SEO IRL 2026 was one of the most useful SEO events I've attended: practical, packed with original data, and refreshingly free of hype. I left Toronto with pages of notes, a clear plan for AI search, and real optimism about where our industry is heading.
Canada's largest SEO conference brought 250 practitioners to the MaRS Discovery District on October 6–7, 2026, and every seat was sold. Founder Steve Toth and his team put together a focused two-day program where nearly every session ended with something you could try the next morning.
I covered eight of the 14 talks: Michał Suski, Lily Ray, Brie Moreau, Steve Toth, Jabez Reuben, Sacha Fournier, Jess Joyce and Joy Hawkins. Below is what I loved about each one, the data worth bookmarking, and the ideas I'm putting to work right away.
In this review:
- Five big ideas I left with
- Michał Suski (Surfer)
- Lily Ray (Algorythmic)
- Brie Moreau (WLDM)
- Steve Toth (SEO Notebook)
- Jabez Reuben (The Blueprints)
- Sacha Fournier (JournoFinder)
- Jess Joyce (Growth Crew)
- Joy Hawkins (Sterling Sky)
- What I'm putting into practice
- Final thoughts
Five big ideas I left with
What impressed me most was how well the talks connected. Speakers with very different specialties kept arriving at the same encouraging conclusions.
- Good SEO is still the foundation of AI search. That's great news: the skills we've built keep paying off. 95% of sources cited by AI engines rank in Google's top 10 for at least one keyword (Suski), and Lily Ray showed Google traffic and AI citations moving together.
- Better measurement is within reach. Track the consumer apps rather than the API, which shares only 26% of brands with them, and run each prompt about 15 times (Suski). Moreau and Hawkins applied the same discipline to citations and local results.
- Brand mentions and consensus are the new links. Domain authority shows no positive correlation with being cited (Suski). Brand mentions outweighed volume links 3.0x to 0.6x (Moreau), and "consensus > links" was Jabez Reuben's core rule.
- Quality settles the listicle question. Self-promoting listicles still make up 20–35% of first-position sources (Suski), but sites that leaned on them too hard were hit in Google's January 20, 2026 update (Ray). Balanced, well-sourced "best for" content is the version that lasts (Reuben).
- There are more ways than ever to earn visibility. 27% of links in AI answers go to news articles (Fournier). Reviews, awards and smart distribution all help models find and trust you (Hawkins, Joyce).
Michał Suski (Surfer): the clearest data on what AI actually cites
Suski's talk was a perfect opener for anyone tired of guesswork. Surfer's research covers millions of AI citations, and the conclusion is reassuring: classic SEO still does most of the work, and the rest is learnable.
Track what real users see. Surfer sent 1,000 prompts to five models twice on the same day, once through the API and once through the consumer app. The two shared only 26% of brands and 15% of cited publishers, so switching to app-based tracking is a quick win for reporting accuracy.
Spend less on prompts, more on repeat runs. Across 3.8M prompt-response pairs, rephrased prompts returned the same brands 78% of the time. That frees up budget to run your core prompts more often instead.
Rankings open the door. 95% of cited sources rank top 10 for at least one keyword. A URL ranking for none of a query's fan-out queries made the AI Overview 9% of the time; ranking for nine raised that to 57%.
Focus where it counts. Across 5M citation sources, domain authority correlated between −0.1 and 0 with being cited, and link counts sat at 0.03–0.05. Of 20 page factors tested, the clear winners were about answering the query well:
Tier Page factors Avg. lift in citation frequency Top Search intent alignment, early query confirmation, early query answer ~20% Middle Heading structure, clear definitions, information density, semantic triplets, experience signals ~10–13% Bottom Author boxes, FAQ sections, expert quotes, tables of contents, statistics ~4–6%
Win your core sources. A small set of pages gets cited run after run, and that set settles after about 15 runs per prompt. Core sources scored a fact coverage ratio of 0.42, against 0.23 for pages never cited, and close to 30% of citations come from a page's opening few dozen words.
Mentions drive visibility. A brand named first in an AI answer appears in 34% of its sources, against 11% for the brand named 20th. Even as ChatGPT cut its sources per prompt from about 31 to 14–16 in 2026, only 1 in 5 listicles dropped out.
My favorite takeaway: "Do you know your niche's core sources? If not, everything else is guesswork." It turns a fuzzy goal like "get into AI answers" into a concrete list of pages to win.
Lily Ray (Algorythmic): why the long game always wins
Ray's session was the clear-headed perspective every marketing leader should hear. Her message is good news for anyone doing quality work: AI engines retrieve from search indexes, so good SEO is good GEO.
