AEO & GEO in 2026: A Guide for Southeast Asian Brands

Almost nobody quotes the single most useful fact about optimising for AI search. In the one peer-reviewed study that actually measured what changes a source's visibility inside AI-generated answers, adding statistics lifted visibility by 37% and adding quotations by 40%. Keyword stuffing, the oldest SEO reflex, reduced it by 9% (Aggarwal et al., "GEO: Generative Engine Optimization," arXiv:2311.09735, ACM KDD 2024, 10,000-query benchmark).

That study is the backbone of this guide, because most AEO/GEO content isn't built on evidence at all. It's built on recycled vendor claims, several of which fall apart the moment you check them. This is a guide for brands in Singapore, India, and Indonesia who want to be cited by ChatGPT, Perplexity, and Google's AI Overviews. It's also, deliberately, a guide that tells you where the data runs out.

Key Takeaways

• The only peer-reviewed GEO study found the highest-impact tactics are adding statistics (+37%), quotations (+40%), and citing sources (+28%). Keyword stuffing hurt (−9%) (arXiv:2311.09735).

• llms.txt, the file every AEO vendor tells you to add, is used by zero confirmed major AI engines — Google has said plainly that Search does not use it.

• Google restricted FAQ rich results to government and health sites back in August 2023. If a guide promises you FAQ snippets, it's years out of date.

• AI Overviews reached India in August 2024 and added Indonesian in October 2024 — but no rigorous study of AI-search usage in SEA/India exists yet. That gap is an opportunity, not a footnote.

AEO and GEO: What They Actually Mean

Two acronyms, often conflated, with genuinely different origins.

GEO — Generative Engine Optimization has a precise, citable source: a 2023 academic paper from researchers at Princeton, Georgia Tech, the Allen Institute, and IIT Delhi (Aggarwal et al.), published at KDD 2024. It defined the problem — how to improve a source's visibility inside the answers that generative engines produce — and, crucially, measured solutions against a 10,000-query benchmark. When we talk about GEO, we're talking about something that was tested.

AEO — Answer Engine Optimization is an industry coinage, popularised by SEO practitioner Jason Barnard around 2017, with documented journalistic usage by early 2018 (Search Engine Watch). It's the broader, looser umbrella term for making your content the answer an AI engine returns. Useful concept, no academic backbone — so treat AEO as the goal and GEO as the part with evidence behind it.

AI Overviews reached India in August 2024 and added Indonesian that October. Photo by NIR HIMI via Unsplash.

In practice you'll see the terms used interchangeably, and that's fine. What matters is the distinction between the tactics that have been measured and the ones that are just asserted — which is exactly where most guides go wrong.

What Actually Moves Visibility Inside AI Answers

The GEO study tested nine content strategies against a benchmark of 10,000 real queries spanning arts, finance, science, law, health, and business, measuring how each changed a source's visibility in generative-engine responses. Here is the whole result, because the ranking is the insight:

Horizontal bar chart of the GEO study's per-tactic results. Adding quotations lifts visibility 40 percent, adding statistics 37 percent, improving fluency 29 percent, citing sources 28 percent, authoritative tone 17 percent, easier-to-understand language 14 percent, technical terms 10 percent, more unique words 5 percent. Keyword stuffing reduces visibility by 9 percent.+40%Add quotations+37%Add statistics+29%Improve fluency+28%Cite sources+17%Authoritative tone+14%Easier to understand+10%Technical terms+5%More unique words-9%Keyword stuffingChange in visibility inside AI answers, by tacticSource: Aggarwal et al., GEO (arXiv:2311.09735), KDD 2024. 10,000-query benchmark

Source: Aggarwal et al., GEO (arXiv:2311.09735), KDD 2024 — 10,000-query benchmark

TacticChange in AI-answer visibilityAdd quotations+40%Add statistics+37%Improve fluency+29%Cite sources+28%Authoritative tone+17%Easier to understand+14%Technical terms+10%More unique words+5%Keyword stuffing−9%

Read the top and the bottom together. The two biggest levers — adding quotations (+40%) and statistics (+37%) — are about making your content more citable. A generative engine reaching for something concrete to attribute finds a stat or a quote and lifts it. Citing your own sources (+28%) works for the same reason; it signals the kind of grounded content these systems prefer to surface.

And at the bottom: keyword stuffing reduced visibility by 9%. The single most durable habit from a decade of SEO actively works against you in generative search. That one data point should reframe how you think about this entire discipline — AEO is not SEO with new keywords. In several respects it's the opposite.

The practical translation: the content that gets cited by AI is content that would survive a fact-checker. Specific numbers, attributed quotes, named sources, plain language. If that sounds like the discipline behind good journalism rather than good SEO, that's the point — and it's a standard we hold every post on this blog to, which is not a coincidence.

