One of the most counterintuitive findings from recent research is this: brand-owned websites typically contribute only about 5-10% of the sources AI models use to generate answers. The rest comes from third-party content, i.e. product reviews, retailer listings, forum discussions, community posts, press coverage.

This flips the traditional content marketing playbook. It's no longer just about what you publish on your own domain. It's about the entire ecosystem of information that exists about your brand across the internet.

Think of it this way: when someone asks an AI "What's the best project management tool for remote teams?", the model isn't just scanning your homepage. It's synthesising information from software review sites, Reddit threads, industry comparisons, and expert opinions. Your owned content is one small input in a much larger picture.

A Practical Playbook for Getting Started

Based on McKinsey's framework and what's emerging from practitioners, here's a sensible approach:

Start with a diagnostic. This is the simplest and most universally recommended first step. Open ChatGPT, Gemini, Perplexity, and Claude. Ask the kinds of questions your actual buyers ask. Document whether your brand appears, how it's described, and whether the information is accurate. You might be surprised: one brand (Pernod Ricard, working with agency Jellyfish) discovered that an AI model was miscategorising one of their mass-market scotch brands as a prestige product. You can't fix what you can't see.

Broaden beyond your own pages. Since AI models lean heavily on third-party and earned media, invest in the sources that feed them: honest customer reviews, partner content, PR and media mentions, presence in authoritative directories, and activity in communities where your audience gathers (including Reddit, which AI models reference frequently).

Make your content machine-extractable. This means structured headings, FAQ sections, concise TL;DR summaries, schema markup, and what researchers call "high factual density", specific, sourced statistics rather than vague marketing language. The Princeton study found that adding concrete data points was among the single most effective tactics for boosting AI visibility.

Build new measurement habits. Track how often your brand is cited across AI platforms, the sentiment of those citations, and whether AI-referred traffic is growing in your analytics. Tools like Jellyfish's "Share of Model" metric, measuring how often and how favourably LLMs mention a brand compared to competitors, are emerging as new standards.

The Window Is Small, but It's Open

Here's what makes this moment interesting: the competitive landscape in AI search is still forming. Unlike traditional search, where dominant players have had decades to build authority, AI answers create a more level playing field, at least for now. Smaller brands with well-structured, factual, widely referenced content can outperform much larger competitors who haven't adapted.

But that window won't stay open forever. As more brands invest in GEO, the bar will rise.

The good news is that the fundamentals aren't mysterious. Be accurate, be specific, be cited by others, and make it easy for machines to understand what you offer. That's not a radical departure from good marketing. It's rather good marketing, optimised for a new kind of audience.