49:48
AEO/GEO
AI/LLM search (ChatGPT, Claude, Gemini, Perplexity) runs on fundamentally different rules than traditional Google SEO, and the presenters lay out an evidence-based 'eight AI behaviors' framework (inconsistent answers, skipped middle content, guardrail/alignment-tax penalties, peak-load degradation, instruction-forgetting, per-platform citation logic, and their compounding effects) that businesses must adapt to via AEO/GEO techniques, framed as urgent given Google's incoming Universal Commerce Protocol.
50% of what a site currently ranks for on Google may actually be hurting its LLM listings, while Google impressions rise and clicks fall as AI Overviews answer queries without a click-through.
Multiple competing claims on one page confuse AI prioritization, so it cites neither; the fix is one clear claim per page.
Client case studies show sites can lose backlinks/keywords while gaining overall traffic and AI Overview visibility, because Google pruning non-core pages boosts authority on the pages that matter — 'red' metrics aren't always bad.
AI does not give the same answer twice, even at zero 'temperature,' undermining the idea of a stable AI ranking (tested across ~490,000 prompts).
AI models heavily favor the beginning and end of content and effectively skip the middle (confirmed by a cited Chroma study of 18 models/194,000 test cases), so answers/FAQs should be front-loaded and back-loaded.
Heavy prompt/content guardrails incur an 'alignment tax' — a 5-15% (or up to ~40% combined with long content) drop in AI output quality/findability, especially in regulated niches like health, finance, legal.
Time of day and platform load affect AI answer quality; querying during peak hours (e.g., 2-5pm) can produce noticeably worse, more generic answers.
AI chat conversations 'forget' earlier instructions after roughly five exchanges, requiring workarounds like Claude 'projects' and restarting chats with pasted-in context.
Each AI platform functions like a separate search engine with different citation logic, training-data lag, and authority signals — ranking #1 on Google does not guarantee LLM citation and vice versa.
A live case study restructured a stalled 2019 article and paired it with syndication plus a press release, triggering cross-engine indexing within about a day and reaching page three within days.
SERP composition ('success leaves clues') reveals what content formats — video, images, FAQs, forum threads — an AI/Google rewards for a given keyword, guiding what content to build.
The strategic goal is to 'own' most of page one (and beyond) for a target keyword across multiple content types and platforms, not just rank a single article.
A cited Google patent suggests Google's AI may auto-generate a replacement landing page for a business whose own site isn't adequate.
Google's Universal Commerce Protocol (announced January, rolling out now) will let users complete purchases inside AI chat interfaces, removing merchant control over checkout — framed as urgent, with roughly 8 months to a year before impact.
SEO and GEO can be pursued together in the same article for an estimated 6-8 more months before the transition fully separates them.
EAT (Expertise, Authoritativeness, Trust) — A ranking framework from traditional SEO that the speakers say still applies to what Claude and other LLMs look for when deciding what to cite. Apply: Build genuine expertise, authority, and trust signals into content and authorship rather than relying only on content volume.
AEO (Answer Engine Optimization) — The named practice of optimizing content specifically to be surfaced or cited as an answer by AI systems, positioned as bridging the current SEO-to-AI transition. Apply: Audit which currently-ranking pages are hurting LLM visibility and restructure content so the direct answer isn't buried before republishing.
GEO (Generative Engine Optimization) — A parallel discipline to AEO/SEO for optimizing content toward generative AI search platforms, used throughout the talk alongside AEO. Apply: Pursue SEO and GEO together in the same article during the estimated 6-8 month window before the two fully diverge.
One-thing-per-page rule — The claim that presenting multiple competing viewpoints or claims on a single page confuses an AI's prioritization, causing it to cite neither. Apply: Limit each page to a single clear claim or answer instead of covering several competing options.
Structured data / schema markup — Technical SEO markup that the speakers say 'still works like a charm' for AI visibility, despite claims elsewhere that it's obsolete. Apply: Keep implementing schema markup on pages rather than dropping it based on newer AEO/GEO commentary.
Content restructuring / rewrite-and-resubmit — Reworking existing long-form content that engines have deprioritized so the answer isn't buried, then resubmitting it for re-crawling. Apply: Take an underperforming older article, move the direct answer/FAQ earlier, trim the middle, and republish so engines re-evaluate it.
Baseline AEO test (5x-prompt exercise) — A DIY diagnostic of typing 'who is the best in [your industry]' five times into ChatGPT and screenshotting the results. Apply: Run this test to capture a personal baseline of your brand's current AI-answer consistency before making AEO changes.
Temperature (AI 'creativity' setting) — A model parameter controlling randomness; the speakers' 490,000-prompt test set it to zero to force reliance on training data only. Apply: Recognize that even at temperature zero, outputs vary heavily, so don't treat one AI answer as a stable ranking signal.
Alignment tax — Dan's term for the accuracy/quality cost (roughly 5-15%, or up to ~40% combined with long content) incurred when a prompt or piece of content carries heavy guardrails/restrictions. Apply: Minimize unnecessary restrictive guardrail language in prompts and content, especially in regulated verticals like health, financial, legal, or comparison content.
Multi-tool workflow (Perplexity to Claude) — The presenter's personal practice of using Perplexity for research and handing that research to Claude for creative writing. Apply: Split AI-assisted content production across tools suited to each stage instead of using one model for the entire pipeline.
Reframing claims / claims substantiation — Backing up assertions with evidence rather than making bare claims, framed as improving whether AI surfaces the content. Apply: Rewrite unsupported claims into evidence-backed statements to increase the odds an AI cites the page.