The llms.txt Question — and Why We're Not Telling You to Add One

Walk through any "AEO checklist" and you'll hit llms.txt near the top: a Markdown file at your site root that summarises your content for large language models. Proposed by Jeremy Howard of Answer.AI in September 2024, it's a reasonable idea.

Here's what those checklists leave out: no major AI engine has confirmed it uses llms.txt, and Google has explicitly said Search does not. Google's Gary Illyes said so at a Search Central event. John Mueller compared it to the old keywords meta tag — a file publishers fill in that consumers ignore. And Google's own AI-optimisation guidance states you don't need any AI-specific text file to appear in Search. Server-log analyses show the major crawlers don't even request the file. It is, right now, a standard that publishers adopt and engines don't consume.

That doesn't make it harmful — it's cheap, and if adoption ever comes you're ready. But spending real effort on llms.txt while believing it drives citations today is optimising for a mechanism that isn't switched on. We'd rather you spend that hour adding statistics and sources to a page, where the evidence is unambiguous.

The FAQ Schema Trap Most Guides Fall Into

This one catches even careful marketers, because the advice was correct — three years ago.

Google restricted FAQ rich results to "well-known, authoritative government and health websites" in August 2023, and deprecated HowTo rich results entirely (Google Search Central, "Changes to HowTo and FAQ rich results"). For an ordinary commercial site, adding FAQPage schema no longer earns the visible FAQ dropdown in Google results. Any 2026 guide still promising you FAQ rich snippets is working from a pre-2023 playbook.

So is FAQ schema pointless? No — and the distinction matters. The structured data still helps machines parse your question-and-answer content, which is a different mechanism from Google's visible rich result. It's exactly why this page carries FAQ schema while claiming no snippet from it — the parsing value is real even where the rich result is gone. Question-shaped content with clean answers is easy for an AI engine to lift into a response. Add the schema for machine-readability and AI citation; do not add it expecting a Google snippet that Google stopped granting in 2023. Precision here is the whole game.

How the Engines Actually Choose What to Cite

A brief, honest map — labelled by what's documented versus what's inferred.

Google AI Overviews (documented): launched in India in August 2024 in English and Hindi, and in October 2024 expanded to 100+ countries, adding Indonesian among other languages (Google). That rollout timing is the cleanest SEA/India primary source available — worth knowing precisely, because most regional AEO content gets the dates vague or wrong.

Perplexity (inferred, not vendor-confirmed): independent technical analyses describe a retrieval-augmented pipeline that pulls roughly ten pages and cites three to four of them inline. Treat that as a reported mechanism, not an official spec — Perplexity hasn't published the details, and we won't dress up a third-party teardown as documentation.

ChatGPT search (undocumented): we could not find an official OpenAI description of how its search feature selects citations. Anyone stating precise ChatGPT citation rules is guessing. The defensible move is to make your content maximally citable — the GEO tactics above — rather than to reverse-engineer one engine's black box.

The pattern across all three: you cannot game the selection mechanism, because it's either undisclosed or unstable. You can only make your content the obvious thing to cite.

Why This Matters More in Southeast Asia and India Right Now

The market backdrop is real, even after we strip out the inflated numbers.

India's D2C ecommerce market is large and growing fast. One syndicated estimate puts it at $108.76B in 2026, growing at a 24.3% CAGR (Mordor Intelligence). We attribute the figure precisely because its decimal confidence reflects one vendor's model, not a consensus.

Column chart of Mordor Intelligence's estimate for the India direct-to-consumer ecommerce market: 87.5 billion dollars in 2025, 108.76 billion in 2026, and 322.1 billion by 2031, a compound annual growth rate of 24.3 percent.$87.5B2025$108.76B2026$322.1B2031India D2C ecommerce market, estimated (24.3% CAGR)Source: Mordor Intelligence, India D2C E-commerce Market (estimate). Retrieved 2026-07-12

Source: Mordor Intelligence, India D2C E-commerce Market (estimate)

Against that growth sits a genuine content vacuum. When we scanned who ranks for "what is AEO" and "AI search optimization," the answer was uniform: capable, US- and globally-framed articles from Western SaaS tools and agencies. Not one referenced India, Indonesia, or Singapore — the regional AI-search rollout, the Hindi and Bahasa Indonesia language coverage, or any local usage pattern. For a SEA-based brand, that's the opening. The engines already serve AI answers in your markets and languages. The optimisation content written for your markets essentially doesn't exist yet — the same gap we mapped for cost data in our analysis of why CAC keeps rising.