SEO and AI visibility rise and fall together. She tracked 11 sites from January 20 to February 18, 2026. The sites that lost Google traffic (−26.7% on average) also lost AI citations (−22.5%), which shows how tightly the two channels are linked.
Google says it plainly. Its guide to generative AI features calls optimizing for them "still SEO," and Brendon Kraham's June 2026 post is titled "Good SEO is good GEO."
She mapped which shortcuts are getting caught, which makes it easy to steer clear:
- Self-promotional listicles at scale. Sites running anywhere from 10 to 2,030 of them lost traffic in the January 20, 2026 update. Being cited isn't the same as being recommended, either.
- Scaled AI content. Of 237+ public case studies from 15 AI content vendors, 65% lost 30% or more of their peak traffic.
- Gaming mentions and assistants. Paid and reciprocal mentions, "Summarize with AI" prompt injection and Reddit "GEO experiments" are now treated as spam.
- Programmatic and parasite pages. The August 2026 spam update targeted them, and Bing removed a 90K-URL site that had been surging in ChatGPT.
She also showed where AI engines are heading. Peec AI data from August 2026 showed ChatGPT's fan-out queries using "pricing" 185% more and "official" 165% more. That's an opportunity for brands with strong official sites, clear pricing pages and fresh content.
What works: improve what you already have before you scale, show real expertise everywhere, keep your brand story consistent across AI answers, and connect your SEO data in one "AI brain" via MCP.
My favorite takeaway: her rule of moderation, borrowed from a NYC fitness trainer: "You can have one slice of pizza, but not the whole pie." One or two bold pages are fine; thousands are a liability.
Brie Moreau (WLDM): a masterclass in entities and lasting citations
Moreau's session was one of the most technical I saw, and one of the most rewarding. She showed how AI engines understand the web through entities, embeddings and co-citation, then gave a precise workflow for earning citations that stick.
Wikidata is "Wikipedia for LLMs." Each meaning of a word gets its own ID: Jaguar Cars is Q30055, the animal is Q35694. That ID tells a machine which Jaguar a page means, and the opportunity is wide open: only 6 of 135 beauty school sites in one WLDM study were in Wikidata.
Doing schema well is a real advantage. WLDM's Common Crawl extraction matched 905 of the 958 types in Google's own schema dataset (94.5%). Most sites paste the same boilerplate everywhere; one audited site had schema on 99.5% of its pages but only 21 unique types and zero rich results.
Google sees embeddings, not DR. Search systems judge a page by where it sits in semantic space and what it sits near. In third-party data compiled for the talk, brand mentions weighed 3.0x, editorial links 1.8x, schema 1.4x and volume links 0.6x.
Smarter links, not more links. The best backlinks come from pages that rank top 20 for your keyword or related terms and get traffic of their own.
Test whether a citation is sticky. WLDM tracks each cited source daily for 41 days:
Label Rule Sticky Cited on 8+ of 41 days and in today's live ChatGPT answer Consistent Cited on 8+ days, but not in today's answer Live only In today's answer, but cited on fewer than 8 days
The workflow: export every cited URL from your AI visibility tool, then keep the sources that are sticky and central to your prompt cluster. On those pages, add one verified, branded statement backed by two credible sources. Where coverage is missing, publish evidence-based content on the most central sites.
My favorite takeaway: the stickiness test. It's a simple, repeatable way to separate one-off noise from the citations worth your time and budget.
Steve Toth (SEO Notebook): AI search is a discovery call
The conference founder delivered one of the most eye-opening talks of the two days. His insight: AI engines qualify the buyer, and the follow-up turns of those conversations are already sitting in Search Console, ready to analyze.
Engines end answers with a question. In the employer-of-record niche, 97.8% of Google AI Mode answers ended by asking about company size, location or legal entity. Perplexity did it 100% of the time, ChatGPT 80% and Gemini 57.8%. "These engines are not answering questions. They are running a discovery call."
The follow-ups show up in Search Console. Replies like "yes, go on" and "yes, pricing" appear as queries, and John Mueller confirmed that AI Mode follow-ups are counted there. On August 31, 2026, all five turns of one conversation landed on the same product page, and the final buying question got the click.
He was open about the limits. 46.77% of Search Console clicks are anonymized (Ahrefs, 22B clicks), and every conversational turn is long tail by definition. Bigger sites will find more journeys.
Map every branch. The chips under an AI Mode answer split buyers into segments, and each branch recommends different brands. Duplicate the browser tab once per chip, follow one branch per tab, and you end up with a tree of buyer profiles and the brands each one sees. It's a brilliantly simple technique.