The 'eight AI behaviors' framework — The talk's overarching structure of eight named AI behaviors (inconsistent answers, skipping the middle, guardrail/alignment-tax penalties, peak-load degradation, forgetting instructions mid-conversation, platform-specific citation logic, and their compounding effects) that affect content visibility. Apply: Use the presenters' 'cheat sheet' of all eight rules as a checklist when producing or auditing content for AI visibility.
Claude 'projects' + five-message rule — A workaround for AI 'forgetting' instructions mid-conversation: upload working files into a Claude project so instructions reload fresh, and treat five exchanges as the practical quality ceiling for one chat. Apply: Store reference material in a Claude project, limit sustained work in a single chat to about five exchanges, then start a new chat and paste in prior context to continue.
Press-release distribution technique — A case-study method where publishing a press release linked to restructured content triggered cross-engine indexing and hundreds of backlinks within about a day. Apply: Pair restructured content with syndication and a press release to accelerate indexing and earn 'white hat' citations quickly.
Rapid companion-site building (Gamma / Lovable) — Using no-code site builders to quickly stand up supporting websites/guides that link back to a press release and main content. Apply: Spin up a Gamma or Lovable site as a linked companion asset to reinforce a keyword push.
SERP feature analysis ('success leaves clues') — Inspecting the mix of result types on a search results page (AI Overview, video, Reddit, Shorts, People Also Ask) for a target keyword to infer rewarded content formats. Apply: Check a keyword's SERP composition before creating content and build matching formats, such as adding images, FAQs, or short-form video.
Cross-platform keyword/citation tool (HFS-based) — A custom tool that generates commercial keyword variants for a niche, pulls Ahrefs data, builds platform-specific URLs (Facebook, Reddit, YouTube, LinkedIn, Quora, TikTok), flags SERP features, and recommends content actions, priced from about $129/month. Apply: Enter a niche or keyword to get the platforms and content most likely to get cited, then build competing content there to displace incumbents.
Owning-the-page / omni-channel domination strategy — The goal of occupying roughly seven of ten page-one positions for a keyword across content types rather than optimizing one asset. Apply: Combine articles, press releases, marketplace listings, creator content, and image/video assets around one keyword to crowd out competitors across pages one through three.
Google's AI-generated-website patent — A Google patent, per the speakers, describing that if a business's website isn't adequate, Google's AI will generate a substitute version of that landing page for the user. Apply: Proactively optimize your own product/landing pages now rather than risk Google substituting an AI-generated version outside your control.
Universal Commerce Protocol (UCP) — A Google framework, announced in January and currently rolling out, that lets users browse, select, and purchase products directly inside an AI chat interface without visiting the merchant's website. Apply: Set up merchant center feeds, product feeds, and offers, and build broad omni-channel content presence to prepare for UCP compliance within the next several months to a year.
A report full of declining SEO metrics can represent a net-positive shift toward AI visibility, since Google appears to prune non-core pages while elevating the ones AI actually cites.
Ranking #1 on Google and being cited by an LLM are governed by loosely correlated, separate algorithms, so SEO and AEO/GEO outcomes can diverge sharply on the very same site.
Combining two individually costly behaviors — long content and heavy guardrails — compounds losses rather than adding them, producing up to ~40% AI visibility loss before a human ever reads the page.
A single, well-timed press release plus syndication push can trigger cross-engine indexing within about 24 hours, far faster than typical organic SEO timelines, suggesting a shortcut for reviving stalled content.
The speakers cite a Google patent describing AI auto-generating replacement landing pages for underperforming sites, implying inaction could cede control of a business's presentation to Google itself.
Universal Commerce Protocol is framed as ending merchant control over the purchase/checkout step itself, restructuring e-commerce businesses into product-and-fulfillment roles rather than storefront owners.
«50% of what you are ranking for right now on Google is actually hurting your LLM listings.»
— 03:25
«Multiple competing viewpoints on one page... confuses the AI and it doesn't know which one to prioritize, so it chooses neither.»
— 05:24
«That's the new organic.»
— 07:51
«AI doesn't give the same answer twice.»
— 11:03
«In the middle it's considering a nothing burger.»
— 13:11
«50-page report... it remembers the first 10 and the last 10 or like going to a movie. You remember the intro, you remember the ending. You're not kind of sure what happened in the middle.»
— 13:19
«The more guardrails you add, the more constraints, the tighter it gets, the worse the content gets. The AI is used to doing what it does. It's a creative engine.»
— 17:11
«Dan calls this paying your alignment tax.»
— 17:38
«If I rank well on Google, all the AI platforms are going to pay attention to me. Doesn't work like that.»
— 24:20
«SEO to GEO is not a temporary thing.»
— 28:40
«Ranking number one does not always indicate that you're going to show up in an LLM and vice versa.»
— 31:10
«Success leaves clues.»
— 35:53
«if your website isn't up to snuff, they're going to go ahead and use AI to generate a version of your website and a landing page that they think best serves that user.»
— 44:07
«You no longer control that buying decision once this launches.»
— 46:07
«It's next year. It's not five years from now.»
— 47:10
«let's pretend you are building a house and your competitor's building a house and you want to have a great party. So you have invite about a thousand people to this incredible party. One your competitor gives them a map. You don't. Your competitor gets a thousand people at this great party and you're sitting there with nobody at your house.»
— 49:01
«And this is real. This is going to happen to your listings. So, your competitor, if they're doing it, they're going to get all the attention, the sales, while you just eat their dust.»
— 49:27
Reception
No comments are available to gauge audience reception.
The talk packages a self-branded 'eight AI behaviors' framework with client case studies, screenshots, and cited studies to project a data-driven authority, but most figures (the 490,000-prompt test, internal client reports, the Chroma study) are asserted from the stage without visible sourcing, so it reads as confident conference thought-leadership rather than published research.
49:48