AI Search / GEO for SMEs in Southeast Asia, a field guide

Most articles about AI search are written for marketers at companies with a dedicated SEO team. This one is for the founder or marketing lead at a SEA startup or SME who just realised their buyers research with ChatGPT before they ever open Google. If that is you, the next 2,000 words are a working playbook based on what I run in client engagements every week, not a glossy theoretical guide.

TL;DR

• Search now has two layers: classic Google rankings and AI answer engines (ChatGPT, Perplexity, Gemini, Google AI Overviews). You need to be visible inside both.
• SEA SMEs are not US enterprise. The playbooks Western marketing blogs sell do not fit smaller teams, mixed-language markets, or local buyer behaviour.
• The minimum work to show up in AI answers is mostly technical hygiene plus deliberate content shape. It is not a giant project.
• Most SEA SMEs need a 90-day starter cadence, not a 12-month transformation roadmap. The plan in section 6 is what I actually run.
• Dashboards lie about AI visibility because the engines do not report referrer data the way Google does. Measure with a manual scoreboard until that changes.

How search now has two layers

When a buyer in Jakarta or Singapore needs to figure out whether they should hire a search consultant, what an AI search audit looks like, or which engine to optimise for first, here is what they do now. They open ChatGPT or Perplexity. They type a paragraph-long question. They read the answer. Then, maybe, they click through to one or two sources the answer cited.

If your site was not one of those one or two cited sources, your business does not exist in that conversation. You will not see the search in Google Search Console because it never happened in Google. You will not see the click because the visitor read the answer and moved on. You will see a small drop in classic organic traffic, a slow rise in AI-referred sessions (when you can identify them), and a growing sense that something is changing under you.

This is what every CMO and growth lead I work with has noticed in the past six months. The data is messy because there is no single dashboard for it. The pattern is real anyway.

The mistake I keep seeing is treating this as a separate workstream. "We have an SEO team and now we need an AI search team." No. They are the same practice now. The same crawl, the same schema, the same structured content, with a slightly different second pass for extractable answers. If your team treats them as two systems, you will either duplicate work or, more often, ship neither well.

Here is the frame I prefer. Classic SEO is for the buyers who still type queries into Google. AI search is for the buyers who already moved up the funnel and want an answer, not a list of links. Most SEA buyers do both, on the same purchase journey, sometimes on the same day. So you do both, with the same content.

Why SEA SMEs are not US enterprise

Everything you read about GEO in your LinkedIn feed was probably written by a US enterprise consultancy, for a US enterprise audience. That audience has a five-plus person in-house SEO team, a dedicated content writer pipeline, a budget that absorbs a six-figure AI search audit, and an English-only market.

SEA SMEs have none of those things. The companies I work with most often look like this. A founder plus a marketing lead, sometimes one or two writers on contract, a budget that has to defend itself against paid acquisition, and a buyer mix that crosses English, Bahasa Indonesia, Singlish, and sometimes Vietnamese or Thai depending on the market.

The Western GEO playbook breaks in three places when you try to apply it here.

First, the localisation. AI Overviews behave differently when the search is in Bahasa Indonesia than in English, even for the same buyer. Perplexity tends to be more aggressive at pulling local-language sources than ChatGPT. Gemini leans on Google's own knowledge graph, which weights English content heavier. You cannot optimise for an aggregate behaviour because there is not one.

Second, the team size. The "appoint a dedicated GEO lead" advice that fills US playbooks assumes you have headcount to spare. A SEA SME with three marketing people cannot afford to dedicate one of them to a discipline that, today, still does not produce a clean attribution number for the CFO. The work has to fit inside the existing roles, or it does not happen.

Third, the budget tolerance. A US founder will spend tens of thousands on an AI search audit because there is enough enterprise revenue downstream to justify it. A SEA SME founder will not. The work has to be cheap enough to start before there is proof it works, or it does not start.

These three constraints are not problems to solve. They are the shape of the market. The right plan respects them.

The minimum work to be visible in AI answers

If you only do five things, do these. I run a version of this list as the opening week of every consultancy engagement.

1. Ship Person and Organization schema. AI engines and Google increasingly resolve who you are before they decide whether to trust you. Without Person plus Organization schema, your site is anonymous in their eyes even if your domain is well-known. The implementation is a JSON-LD block in your site's head section. It takes one afternoon. Most sites I audit do not have this.

2. Confirm AI crawlers can read your site. Open your robots.txt. Look for crawler names like GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Applebot, CCBot. They should not be blocked. If your robots.txt has a blanket Disallow for any of these, your site is not in those engines' indexes. Reverse it.

3. Front-load the answer. AI engines lift the first 100 words of a page more often than any other section, in every audit I have run. Whatever the page is about, the first paragraph should answer the question the page exists to answer, in a sentence the engine can quote verbatim. No introductory throat-clearing, no scene-setting. Answer first, expand after.

4. Make claims self-contained. "As mentioned above" or "see chart" kills your extraction odds. Each claim should make sense on its own, with the numbers, names, and dates inline. If you want a sentence to be quoted, write it like the only one the engine will read.

5. Submit a sitemap to Google Search Console and Bing Webmaster Tools. Bing's index feeds parts of ChatGPT and Copilot. Most SEA SMEs I audit have GSC set up but not Bing, then wonder why their content does not appear in ChatGPT answers. Two minutes of setup, real downstream impact.

That is the minimum. None of it is glamorous. All of it is operational hygiene that compounds.

When AI search actually matters for your business

Not every SME needs to invest in AI search this quarter. Here is the rough decision tree I use.