We'll state a gap as plainly as we can, because it's a credibility asset rather than an embarrassment. There is no methodologically rigorous, SEA- or India-specific study of AI-search usage share or citation behaviour. We looked. What circulates as "AI traffic in India converts at X%" traces to vendor extrapolations, not regional research. The rollout dates are documented; the regional behaviour is not measured. A brand — or an agency — that runs the first honest measurement of how AI search actually behaves in these markets would own a citation magnet nobody else has. It's the same bet this blog is built on: publish the data that doesn't exist yet.

What the Data Says About Traffic — and What It Doesn't

Two facts about AI search get repeated constantly. One is solid; one is inflated; and a third, widely-quoted "conversion" claim is simply invented. Here's the honest version of each.

AI Overviews are prevalent but volatile. Across 2025, the share of Google searches showing an AI Overview swung from about 6.5% in January to a peak near 24.6% in July, then fell back toward 15.7% by November (Semrush, 10M+ terms). Pew Research, using a cleaner sample of 68,879 real searches, put it at 18% in March 2025.

Line chart of the share of Google searches showing an AI Overview across 2025, per Semrush tracking of 10 million-plus terms: about 6.5 percent in January, rising to a peak near 24.6 percent in July, then falling to about 15.7 percent by November. A Pew Research reference line marks 18 percent for March 2025.0%10%20%30%Pew: 18% (Mar 2025)JanMarMayJulSepNov24.6%Share of searches showing an AI Overview, 2025Sources: Semrush (10M+ terms) via Search Engine Land; Pew Research. 2025

Sources: Semrush (10M+ terms) via Search Engine Land; Pew Research, 2025

The single stat worth internalising: "AI Overviews appear in 25%+ of searches" cherry-picks the July peak. The honest statement is a volatile 15–25% range, method-dependent. We flag this because the precise-sounding "25%+" is exactly the kind of number that gets laundered into strategy decks as settled fact.

Zero-click is real and measurable. Pew's data shows the behavioural shift cleanly: when an AI summary appears, users click a result link far less often and end their session more often.

Paired bar chart from Pew Research. When no AI summary appears, 15 percent of users click a result link and 16 percent end the session. When an AI summary appears, only 8 percent click a link and 26 percent end the session.15%8%Clicked a result link16%26%Ended the browsing sessionNo AI summaryAI summary shownWhat users do when an AI summary appearsSource: Pew Research Center, 68,879 searches, March 2025 (published Jul 2025)

Source: Pew Research Center, 68,879 searches, March 2025

That is the actual case for caring about AI citation. Not a five-times conversion multiplier — see below — but a documented shift in what users do when the answer is served on the results page. If the answer is going to be assembled for the user, you want your brand in the answer.

The Statistics We Rejected — Including One From a Strategy Deck

House rule for this blog: every number traces to a named, dated, primary source, and the ones that don't get named as rejected. These circulate widely in AEO content — including in strategy documents — and failed verification:

1. "AI referral traffic converts at 14.2% vs 2.8% for Google organic (5x)." The most-quoted AEO stat of 2026, and it traces to a single agency's self-reported data on its own clients (Opollo's "2026 AI Search Benchmark Report") — reframed as an industry benchmark. The "independently confirmed across 12 million visits" line that travels with it does not exist in any checkable form. One case study wearing a benchmark's clothes. We don't build arguments on it, and neither should you.

2. "AI search converts five times better than organic." Same root as above; same problem. Directionally, AI-referred visitors may well convert better — higher intent, pre-qualified by the answer — but there is no credible, methodology-disclosed number, and the specific "5x" is invented.

3. "India social commerce growing 60%+ year-over-year." Traces to a blog, not a research house. Credible estimates cite a ~37.5% CAGR (Mordor) — real growth, different number, and a CAGR is not a YoY figure.

4. Gartner's "search volume will drop 25% by 2026." This one is real — a genuine February 2024 Gartner prediction. But it's a forecast that missed: Google adapted with AI Overviews and held roughly 90% search share through 2026. Cite it, if at all, as a prediction that didn't land — which is far more useful than repeating it as though it came true.

If an AEO stat you've seen isn't here, this discipline is often why.

An Honest AEO Checklist for 2026

Everything above, reduced to what the evidence actually supports:

1. Put real statistics and attributed quotes in your content. The highest-impact, best-evidenced move there is (+37% and +40% in the GEO study). This is the one to start with today.

2. Cite your sources, in-line and named. +28%, same study, same mechanism — grounded content gets surfaced.

3. Write in plain, fluent language. Fluency optimisation lifted visibility 29%; "easier to understand" added 14%. Density and jargon don't help.

