How Google's Open Knowledge Format, AI citation patterns, and spam detection are quietly redrawing the rules of generative engine optimisation in Southeast Asia.
The rules of search discoverability are being rewritten — not by an algorithm update you’ll read about in a press release, but by the quiet structural shift happening inside how machines understand content rather than merely index it.
Three developments landed within 24 hours last week that, taken together, tell a coherent and urgent story for any brand that cares about being found in LLM-powered search environments. Here’s what they mean and what to do about them.
Google’s Knowledge Graph Just Got a Front Door for Websites
Search Engine Journal’s Slobodan Manic surfaced something that deserves more attention than it’s received: Google’s Open Knowledge Format (OKF) is essentially a specification for turning your website’s content into a traversable, linked semantic graph — one that AI agents can navigate with intent rather than just crawl for keywords.
The strategic implication is significant. Traditional SEO treated pages as discrete documents competing for ranking positions. OKF treats your entire site as a knowledge network where entities — your brand, your products, your subject-matter claims — exist in relationship to each other and to the broader web of established knowledge.
For brands in Southeast Asia, this matters acutely. If your site discusses, say, embedded finance for SMEs in Indonesia, OKF allows you to formally declare that your content entity “embedded lending” is related to Bank Indonesia’s regulatory framework, to specific fintech operators, and to documented market size figures. An AI agent traversing that graph doesn’t guess at relevance — it follows declared semantic relationships. Brands that structure their content this way become legible to generative engines. Those that don’t remain opaque.
Implementation starting point: audit your existing schema markup, identify your top 10–15 core topic entities, and map their relationships explicitly using structured data. This is less a technical sprint than an editorial architecture decision.
The AI Citation Game Has a Ceiling — and a Trap
Ahrefs published findings from two recent studies, including Glen Allsopp’s analysis of 750 ChatGPT prompts, that confirm what many GEO practitioners suspected: “best [category]” listicle-style content is the most frequently cited source type in AI-generated answers. Self-promotional content does get picked up by large language models.
But here’s the trap the same research exposes: self-promotional content that lacks independent corroboration creates a brittle citation profile. AI systems are increasingly cross-referencing claims against third-party sources. Content that appears only in brand-owned channels — with no external validation, no mention in independent publications, no entity associations in the knowledge graph — gets cited initially, then deprioritised as models refine their source weighting toward authoritative corroboration.
The practical read: publishing a “Best [Category] Tools in Southeast Asia” listicle that features your own product will earn you short-term AI visibility. But if that claim isn’t reflected in independent reviews on G2, regional tech publications like Tech in Asia, or analyst commentary, the citation half-life is short.
The GEO strategy that holds: earn mentions in third-party content that AI models treat as authoritative, then structurally connect those mentions back to your entity graph through consistent naming conventions and cross-domain schema.
Google Is Getting Sophisticated About What Spam Looks Like at Scale
Semrush reported on new Google research that detects AI-generated video spam not by analysing individual videos but by clustering accounts and identifying behavioural patterns across networks. A single AI-generated video might pass content quality checks. A network of accounts exhibiting coordinated posting cadence, similar semantic fingerprints, and shared entity associations does not.
This signals something broader than video spam detection. Google is moving toward pattern-level quality assessment across content formats — which means the GEO risk isn’t just publishing thin AI content on your own site. It’s participating in content ecosystems that look coordinated, inauthentic, or semantically hollow at the network level.
For brands running influencer programmes, content partnerships, or affiliate SEO in markets like Thailand, Vietnam, or the Philippines — where coordinated content distribution is common — this is worth taking seriously. If your brand’s entity appears in a cluster of sites that exhibit spam-pattern characteristics, the association itself may carry reputational cost inside Google’s entity evaluation layer.
The mitigation isn’t to avoid content partnerships. It’s to be selective about the entity neighbourhoods your brand inhabits, and to ensure that external content mentioning your brand meets a genuine quality threshold — not just a word-count or keyword-density threshold.
What’s Actually Being Optimised For Now
Taken together, these three developments describe a search environment where the unit of competition is no longer the individual page — it’s the entity and its semantic neighbourhood. Who mentions you, in what context, with what structured relationships declared, and whether those mentions form a coherent and independently verifiable picture of your brand’s authority in a defined domain.
This is a fundamentally different optimisation problem than traditional SEO. It’s closer to reputation architecture than content production. The brands that will win AI-driven discoverability in Southeast Asia’s increasingly LLM-mediated search landscape are those that treat their knowledge graph presence as a strategic asset — not an afterthought managed by a developer once a year.
The question worth sitting with: if an AI agent were to traverse the entity graph surrounding your brand right now, what relationships would it find — and would they be the ones you’d choose?
At grzzly, we spend a lot of time thinking about exactly this layer of search — the structural, semantic foundations that determine whether a brand gets cited, trusted, or ignored by generative engines. If you’re working through what GEO actually looks like in practice for a Southeast Asian brand, we’d enjoy that conversation. Let’s talk
Sources
Written by
Sneaky GrizzlyTracking the quiet revolution inside LLM-powered search — where brand mentions, structured semantics, and entity authority rewrite the rules of discoverability before most marketers notice.