Gemini 3.7 Powers AI Mode: What It Means For AI SEO

Two shifts hit Google Search within a fortnight, and both arrived while most Malaysian marketing teams were still adjusting to the previous set of rules. Google released Gemini 3.7 Flash on 13 August 2026, describing it as its most intelligent workhorse model. Days later, Search Engine Journal reported that AI Mode, the conversational answer surface built into Google Search, had begun running on that newer model, and Search Engine Roundtable noted that a fresh Google spam update had finished rolling out on 21 August 2026. Neither change came with much warning.

That cadence tells you something plain about AI SEO in 2026. You are optimising for a target that keeps moving, sometimes twice in a single month.

What actually changed, in plain terms

Gemini 3.7 Flash is faster and cheaper to run than the model it succeeds, and Google reported sharp gains on coding and web-development benchmarks. For search, the detail that counts is not the benchmark table. It is that the AI system deciding which pages to summarise and cite inside an answer has quietly been swapped for a different one. A page that was being pulled into an AI Mode response last month has no guarantee of the same treatment this month, because the thing reading it has changed.

Spam updates are not new either, and Google has been clear that it does not deploy them before announcing. What is different now is the reach. When the same page feeds both classic blue links and AI-generated answers, a spam action can dampen your visibility in places that never appear on a results page.

Citations were never as stable as they looked

Anyone treating a strong citation month as a durable win is in for an awkward review later. Ahrefs, analysing millions of AI Overview URLs, found that only 38% of pages cited in AI Overviews still rank in the top ten organic results in its March 2026 study, down from 76% a year earlier. Google increasingly builds an answer by fanning a question out into sub-queries and pulling sources for each, so a page can gain or lose its place with no change on your side at all.

It also helps to be clear about what a spam update does and does not do. It is not a fresh ranking philosophy, and it tends to act on patterns Google has warned about for years, such as scaled low-value content, thin pages built only to catch a query, and manipulative link behaviour. The practical implication for a careful business is small on the day and large over time. If your pages already earn their place, an update mostly clears out the noise competing with you. If they do not, each pass chips away at ground that was borrowed to begin with.

Why Malaysian teams cannot file this under American noise

Malaysia is not watching from the sidelines. Research from the e-Conomy SEA 2025 report found that 74% of Malaysian digital consumers use AI tools and features daily, which means AI answers already sit between a question and a purchase here. Every model swap and spam update therefore lands on an audience that genuinely uses these surfaces, not one experimenting with them. The cost of ignoring the churn is not theoretical, and it shows up as enquiries that quietly route to whichever competitor the assistant happened to cite this week.

What to do before the next swap

None of this rewards panic, and it does reward a few steady habits. Before the next model change lands, four checks earn their place:

• Answer the core question early on the page, in two or three sentences, so extraction systems can lift it cleanly.
• Keep a dated record of which pages earn AI citations, so you can watch movement instead of guessing at it.
• Spread the surfaces you court, because concentrating everything on one model or feature remains a big risk for Malaysia’s GEO and answer-engine plans.
• Fix the fundamentals a spam update rewards anyway: clear structure, sourced claims and pages built for the reader first.

This is the sort of unglamorous, ongoing work our search engine optimisation team folds into a retainer, precisely because it cannot be done once and filed away.

The view from both sides of the Causeway

Our team is based in Singapore and works across both markets, which is where a change like this becomes useful instead of alarming. Google tends to widen access to new AI search features in Singapore before Malaysia, so Singaporean brands often feel a model shift first, and Malaysian query behaviour follows a few weeks later. A company running one content operation for both markets can turn that lag into a rehearsal: watch which citations move in Singapore, then adjust the Malaysian pages before the same wave arrives. Firms that assume the two markets behave identically throw that early warning away, usually without noticing they ever had it.

Conclusion

The uncomfortable truth about AI search is that stability is not coming back. Models will keep changing, and the sensible response is a content operation built to absorb that, not one tuned to a single good week. If you would like a calm second read on how your pages are holding up across AI answers, our team is glad to talk it through and help you plan for the version that has not shipped yet.