4. Stop keyword-stuffing. It measurably hurts (−9%). The SEO reflex is now a liability.

5. Add FAQ schema for machine-readability — but expect no Google rich snippet. The parsing value is real; the visible dropdown died in 2023 for non-government sites.

6. Add llms.txt if you like, but don't believe it drives citations yet. No major engine consumes it. Low cost, currently zero confirmed benefit.

7. Write for your region, because nobody else is. The engines serve AI answers in Hindi and Bahasa Indonesia; the optimisation content for those markets is a vacuum.

8. Measure what you can, and say where the data ends. Nobody has rigorous SEA AI-search behaviour data. Being the source that admits that — and then goes and gathers it — is itself the strategy.

Frequently Asked Questions

What is the difference between AEO and GEO?

GEO (Generative Engine Optimization) comes from a specific 2023 academic paper that measured which content tactics improve a source's visibility inside AI-generated answers. AEO (Answer Engine Optimization) is a broader industry term, popularised around 2017, for making your content the answer an AI engine returns. In practice they're used interchangeably; the useful distinction is that GEO's tactics have been tested and AEO's are often just asserted.

What actually works to get cited by AI search engines?

The only peer-reviewed study found the highest-impact tactics are adding quotations (+40% visibility), adding statistics (+37%), and citing your sources (+28%), with fluent, plain language also helping. Keyword stuffing — the classic SEO habit — reduced visibility by 9%. In short: make your content genuinely citable with specific, attributed, well-sourced information.

Do I need an llms.txt file for AI search?

No major AI engine has confirmed it uses llms.txt, and Google has said explicitly that Search does not use it. It's a publisher-side proposal from September 2024 with no confirmed consumption by any major engine. It's cheap to add and harmless, but it does not currently drive AI citations, and any guide implying otherwise is overstating it.

Does FAQ schema still work in 2026?

Not the way it used to. Google restricted FAQ rich results to authoritative government and health sites in August 2023, so ordinary commercial sites no longer get the visible FAQ dropdown in search results. FAQ schema still helps machines parse your question-and-answer content for AI citation, but don't add it expecting a Google rich snippet.

Is it true that AI traffic converts five times better than Google organic?

There's no credible, methodology-disclosed source for it. The widely-quoted "14.2% vs 2.8%" figure traces to a single agency's self-reported data on its own clients, reframed as an industry benchmark, and the "independent confirmation" cited alongside it doesn't exist in verifiable form. AI-referred visitors may convert better because they arrive pre-qualified, but the specific 5x multiplier is invented.

When did AI Overviews launch in India and Indonesia?

Google AI Overviews launched in India in August 2024 in English and Hindi, and expanded in October 2024 to more than 100 countries, adding Indonesian among other languages. The rollout timing is well documented; what isn't documented is any rigorous study of how AI search is actually used, or who it cites, within Southeast Asia and India.

Conclusion

AEO in 2026 is less mysterious than the content around it suggests. One peer-reviewed study tells you the high-value moves — statistics, quotations, cited sources, plain language — and tells you the classic SEO reflex now backfires. The rest is discipline: don't add files engines ignore, don't promise snippets Google killed in 2023, and don't repeat conversion stats that trace to a single vendor's marketing. For brands in Singapore, India, and Indonesia there's an unusual window: the engines already serve AI answers in your markets and languages, and almost nobody has written the optimisation playbook for them. The way to get cited by an answer engine turns out to be the way to deserve it — publish specific, sourced, checkable content. We built this whole blog on that bet.

How this guide was compiled. The load-bearing evidence — the per-tactic GEO results — comes from Aggarwal et al. (arXiv:2311.09735, KDD 2024), verified against the paper. AI Overview prevalence is triangulated across three independent sources: Pew, Semrush, and the reported BrightEdge range. The llms.txt and FAQ-schema positions are verified against Google's own statements and Search Central documentation, and the AI Overviews rollout dates come from Google's blog. Four widely-circulated claims failed verification and are named as rejected in the article rather than omitted — including the "14.2% vs 2.8%" conversion stat, which traces to a single agency's self-reported client data. Where a figure rests on one vendor's model (India D2C market size, Mordor Intelligence) we say so on the chart. And we state plainly that no rigorous SEA/India AI-search behaviour study exists. Written by Abhilash LR, founder of Coact, a performance marketing agency working with ecommerce and app businesses across Singapore, India, and Indonesia.

Continue Learning

• What is AEO? Answer Engine Optimization Explained — the plain-language primer

• What Attribution Model Should You Use? — measurement in an age of thinner data

• SEA & India Growth Benchmarks: What's Real — the same verify-everything discipline, applied to regional data

• Our editorial standards — how we verify every statistic in guides like this one.

• Talk to our team about getting your brand cited by AI search in your market.