AI Search ROI: Measuring What LLMs Actually Send You

Forrester’s latest research puts AI search adoption among B2B buyers at 94% — ranking it above vendor websites, product experts, and sales reps as a research tool. NielsenIQ adds that 42% of consumers are now using AI to research product purchases. Those numbers aren’t a trend. They’re the new baseline. And most Southeast Asian marketing teams are measuring exactly zero of it.

The Attribution Gap Nobody Wants to Admit

Here’s the uncomfortable reality: your GA4 dashboard is not built for generative search. When ChatGPT, Perplexity, or Google’s AI Overviews send someone to your site, that visit often arrives as direct traffic or gets lost in referral noise — not cleanly tagged as “AI-referred.” Ahrefs’ Louise Linehan documents this problem precisely: AI search ROI is structurally messy because the referral chain is broken or obscured by the LLM interface itself.

The practical fix isn’t waiting for platform consensus on tracking standards. It’s building a parallel layer now. Segment your direct traffic by landing page — pages that rank well for informational queries but sit deeper in your architecture are disproportionately likely to receive AI referral traffic. Cross-reference against your GEO-targeted content (structured FAQs, entity-rich explainer pages) and watch for anomalous direct spikes when you publish. That correlation isn’t proof, but it’s signal — and signal is what you’re working with right now.

What Google’s 14-Day Spam Update Actually Signals

Google’s September 2026 spam update rolled out over nearly two weeks — longer than any of the three earlier spam updates it ran this year, according to Search Engine Journal. The extended timeline isn’t an operational footnote; it reflects the increasing complexity of what Google is trying to isolate.

The updates this year have consistently targeted scaled, low-quality content and manipulative link patterns — precisely the tactics that spiked when generative AI made content production nearly free. The implicit message for GEO practitioners: entity authority and topical depth are being stress-tested harder than ever. A brand that ranks via genuine expertise and consistent entity presence across structured sources — Wikipedia, industry databases, authoritative local directories — is more insulated than one that ranks through content volume alone. In Southeast Asia specifically, where local-language content is still thin in many verticals, the temptation to flood thin markets with AI-generated pages is high. The September update is a reminder that Google is explicitly hunting for that pattern.

---

Multi-Location SEO as a GEO Infrastructure Problem

Semrush’s framework for scalable multi-location SEO surfaces something that GEO strategists should read carefully: the recommendation that each branch location starts with one approved canonical data record that feeds its location page, Google Business Profile, schema markup, and third-party listings simultaneously. That’s not just operational hygiene — it’s entity architecture.

For brands operating across Southeast Asia’s fragmented markets — say, a financial services firm with offices in Bangkok, Kuala Lumpur, Jakarta, and Manila — inconsistent NAP (name, address, phone) data across platforms doesn’t just hurt local pack rankings. It creates entity ambiguity that LLMs inherit. When Perplexity or Gemini tries to answer “best [category] in KL,” it’s synthesising from structured sources. If your entity signals are contradictory across those sources, you’re less likely to appear in the generated answer — regardless of how well your page ranks in blue-link search.

The implementation priority: audit your entity consistency before your content strategy. Tools like Semrush’s Listing Management or BrightLocal can surface the discrepancies. Fix the foundation before building upward.

Building an Attribution Stack That Doesn’t Lie to You

SEO.com’s Dan Shaffer outlines the core mechanics of SEO attribution clearly: tracking the path from organic touchpoint to form submission, lead, and revenue conversion. The framework is sound for traditional search. For AI search, it needs extension.

Three layers worth adding to your current attribution model: First, UTM-tag every asset you publish that’s designed for AI visibility — structured FAQs, entity pages, comparison content — so you can track when those specific pages drive conversions, even if the referral source is obscured. Second, run quarterly brand mention audits across LLM outputs. Prompt ChatGPT, Gemini, and Perplexity with the queries your target audience actually uses and record whether your brand appears, in what context, and with what framing. That’s qualitative, but it tells you whether your GEO work is landing. Third, track branded search volume as a proxy. When AI tools mention your brand in a response, users often follow up with a branded Google search. A rising branded search trend, correlated with GEO content investment, is one of the cleaner signals available right now.

For B2B teams in Southeast Asia, where deal cycles are long and multi-touchpoint, this matters especially in the research phase. If a procurement manager in Singapore is using AI search to shortlist vendors and your brand isn’t appearing in generated answers, you’re not in the consideration set — regardless of your keyword rankings.

---

Key Takeaways

• Build a parallel attribution layer using landing page segmentation and branded search tracking — don’t wait for platform-level AI referral data to become reliable.
• Entity consistency across all local listings, schema, and structured sources is GEO infrastructure, not just local SEO hygiene — fix it before scaling content.
• Google’s extended September spam update signals sustained pressure on content-volume tactics; topical authority and entity depth are the durable differentiators.

The honest provocation here: most brands are still measuring AI search with tools built for a world where Google’s blue links were the only game in town. That gap compounds quietly — every month your brand is invisible in generated answers is a month a competitor is getting cited instead. The question isn’t whether to invest in GEO measurement. It’s how long you can afford to fly blind.

---

At grzzly, we work with marketing teams across Southeast Asia who are grappling with exactly this — building attribution frameworks that capture AI-influenced discovery, and structuring entity and content strategies that show up in generated answers, not just ranked pages. If your current measurement stack can’t tell you what the LLMs are sending you, that’s the conversation to start. Let’s talk