He shared a full workflow, with free tools:
- Find the conversational turns with the free stevetoth.ai Search MCP and his Follow-Up Journey skill.
- Record HAR files of the same journey across different LLMs.
- Clean personal data out of them locally.
- Parse what was searched, fetched, cited and dropped.
- Optimize the pages that weren't cited and the ones that were.
My favorite takeaway: "An AI answer is the average of everything it has read. We never accepted average advice in SEO, and we should not accept it in AI search."
Jabez Reuben (The Blueprints): a step-by-step playbook for LLM visibility
Reuben's "LLM Domination" session was the most hands-on playbook of the event. His core idea: when enough credible pages agree about a brand, LLMs repeat it. So study what each model searches and cites, then publish balanced "best for" content where those models read.
LLM search is a valuable new channel. Queries run about 17x longer, about 70% are unique, and context carries across turns. Best of all, LLM traffic converts 4.4x to 23x better than organic.
Each engine retrieves differently. Per the studies he cited, ChatGPT fans out into 10–18 sub-queries via Bing and Bright Data, Claude runs 1–4 through Brave Search, and Perplexity runs none.
Research the answer before you publish. For each target prompt, he logs five things: fan-out queries, reasoning, winners, "best for" categories and sources. For "best red light therapy for shoulder pain," ChatGPT checked site:fda.gov and systematic reviews before it searched a single product, which tells you exactly what your content should cover.
The opportunity is real. For "best standing desk australia," one Medium article earned Desky Dual the AI Overview's "best for cable management" slot. Models split every market into categories, and many of those wins are still up for grabs.
His content formula works for listicles, comparisons, reviews, alternatives pages and press releases. It calls for a clear answer in the first 30%, a TL;DR, tables, balanced coverage of every competitor, all brands in H2s, backed-up claims and semantic triples. A final footprint check keeps placements from looking templated, which LLMs now notice.
Placement beats prestige. Pick sites that don't block LLM crawlers and that have AI citations and real traffic. Medium, LinkedIn, Reddit, YouTube and Substack all work, and even YouTube chapters can carry "best for" labels.
He built ethics into the method. Follow the FTC's Consumer Review Rule (16 CFR Part 465), avoid "tested and tried" claims you can't back up, and add disclosures.
My favorite takeaway: specificity wins. You don't need to be "the best" overall, just clearly the best for someone.
Sacha Fournier (JournoFinder): reactive PR made simple
Fournier's talk was one of the most energizing of the conference. He showed SEOs a fresh, practical way to earn news coverage, and news coverage feeds AI answers directly.
News is a fast lane into AI answers. 27% of the links in ChatGPT, Claude and Gemini answers go to news articles, according to a May 2026 study of 25M+ links across 17 industries.
Get ahead of the crowd. Journalist requests grew slowly to about 10K a month, while replies climbed to 60–65K a month in 2025. By the time a request is posted, the story has already broken, trended and been planned. Reactive PR lets you step in at the start instead.
One pitch, many journalists. His Eggspensive project shows how fast it can work:
- US egg prices hit a record $4.82 a dozen in February 2023.
- In three days he built a live, embeddable map of egg prices at every Walmart in the US.
- A four-line email went to about 50 reporters already covering egg prices.
- Within 48 hours came TV news, The Takeout, Hacker News, local TV and 11 more pieces of coverage.
Let AI watch the news for you. JournoFinder's free API groups 150,000+ stories into topics every hour. A setup prompt turns ChatGPT or Claude into a daily reactive PR alert, and nothing is drafted or sent without your approval.
A simple filter for what's worth it: the story is fresh, your expert would talk about it on live TV, and your expert can add something.
My favorite takeaway: the session ended with a working alert you can set up in about 10 minutes. It's rare to leave a talk with a tool that's ready to use the same day.
Jess Joyce (Growth Crew): understanding the pipes behind AI search
Joyce gave the most thought-provoking talk I saw. She explained, with great data, how content actually reaches a model, and why your distribution choices now matter more than ever.
The traffic picture, made clear. Google sends 276 clicks per 1,000 searches, down from 374 in 2024 (SparkToro). ChatGPT is the world's #5 website, yet it sent 4.3M visits to the sites Ahrefs tracked in September 2026, against Google's 301.8M.
Know where the open web ends. Facebook, TikTok, LinkedIn, Instagram, Reddit and X each had zero user-content pages in Common Crawl in August 2026. Reddit still reaches models through licensing deals: "The content didn't disappear. The free route to it closed."
