I Spent Two Days at SEO IRL 2026. Here Are the 5 Things I Learned.

The gap between conference rooms and LinkedIn feeds has never been wider. I wrote that after day one in Toronto, and it held through day two.

I was at MaRS Discovery District on October 6 and 7 with the First Rank team, in rooms with 14 speakers, about half working directly in my lane: how AI models pick their sources. Some attendees came away disappointed (the feedback threads collect the gripes), fair enough if you wanted button-by-button tutorials. If you wanted data on where AI search is going, the room delivered. These are the five things I wrote down.

1. Cosine similarity is the new link metric

Brie Moreau brought the largest dataset I saw at the conference: more than 300,000 AI citations analyzed. His thesis is that inside LLMs, link value is decided by cosine similarity, page-level traffic, and topical clustering. Models cite pages that sit semantically close together in topically dense clusters. The traditional link metrics your reports are full of are dead weight inside a language model.

He has worked the idea in public too: at SEO Week in April he showed how cosine similarity and page-level traffic decide citation value, and why that breaks traditional link building practices (the full talk is on YouTube).

Independent data points the same way: Hallam's roundup (Ahrefs, SEMrush, Profound, Zyppy; 75,000 brands) found backlinks correlated with AI visibility at 0.218, branded web mentions at 0.66 to 0.71, YouTube mentions at 0.737 (Hallam). Correlation is not causation, but the direction holds across datasets, and it matches what Moreau measured.

My read for Monday: evaluate your citation profile by cluster membership and page-level traffic, not domain-authority scores.

2. Consensus beats links

Jabez Reuben, founder of The Blueprints, built his talk around one idea: LLMs cite what independent sources agree on. Consensus means topical authority about your brand on third-party sites. The model trusts you when it sees you everywhere, the same way a person would.

He laid out the mechanics in an earlier interview: a Google top-10 or top-20 ranking substantially raises your odds with Claude, Perplexity, and Gemini; citation weight leans on search results first, Reddit second; and third-party content shapes answers even when never cited (Jesper Nissen's interview). He starts client work at five supporting posts per keyword, because LLMs run varied fan-out queries before answering.

The practical shift: move off-page budget from link acquisition to consensus building. Unlinked brand mentions, expert commentary on third-party sites, content answering the fan-out queries your money pages miss. Links still matter for Google. Mentions matter for the models.

3. Self-promotional listicles are an unreliable asset inside the answer

Lily Ray closed the conference on Wednesday with fresh numbers from her October 4 Substack essay: across 100 business software queries, brands that ranked themselves first on their own "best of" pages were left out of the recommendation in 83% of the Google AI answers citing those pages. Versus June, AI Overviews cite 38% fewer vendor-written pages. In 70% of cases, the brand's listicle helped a competitor get recommended while the brand was left out (ppc.land's summary).

She included the nuance: Kevin Indig found 62% of vendor-authored list pages gained organic traffic since January (median +38%) across 5.32 million organic rows, and Scrunch found a cited self-ranking page lifts its author's recommendation rate from ~4% to ~7% while a competitor gets recommended in 24.3% of those answers. The read: self-serving listicles are not dead, but they are an unreliable asset and often an own-goal.

Google warned us in January: self-promotional listicle sites lost 30 to 50% of organic visibility within weeks (Search Engine Land). The crackdown moved from rankings to answers. If your team still publishes "best X" pages where you rank first, audit them this quarter.

4. "Truth, Not Traffic" is the new reporting brief

Steve Toth runs SEO IRL, and his directive was blunt: Truth, Not Traffic. Agency reporting moves from rank positions to citation visibility and answer accuracy: when the model answers your buyer's question, are you in the answer, is the claim about you right, and did it link you.

I wrote about the measurement problem earlier this week: one-shot AI visibility dashboards lie because personalization changes the answer (AI Visibility Scores Are Lying to You). Toth's framing gives the replacement. What I am changing Monday at First Rank: citation-visibility tracking, answer-accuracy checks, and branded-query AI Overview presence as standing line items. Rank tracking stays. It stops being the headline.

5. The fundamentals did not die. They got harder.

That was my day-one line and the through-line of the whole conference. E-E-A-T, structured content, and real brand presence are what show up in AI results. The feed still sells shortcuts. The room has receipts on what survives them.

Gary Illyes said it at Search Central Live in Barcelona: scaled content is becoming a bigger problem than link spam. Google filters 40 billion spam pages a day, and core updates judge content at page level. Then Google rewrote its documentation on October 1 and 2: accuracy is now "critical," raters look for effort, originality, talent or skill, and accuracy, and fabricating creator profiles is "a form of deception" (the ppc.land timeline tracks the paper trail). I ran the byline audit on my own site a few days ago and fixed it (my write-up).

What I am changing on Monday

Conferences only earn the ticket if something changes when you land:

First, per-engine citation tracking already runs for my clients: ten money questions, four engines, logged in and out. This month I am adding Reuben's consensus layer on top: third-party brand mentions tracked like rankings (my source-diet breakdown).

Second, every client with "best X" content gets a listicle audit against Ray's numbers. Self-ranking pages get rewritten, retired, or replaced with genuinely independent coverage.

Third, off-page budgets shift toward consensus: expert commentary, original data, and third-party pages about our clients on domains the engines already cite.

Fourth, reporting at First Rank adds citation visibility and answer accuracy as standing items. Rankings were a proxy for being seen. Citations are the position now.

Two days and one notebook page produced four changes, and the rooms were worth it.