduction that addresses search intent • 5-7 main sections (H2 headings that support the entities needed) • Key points under each section • FAQs based on "People Also Ask" • Meta description (155 chars) • Make it comprehensive enough to satisfy search intent and outrank existing content. [Result] Or this master prompt
69. AI Citations Factors
70. AI Citation Factors “...traditional SEO fundamentals still matter. Crawlability, metadata, internal linking, and backlinks remain essential for ensuring your content is discoverable.” - Microsoft
71. URL Accessibility
72. Is your URL Indexed? Can LLMs see it? • Check indexing via Google Search Console or Bing Webmaster Tools • Check JS visibility with SERPRepCon Chrome Extension.
73. Search Rank
74. ChatGPT - Which Search Engine?
75. Google AI Answers Uses Google
77. Fix Search Engine Warnings From Bing:
78. Fan Out Rank
79. Query Fan out Tools • Chrome Extension - AI Search Fan-out Tracker • https://aicoverage.locomotive.agency/ • https://geo.otterly.ai/geo/ai-query-fan-out/ • Content AI-Mode Readiness • https://queryfanout.ai/ • https://marketbrew.ai/tools-and-guides
80. Query Fan out for ChatGPT Options for seeing queries used in fan out in ChatGPT: • ChatGPT search query extractor (bookmarklet) • Better GPT (Chrome plugin) • ChatGPT Path (Chrome plugin)
81. Query Answer Match
82. Semantic Closeness • Semantic alignment between a user's search query and the text on a webpage or in an AI-generated answer. • This often means that page titles, subheads, and content closely match both the search query and the AI answer.
83. Tools to measure cosine similarity AI Overview visualizer AI Mode visualizer Embedding Similarity Calculator AI Content optimizer - custom GPT To measure Cosine Report SEO custom GPT
84. Market Brew’s Output
85. Measure Semantic Relevance-SERPRepCon Chrome Plugin
86. Intent-Format Match
88. Evaluate Searcher’s Intent 1. Open up incognito, and check to see if Google is serving results like yours. ◦ What type of content is there? Videos? HTML? 2. Prompt LLM with your persona: “If [persona] searched for [keyword] and found this [URL/text] would they find that the copy answered their question? If not, please provide 6 other keyword alternatives for [persona] interested in [jobs to be done].
89. Topic Cluster Ranking
90. Leveraging Topic Clustering Why topic clusters matter: • LLMs only crawl 3 links deep - interlinking encourages more discovery. • AI & Search assess how well your site covers a broad topic comprehensively. Google algo leak & topical authority
91. Why Create Topic Clusters?
92. What Topic Clusters look like SEMRush guide
93. Prompt to Generate Topic Cluster Prompt: “I have a list of prioritized seed keywords related to [Your Topic]. Can you help me cluster these keywords based on their semantic relationships and search intent? The aim is to create comprehensive content that enhances user experience and SERP dominance."
94. Answer Near the Top
95. Be up front - 44% comes from intro
96. How to check your snippet?
97. AI ready structure
98. One Primary Intent. One Entity Cluster
99. Use Declarative Headings. Clear Entities
100. Factually Specific
101. Support Key Passages with Evidence
102. Facts and Statistics • Comprehensive guides with data tables achieve a 67% citation rate • Pages using FAQ schema see a 3.2x advantage in appearing in AI Overviews. • Semantic HTML tables and verifiable author credentials boost citations by 2.3x
103. Explicit Phrasing
104. Be Direct- Channel Simon Cowell “Some people reduce emissions by changing their diet, driving less, buying offsets, or switching energy providers...” Better: Climate change action starts at home: switch to clean electricity, replace gas appliances with electric ones, reduce car trips, and cut food waste.
105. Cites Sources
106. Citations Link to authoritative external sources that support your claims Expert Quotes Include direct quotes from recognized authorities in your field Statistics Present relevant data points from credible research Aim for no more than 5-6 citations per content fragment to maintain focus Credibility is Key
107. Aim for no more than 5-6 citations per content fragment to maintain focus Cite Named Standards and Sources
108. Self Contained Passages
109. Think like Featured Snippets Our writing guide Provide a clear structure with headings, sections, tables.
110. Content Visibility
111. Check for these issues: • Heavy JS or page loads over 500 ms • Don’t block LLM bots via robots.txt or CDN - CrawlerCheck.com • LLMs and search bots don’t click. • Use this free tool to check.
112. Areas of Improvement Avoid these: • Content needs to be visible when the searcher lands on the page. • Overloaded sentences • Decorative symbols: Arrows (→), stars (★★★), or strings of punctuation (!!!) • Unanchored claims
