How AI Decides Which Brands to Surface in Search Results

The brands winning in AI-generated recommendations aren’t the ones spending the most on ads. They’re the ones language models have the most to say about.

Association Strength Is the New Domain Authority

Research on how large language models construct recommendations — covered in depth by Search Engine Journal’s Dixon Jones — points to a concept that most SEO playbooks haven’t caught up to yet: relational knowledge. LLMs don’t evaluate brands the way a search index does. They build probabilistic associations between entities. A brand that appears consistently alongside high-authority topics, credible sources, and relevant industry concepts builds what you might call semantic equity — and that equity directly influences whether the model reaches for your brand name when generating a recommendation.

Think of it less like PageRank and more like reputation in a room full of experts. If every knowledgeable person in that room instinctively connects your brand to a specific problem domain, you get mentioned. If your brand only exists in your own content, you’re essentially unknown to the model, regardless of how well-optimised your website is.

For Southeast Asian brands, this has a specific implication: presence on third-party platforms — Shopee editorial content, LINE Today articles, Grab’s merchant ecosystem — isn’t just distribution. It’s entity-building infrastructure.

Sundar Pichai Just Told You Where Search Is Going

In a recent interview analysed by Search Engine Journal, Google CEO Sundar Pichai described Search’s future as an “agent manager” — a system that completes multi-step tasks on a user’s behalf rather than returning a list of links for the user to navigate themselves. This is not a distant roadmap item. Google’s AI Overviews, now broadly rolled out, are already compressing what used to be a three-click research journey into a single synthesised answer.

The strategic consequence is uncomfortable but clear: if your brand isn’t in the synthesised answer, you’re not in the consideration set. You don’t get a consolation click from position four.

For marketers still measuring success primarily through organic click-through rates, this is the moment to recalibrate. The question is no longer just “do we rank?” but “does the model know we’re authoritative on this topic, and does it have enough relational context to surface us unprompted?” Topical presence — consistent, substantive coverage of a domain — is what earns that unprompted mention.

---

Building Topical Presence That LLMs Actually Register

Topical presence isn’t the same as content volume. Publishing 50 thin blog posts on a keyword theme does less for your entity authority than a smaller body of work that genuinely advances a conversation — and gets picked up, cited, or referenced by other credible sources.

Practically, this means three things for growth teams:

First, treat structured data as entity signalling, not just a technical tick-box. Schema markup for your organisation, your products, your people, and your topic coverage tells crawlers and models how to categorise you. Brands like Grab and Sea Group have dense, well-structured entity footprints across their digital properties — that’s not accidental.

Second, earn mentions in sources the model trusts. Trade press, industry reports, Wikipedia references, and high-authority regional publications carry disproportionate weight. A single well-placed feature in a credible outlet does more for your GEO position than a month of self-published content.

Third, define your topical moat explicitly. What is the specific domain your brand should own in AI-generated responses? Not “digital payments” — too broad. Something closer to “cross-border e-commerce payment solutions for SMEs in Indonesia.” The narrower and more consistently reinforced the association, the stronger the relational signal.

The Compliance Floor Is Rising Too

While GEO strategy dominates the strategic conversation, Google’s concurrent update to its spam policies is a useful reminder that the technical hygiene floor is also moving. As of June 15, 2026, back button hijacking — where sites intercept or manipulate browser navigation to trap users — becomes an explicit violation of Google’s malicious practices policy, per Search Engine Journal’s reporting.

This matters beyond the narrow technical point. Google is systematically closing the gap between what constitutes a bad user experience and what triggers a ranking or indexing penalty. The direction of travel is consistent: practices that degrade user trust — whether that’s manipulative navigation, thin AI-generated content, or opaque redirect chains — are being brought inside the enforcement perimeter.

For teams auditing their technical SEO posture in 2026, back button hijacking is worth a specific check, particularly on campaign landing pages built by third-party vendors or older CMS templates. Two months is enough time to remediate if you start now.

Key Takeaways

• Build your brand’s entity presence across third-party platforms and credible publications — not just your own domain — to accumulate the relational associations LLMs draw on when generating recommendations.
• Redefine your topical focus narrowly and explicitly, then reinforce it consistently across every content surface, so AI systems can categorise your authority with confidence.
• Audit your site for back button hijacking before June 15, 2026 — especially campaign pages and vendor-built templates — to avoid falling inside Google’s expanded spam enforcement perimeter.

The deeper provocation here is one worth sitting with: if the model’s training data is already baked, and your brand didn’t make the cut, how long does it take to rehabilitate that position — and what’s the compounding cost of invisibility in the meantime?

---

At grzzly, we work with growth and marketing teams across Southeast Asia who are navigating exactly this shift — from keyword optimisation to entity authority, from link-building to generative engine positioning. If your brand’s AI discoverability strategy is still a question mark, we should probably talk. Let’s talk