The talk I gave at SEO Week 2026 was called “Watch the Party Die: Engineering Answerability for Prospective Students.” The first slide is a tombstone. R.I.P., the ten blue links, 2000 to 2024.
iPullRank put me on Day 3, the day they themed “The Ecosystem,” and they were kind enough to feature the session in their Day 3 recap and their best quotes roundup. For everyone who asked for the full story, here it is.
SEO Was an All-Out Rager
For two decades, organic search worked like a great party. You knew the address, you knew the dress code, and if you put in the work, you got in. Our traditional approach at Archer Education was built for that world, and it delivered. One university partner earned 70 pieces of coverage with 4.3 million estimated views and $680,000 in publicity value, at an average domain authority of 74, without a dollar of paid spend. Life was good.
Then the Lights Came On
Google AI Overviews and LLMs ended the party as we knew it. A page that holds the exact same ranking year over year can now receive 25% less traffic from Google, because there are simply fewer blue links for a prospective student to click. Meanwhile, 70% of students are using AI to explore their program options and make quicker decisions. We watched a nursing program that once dominated the SERPs for its program keywords see traffic and leads tank. When we audited the build, we found indexability and rendering issues that never mattered in the old world and suddenly mattered a great deal.
There’s a Tim Wu line I used in the talk: “We sometimes treat the information industries as if they were like any other enterprise, but they are not, for their structure determines who gets heard.” The structure changed. Who gets heard changed with it.
Search vs. Fetch: The Diagnostic
The most practical framework in the deck is a simple question: is your problem visibility or comprehension? AI systems retrieve your content in two distinct modes. AI web search, the mode behind Perplexity, ChatGPT Search, and Gemini, still runs on the classic signals: crawlability, domain authority, backlinks, traditional rankings. LLM fetch, the direct URL retrieval that feeds training and responses, cares about different things entirely: clean parse-able HTML, markdown readability, low JavaScript bloat, and a clear heading hierarchy. LLMs don’t render JavaScript the way Googlebot does. If your content only exists after a script runs, to an LLM it may not exist at all.
Here’s the part that surprises people: roughly 80% of AI eligibility is still traditional SEO foundations. The genuinely new work, passage retrieval and source citation, is about 20%. And then query fan-out arrived, where one student question becomes dozens of synthetic sub-queries behind the scenes, and suddenly relevance matters at a resolution keywords were never built for.
Engineering Relevance
We didn’t change the vibe, we evolved it. The talk laid out three levers for closing the answerability gap. First, semantic scoring: use LLM embeddings to measure how closely a page aligns with a student’s actual intent instead of counting keywords. I shared a working Apps Script that scores topic-to-program relevance with cosine similarity, and a simpler test anyone can run: crawl your pages, then ask an LLM what question each page best answers. Second, entity gaps: map which entities, programs, outcomes, and skills your competitors own in LLM responses that you don’t. Ask ChatGPT for the best programs in your category and see whether you’re in the answer. That gap is your roadmap. Third, multimodal content and digital PR built around faculty expertise, because named experts are the strongest authority signal an institution has.
The Student-First Knowledge Graph
The core shift is from keywords to entities. We organize it around four pillars: programs (degree names, formats, credits, tuition, each one an entity, not just a page), outcomes (job titles, salaries, employers, alumni paths), skills (competencies and certifications in the language of job postings, not academic catalogues), and faculty (research, publications, industry experience). We built the ArcherEDU Higher Education Schema, an 85-entity model, to audit against. One MBA program we assessed covered 59% of it, 50 entities out of 85, and had a freshness problem on top: a sitemap with one static date for every page and no date metadata anywhere. LLMs want to know your content is current. Tell them.
The Receipts
This is the part of the talk where the metaphor earns its keep. One regional university went from 3 AI citations to 362 in 19 months, peaking at 194,000 AI impressions a month, with links earned from domains like columbia.edu and ca.gov. A microsite we launched for a partner who couldn’t sustain paid CPLs generated 3,726 leads, 1,789 applications, and 627 enrolled students over 24 months, entirely from organic. Another partner launched six online programs at once; the semantically structured microsite we built produced 314 enrolled students in 20 months. And our Academic Thought Leadership program, which turns faculty into the face of the brand, drove a sharp rise in AI citations for a college of nursing and landed two partners a combined 128 coverage pieces worth 13.5 million estimated views.
We Tested It on Ourselves First
Before we asked any partner to trust this playbook, we ran it on our own sites. We refreshed program content, added structured data, grew inbound links, and built WordPress plugins that automate LLMs.txt and markdown versions of our pages, work I’ve written about before. Organic traffic fluctuated. LLM conversions climbed steadily anyway. We also built Bullseye, our internal research agent, a multi-agent system wired into APIs and MCP tools that interacts with content the way AI systems do, so our strategies come from observation rather than guesswork.
Three Steps to Start the Party
I closed with homework anyone can start this week. One: crawl your five most important pages in Screaming Frog and check whether key content survives the fetch test, cleanly extracted, no JavaScript required. Two: run entity-based queries for your program type and target outcomes, track what shows up, and treat what’s missing as your knowledge graph map. Three: build one thing, a tool or a content asset, that tests your program pages and sharpens your stack.
The party didn’t die. The venue changed, the guest list is curated by machines now, and the promoters who understand the new door policy are doing just fine. If you want the shorter version, my Inside SEO Week interview covers the highlights, and if you want help throwing your own party, you know where to find me.