A page reaches AI through four separate pipes:
Pipe What it is Where you can check it Fetching Bots requesting your page Your server logs Indexing Your page stored in a search index Search Console, Bing Webmaster Tools Retrieval Your page pulled to ground a live answer Bing Webmaster Tools grounding queries Training Your page in a model's training set Common Crawl only, in part
Take control of your robots.txt. By Cloudflare's count, 9.3% of AI bot requests are search and 51.8% are training. Allow Googlebot, Bingbot and the answer-time agents, then decide on purpose about GPTBot, ClaudeBot and CCBot. "Four decisions. Most robots.txt files contain none of them."
Give your best work a second life. A link from a crawled page counts, and a YouTube description reaches every model. So re-promote the strong content that went out once in 2023 and was never mentioned again.
My favorite takeaway: much of the best work still comes from small Slack and Facebook communities. "You cannot optimise any of it. You can only be there." It was a warm reminder that relationships remain the heart of this business.
Joy Hawkins (Sterling Sky): winning the new local SERP
Hawkins gave a wonderfully actionable local SEO session, grounded in Sterling Sky's client data. Her guiding principle set the tone: "The businesses I want to help win are the ones customers should choose."
Meet the AI local pack. In early 2026 Google began replacing the local pack with an AI version, mobile-only and US-only, on 13% of queries. In 88% of 322 markets it showed fewer unique businesses, which rewards the businesses that earn their spot.
Every local feature competes for attention. Between April 6 and June 1, 2026, 3-pack ads jumped from 26 to 158 appearances, so a full local strategy now looks at organic, AI and paid results together.
Reviews still lead: frequency first, then volume, then rating. One business had a 4.8 rating but 54% fewer total reviews than its top five competitors and 26 days since its last one. The fix is clear and doable: ask for reviews steadily.
AI Overviews love awards. Descriptions read like "a four-time 'Best of Portland' winner" or "'Best Law Firm' by U.S. News & World Report." Find the awards you don't have yet, and promote the ones you do everywhere.
Use one name everywhere. A Las Vegas firm called itself "The Abrams and Mayo Law Firm" on its website and "The Abrams Law Firm LLC" on its Business Profile. Matching them helps AI connect the two.
Track with care. Run each keyword dozens of times, take screenshots and include ads, because AI local results vary from run to run.
LLM leads are growing. In one data set, LLMs drove 81 of about 4,100 leads (2%), with ChatGPT leads jumping to 31 in July 2026. Pairing SEO with paid search pays off too: one client went from 51 monthly leads in April 2025 to 639 in July 2026 after ads were added.
My favorite takeaway: AI can only repeat the awards it can find. That one idea gives every local business an easy, positive project to start this week.
What I'm putting into practice
The best compliment I can give a conference is a to-do list. Here are the ten ideas I'm most excited to act on, in roughly the order I'd tackle them.
- Upgrade AI tracking. Measure the consumer apps, not the API, and run each prompt 15+ times before trusting the results.
- List our niche's core sources. Keep the citations that persist across runs and days.
- Earn mentions on those pages. Add one verified, branded fact backed by two credible sources, and aim for balanced "best for" placements.
- Rewrite key pages answer-first. Confirm and answer the query in the opening lines, and pack in useful facts.
- Keep quality high. Review self-promotional listicles and scaled pages: one slice, not the whole pie.
- Mine Search Console for follow-up turns. Rebuild buyer journeys and walk every AI Mode branch.
- Strengthen entity basics. Claim a Wikidata entry, use schema that describes what you are, and keep one business name everywhere.
- Make robots.txt choices on purpose. Search crawlers, answer-time agents and training bots are separate decisions.
- Set up a reactive PR alert. Offer your expert to every reporter on a fresh story.
- Grow local the proven way. Collect reviews steadily, promote your awards, and pair local SEO with paid search.
Final thoughts: highly recommended
SEO IRL 2026 is an easy recommendation. The talks ran on real data, the speakers shared their methods and tools openly, and almost every session ended with something practical to try.
I especially appreciated how honestly the program handled open questions. Suski's data shows self-promoting listicles still earn citations, while Ray showed the risks of overdoing them. Hearing both sides made the advice more useful, and it's the kind of nuance you rarely get from a single blog post.
If you're an in-house SEO lead, an agency owner shaping your AI search offer, or a local SEO, put SEO IRL on your calendar for next year. Seats are capped at 250 and 2026 sold out, so join the waitlist early at seoirl.com.
A big thank-you to Steve Toth, the speakers and everyone who made SEO IRL 2026 such a valuable two days. I'm already looking forward to the next one.
Sources
- My conference notes and slides at SEO IRL 2026
- SEO IRL official site: dates, venue, 250-seat cap, 14 speakers