GEO Tactics, Agentic Commerce, and the Signals That Actually Matter

The search industry has a recurring habit: when a new surface emerges — featured snippets, voice search, AI Overviews — someone invents a new file convention, sells it as the unlock, and the rest of us spend six months debating whether it works. In 2026, that file is llms.txt. Spoiler: the evidence says it probably isn’t doing what you think.

llms.txt Is SEO Astrology — and the Proof Is a Cat File

Mark Williams-Cook at Search Engine Journal ran a genuinely useful test: he took the four core arguments used to validate llms.txt as a GEO signal and applied them verbatim to a text file about cats. Every argument held. That’s not a minor methodological quibble — it’s a falsifiability problem. If the same logic that justifies your GEO tactic also justifies a file about tabby cats, you don’t have evidence. You have a hypothesis dressed in cargo-cult clothing.

For brands in Southeast Asia, this matters more than it might in markets where GEO is still theoretical. AI-generated answers are already appearing in Google Search for Bahasa Indonesia, Thai, and Tagalog queries. Marketing teams are being asked to “optimise for AI.” The llms.txt pitch lands well in that environment. But deploying tactics without a causal mechanism isn’t a strategy — it’s a ritual. Proximity to the right answer still requires you to know what the right answer looks like.

Context Engineering Is the Actual Lever

Google’s former Chief Scientist Jeff Dean recently made an observation that should recalibrate where GEO effort goes: the model matters less than the context you give it. As Search Engine Journal reports, Dean’s framing places context engineering — how information is structured, surfaced, and made retrievable — above model capability as the determinant of output quality.

For search practitioners, this is less revolutionary than it sounds and more actionable than most GEO advice. If LLMs are retrieval systems operating on context windows, then your job is to make your brand’s information the highest-quality, most unambiguous input in that window. That means:

• Structured, atomic content — answers written as discrete, self-contained facts, not buried in 800-word introductory paragraphs
• Entity clarity — your brand, products, and locations consistently named and described the same way across every crawlable surface
• Authoritative attribution — content that is cited, linked, and associated with named experts, not anonymous brand voice

For local and hyperlocal SEO, this is already familiar territory. Google Business Profile consistency, NAP accuracy, and category specificity are all forms of context engineering. The difference now is that the same discipline needs to extend into your CMS, your product feed, and your schema markup.

---

Agentic Commerce Is Changing What “Optimised” Means

Moz’s Miracle Inameti-Archibong laid out something that every e-commerce team in Southeast Asia should read before their next product catalogue review: AI agents are increasingly acting as purchasing intermediaries. They don’t browse. They query, evaluate structured outputs, and transact — often without a human reviewing the product page at all.

The implications for optimisation are significant. Inameti-Archibong highlights four surfaces that agentic systems actually consume: Markdown-formatted content, product feeds, structured data (schema), and MCP (Model Context Protocol). Of these, product feeds and schema are the most immediately actionable for teams already running on Shopee, Lazada, or regional e-commerce infrastructure.

Specific steps that matter now:

1. Clean product feeds with explicit attribute fields — size, colour, material, compatibility — not inferred from titles
2. Product schema with offers, aggregateRating, and availability properties populated — agents weight completeness
3. Markdown landing pages for high-consideration categories — structured headings, comparison tables, and clear CTAs that agents can parse without rendering JavaScript

For brands operating across multiple Southeast Asian markets, the multilingual dimension adds a layer: your structured data needs to be language-specific, not just translated. An agent querying in Thai should retrieve Thai-language schema, not English metadata with a translated product title bolted on.

What Local SEO Teams Should Actually Do This Quarter

Pull these three threads together and a clear local search priority list emerges — one that doesn’t require betting on file conventions that haven’t passed a basic falsifiability test.

First, audit your entity consistency. Run your brand name, address variants, and category descriptors across your GBP, website, and any third-party directories relevant to your market (Wongnai in Thailand, Zomato in Indonesia, iProperty across the region). Inconsistency here is context noise — it degrades retrieval quality across both traditional and AI-mediated search.

Second, restructure your highest-traffic local landing pages for scannability. If an agent — or an AI Overview — is pulling a direct answer about your opening hours, service radius, or pricing, it should find a clean, structured fact, not a paragraph of brand narrative.

Third, get your product or service schema to completeness, not just presence. A schema tag that’s technically valid but missing key attributes is the equivalent of a GBP listing with no photos. It signals effort without providing signal.

Key Takeaways

• Tactics without a falsifiable causal mechanism — like llms.txt — are resource drains; invest that effort in context engineering instead
• Agentic commerce demands clean product feeds, complete schema, and Markdown-formatted content that AI intermediaries can parse without a browser
• Local SEO’s core disciplines — entity consistency, structured answers, complete attribute data — are exactly what context engineering requires at scale

The question worth sitting with: if AI agents are increasingly the first point of contact between a consumer and your brand, does your current content architecture make your business legible to a system that doesn’t read — it retrieves? That’s not a rhetorical question. It’s a site audit waiting to happen.

---

At grzzly, we work with brands across Southeast Asia on exactly this intersection — where local search mechanics meet the emerging demands of AI-mediated discovery. Whether that’s GBP infrastructure, schema implementation, or building content architectures that hold up under agentic query patterns, we’ve been in the weeds on this so our clients don’t have to be. Let’s talk