113. Freshness
114. Freshness Even recently created YouTube videos can rank.
115. Brand Entity/Trust
116. Brand Authority Moz’ Brand Authority
117. Length
118. Longer Content is Better It’s best to see how long the ranking URLs are by keyword.
119. Entity Consistency
120. Entity Research • Custom GPT - Entity Salience • Entity Extractor • Word Lift’s Entity Extraction
121. Fixing entities in content Prompt: You’re an SEO strategist focusing on improving content relevance and depth. I’m building a [landing page, blog, product page] targeting [target keyword], and want to include semantically related terms to support ranking. My audience is [brief description of target audience]. Please return: • List of 10-15 semantically related keywords • Any named entities (brands, tools, concepts) relevant to this topic • Suggestions on where/how to incorporate these entities and semantic keywords into the page
122. Structured Data
123. The Value of Schema. • Google's own documentation confirms that JSON-LD structured data is the recommended format for communicating entities, attributes, and relationships to its systems (Google Search Central, 2024). • It turns plain text into structured data that machines can interpret with confidence. -Microsoft "@context": "https://schema.org", "@type": "Person", "name": "Katherine Watier Ong", "image": "https://wostrategies.com/wp-content/uploads/201 7/07/KatherineWatierOng-speakingopt1.jpg", "jobTitle": "Owner of WO Strategies LLC, Writer, Trainer, Podcast Host", "worksFor": { "@type": "Organization", "name": "WO Strategies LLC", "url": "https://wostrategies.com/", "logo": "https://wostrategies.com/wp-content/uploads/2016/09/ WO-LOGO-LowResTagline-Grey.jpg", "email": "katherine@wostrategies.com", "description": "WO Strategies LLC is a boutique organic traffic consultancy specializing in science-based, enterprise sized organizations.", "telephone": "202-930-1744"}, "url": "https://wostrategies.com/", "sameAs": [ "https://www.linkedin.com/in/katherinewatier", "https://www.facebook.com/wostrategies ", "https://in.pinterest.com/kwatier/ ", "https://www.youtube.com/channel/UCQFt3vXfbvQ oRJmfhzSDqAQ ", "https://www.twitter.com/kwatier ", "https://www.crunchbase.com/organization/wo-str ategies "
124. Known Source in the Training Data
125. Your URL is in the training data Sometimes (quite often, in fact) an AI will cite a URL simply because it knows about it from its existing training data. To improve: You start by searching for your “brand + topic.” Record the mentions and then use this prompt: On a scale of 0 (not at all likely) to 100 (almost certain), how likely are the following websites to be used in the training data of LLMs like Gemini, ChatGPT, and Perplexity? [add the list of mentions].
126. Creating a Testing and Optimization Framework
127. Building a Testing Plan • AI Overviews checklist to track your testing plan. • Measure impact on traffic (now and based on predictions) • A/B test your AI-optimized content • Measuring visibility on AI platforms
128. Convert the User! -Engineer Long Clicks • Make sure you’re capturing their email or phone # • Free Microsoft Clarity to test your conversions • Google is watching your users in the Chrome browser (thank you Google algo leak and DOJ trial) - so you should too!
129. Conversion Rate Optimization Tips • Google’s UX Playbook for news and content. • Track your key events in GA4 • Install Microsoft Clarity (Free) • Use LLMs to test your UX. • Use this sheet to track your improvement ideas
130. Organization Adaptability • Join masterminds • Share updates via brown bag sessions • Take advantage of free AI training
131. Measuring your results
132. AI Measurement Framework Success metrics for the AI search era
133. Picking Prompts - 1 (Gemini)
134. Picking prompts - 2 Data sources: • Anonymized Clickstream & Web Panel Data: • Search Engine Results Pages (SERPs): • Public Social Profiles We fed search keywords (that we know audience’s search for in Google/Bing/etc.) alongside other audience data (like the description, websites, etc.) into the AI and requested, via API, a list of likely topics, scored by affinity, and based on the real audience data.
135. Picking Prompts -3, Includes Bots Run RegEx in your Google Search Console to find longer queries: • You can use RegEx to search for query length: 4-8 word queries: ^([^" "]*\s){3,7}[^" "]*$ • 9-12 word queries: ^([^" "]*\s){8,11}[^" "]*$ • 13-20 word queries: ^([^" "]*\s){12,19}[^" "]*$ • 20+ word queries: ^([^" "]*\s){19,}[^" "]*$ ** can include LLM rank tracker queries.
136. Picking Prompts 4- Brand Management Check out the brand prompts here. Samples: What is [brand] ? How would you describe [brand] ? Why is [brand] important? What is [brand] known for? How would you describe [brand] 's main mission or purpose? Who is the primary audience for [brand] ? What products, services, or experiences does [brand] offer?
137. Prompt Tracking So many options! Things to keep in mind: • Tools that only use the API don’t match what users see (UI). (list of which use API here) • Each prompt needs to be run at least 10 times to get statistically significant results. • Check out our list of LLM trackers (this one is free) • You will need SEO software to measure your page in RAG.
138. Microsoft Clarity’s Bot Traffic Reports Free to install. • Question to ask: What pages are *not* getting bot traffic? • Visiting 404s that need redirects? • Bots that need blocking?
139. Resources
140. Advanced Reading & Resources • Beyond Cosine similarity • LLM Security Issues • Dark SEO • Writing for & Ranking in AI Overviews, AI Mode and LLM Search • AI SEO writing checklist
141. Bonus Items • Keyword Content Mapping Sheet • SEO Writing Guide/Content Brief per main topic • SEO Writing Checklist • Searcher’s Intent tutorial and checklist • Featured Snippets writing guide and checklist • Image Optimization Checklist • Fraggles Writing Checklist • Guide for improving CTR • Optimizing the SERP Snippet Generated with Gemini Advanced
142. KATHERINE WATIER ONG 202-930-1744 katherine@wostrategies.com www.linkedin.com/in/katherinewatier www.WOStrategies.com Daily SEO tips via Alexa