LLMs don't rank anything. So what are you optimizing for?

When an LLM trains on your content, it doesn't remember where it came from.

That's the gap at the center of most "LLM SEO" advice. The industry is treating LLMs like a new kind of search engine with ranking factors, domain authority, and an algorithm you can reverse engineer.

LLMs don't work that way. There's no ranking inside. They don't remember URLs on their own. They're not doing source quality assessment the way a search engine does. When they do pull in fresh information, they fan out live queries to search engines at runtime, then use this information alongside their training data.

What they carry forward from training data is co-occurrence: which brands and concepts appeared together, how often, in what contexts. The more the industry tries to squeeze LLMs into the old playbook, the further the advice drifts from how the technology actually works.

Britney Muller has spent years close to this technology, including red-team time on BLOOM, the largest open-source multilingual language model before ChatGPT existed. She talks about AI without the hype. She joins Will Critchlow for 45 minutes to work through what that means for the way you think about AI search.

WEBINAR

Free · 45 min · Replay included

Wednesday 29th April • 5 pm BST / 12 pm ET / 9 am PT

Register Now

How LLMs actually work

What happens inside an LLM when you type a prompt, and why "continuation engine" is a more useful framing than "answer machine" for anyone thinking about optimization.

Mentions are the new backlinks

What LLMs carry from training data isn't URLs. It's brand mentions and co-occurrence, which shifts SEO priorities toward a different kind of visibility play.

What "grounding" actually means

A real AI research term borrowed from statistics and cartography. The SEO industry has repurposed it, and Britney explains why that's worth pushing back on.

Why Google is better positioned than people think

Easy to assume Google is on its heels after OpenAI got to chat first. Will argues that underweights their chip-level vertical integration, spam-fighting experience, and direct search attachment.

Who Should Attend?

SVP Digital & Ecommerce
VP of Marketing / Global Head of Marketing
Head of SEO
Director of Ecommerce, Director of Digital Marketing
Head of Product
Anyone responsible for ambitious SEO targets who needs a practical, proven framework

Who's Britney Muller?

Britney Muller has been working in machine learning and generative AI for over ten years. At Hugging Face she led the launch of BLOOM, the largest open-source multilingual language model at the time, with coverage in The New York Times, The Washington Post, and on 60 Minutes.

She's built more than a dozen in-house AI applications for large marketing teams and wrote The Beginner's Guide to LLMs and BERT 101 to make the technology less opaque for non-technical readers. She's the founder of the Orange Labs community, speaks internationally, and has a useful habit of calling out industry claims that don't hold up.

Her work focuses on what works in practice rather than theory, especially for marketers, entrepreneurs, and business leaders who want to use AI without the technical overwhelm.

Who's Will Critchlow?

Will Critchlow is the CEO and co-founder of SearchPilot, the SEO experimentation platform that helps enterprise websites, specifically large ecommerce, take search out of React Mode and put it into Control Mode. After 15 years running Distilled, an agency he built into a global search powerhouse, Will founded SearchPilot in 2020 to give marketing and engineering teams a way to run statistically-rigorous SEO A/B tests across thousands of pages and see uplift in weeks.

Today, the platform powers rapid-fire testing for $100M+ ecommerce, marketplace, and travel brands, safely optimising millions of product pages through controlled testing. Will’s data-driven approach has helped hundreds of companies turn their biggest, least-understood channel into a controllable performance engine with provable ROI.