Yes, prioritise it now.

• B2B SaaS, especially anything tools-related (analytics, CRM, marketing tech, dev tools). Your buyers research with AI first because the category is full of jargon and AI handles jargon well.
• Categories where buyers ask "what is X" or "how do I Y" before they evaluate vendors. AI handles those questions far better than a Google SERP, so the conversation has shifted there.
• Anything where the purchase decision involves comparing two or three options. AI answer engines now produce comparison tables on demand. Be in those tables.

Probably yes, but slower.

• Consumer brands with a research-heavy purchase (high-ticket fashion, finance, health). AI is in the consideration phase, less in the awareness phase.
• Local services with national reach. Worth doing because competition is light, but Google Maps still owns most of the demand.

Probably not yet.

• Purely local services with low search volume (a single-location restaurant, a neighbourhood dentist). Classic local SEO is still most of the win.
• Categories where buyers want to see the product before they commit (most physical retail). AI answers can describe the product; they cannot show it the way a high-quality image search can.
• Enterprise B2B where buyers go straight to vendor demos. AI search does not reshape that funnel much yet.

If you are in the first bucket, the rest of this article is your priority list. If you are in the second, do the first three items in the next section over a quarter, not a month. If you are in the third, save this and revisit in six months.

A 90-day starter plan

Here is what I run for SEA SMEs who decide to take this seriously. Three months, one person's time at roughly twenty percent allocation, no big external spend.

Days 1 to 30, foundation. Verify your site in Google Search Console and Bing Webmaster Tools. Submit your sitemap to both. Implement Person and Organization schema. Confirm AI crawlers are allowed. Audit your top ten existing pages and front-load the answers. Define the twenty specific search queries you most want to be cited for. Manually test each query in ChatGPT, Perplexity, Gemini, and Google AI Overviews. Record what comes back. This is your baseline.

Days 31 to 60, content base. Publish four blog posts that directly answer four of the twenty queries from your scoreboard. Not generic "ultimate guide" pieces. Specific, operational answers a reader can act on. Each post should link up to the canonical reference page on your site (a pillar) and sideways to two or three related posts. Build out a glossary of the terms your buyers actually use, because AI engines lift glossary definitions verbatim. Each glossary entry is its own opportunity to be cited.

Days 61 to 90, first audit, iterate, expand. Re-run your twenty-query scoreboard. Compare to the day-1 baseline. You should see the first one or two queries flip from "not cited" to "cited" on at least one engine. Identify which posts are driving citations and write more in that shape. Identify which queries are still empty and either write a post for them or accept that the keyword is wrong for your business.

This is the operational version of what I describe in the consultancy and training offerings. The difference between the two is whether I run the cadence or whether I teach your team to. Both work; pick by which constraint matters more, time or knowledge transfer.

If you would rather have it executed end to end without managing the workstream yourself, the in-house agency mode is built for exactly that.

What to measure when dashboards do not show it

The honest reality is straightforward. There is no good dashboard for AI search visibility. The engines do not consistently send referrer data. Google Analytics will sometimes show traffic from chat.openai.com or perplexity.ai, but it undercounts heavily because most users copy answers without clicking. Most "AI traffic" dashboards you see in marketing tool ads are estimating, not measuring.

Here is what I actually instrument for clients.

Google Search Console is the primary signal, not third-party tools. Impressions are the leading indicator. When a query you target starts showing impressions even before clicks, AI Overviews is probably surfacing your page. This shows up in GSC's Search Type filter under "Web" and increasingly under "Web with AI Overviews" once it lands in your region.

A manual twenty-query scoreboard, run monthly. Open ChatGPT, Perplexity, Gemini, and AI Overviews. Type each of your twenty target queries. Record whether your site is cited, paraphrased, or absent. Yes, this takes an hour a month. Yes, it is more reliable than any dashboard you can buy right now. I run this for every client; the scoreboard catches movement that classic SEO tools miss entirely.

Bing Webmaster Tools. Bing is an under-counted source of AI search traffic because ChatGPT and Copilot lean on Bing's index for portions of their retrieval. If you only have GSC, you are blind to a meaningful slice.

Server-side referrer logs. Look for hits with Referer headers from chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com. These are the rare clicks. Their count is a floor; the actual influence is higher.

Brand search in GSC. When someone reads about you in an AI answer and then searches your brand name in Google to find your site, GSC shows the brand search. A rising brand search line with no paid driver is one of the cleanest AI visibility signals you can get.

You will not get a single number. You will get a triangulation of signals that, together, tell you whether the work is moving. Make peace with that, or you will spend more time looking for dashboards than doing the work.

The two biggest mistakes I see

The biggest mistake I see SEA SMEs make is waiting for proof before starting. The proof for early-stage AI search work is slow and noisy. By the time you have a clean attribution number, the competitors who started a year before you have a 12-month head start in entity recognition and cited content. That gap compounds.

The second biggest mistake is the opposite: hiring a US-style agency, spending tens of thousands on a 200-page audit, and never executing the recommendations because the implementation falls on the same two-person marketing team that already could not keep up. Audits do not move the needle; published, cited content does.

What I have found works for the companies I run engagements with: a small foundation in week one, four posts in the next two months, a measurable scoreboard, and a willingness to iterate on what gets cited. Three months in, you have signal. Six months in, you have momentum. Twelve months in, you are reliably cited for several queries your competitors have given up on.

If you want a structured version of this for your specific business, book a call or read the consultancy offering. I run these engagements weekly across Southeast Asia. The debrief alone is usually enough to save the next three months of misdirected work.