A practical, technical walkthrough of how to get cited by ChatGPT, Gemini, and Google AI Overviews — schema, content structure, and the checks that actually move the needle.
Suncorp AI Search Optimization
AI Growth Solutions, Bangkok
There’s no “rank #1 in ChatGPT” the way there’s a rank #1 on a Google results page — there’s no fixed position, no stable SERP, and the same question asked twice can return two different answers with two different sets of citations. That doesn’t mean it’s ungovernable. It means the mechanics are different, and this guide walks through the ones that actually move a brand from invisible to cited: content structure, technical schema, source authority, and the testing loop that tells a business whether any of it worked.
This isn’t a theory piece. It’s the practical, step-by-step version of the concepts covered in our guide to GEO vs SEO — what to actually change on a page, in what order, and how to verify it made a difference. Built by Suncorp AI Growth Solutions, which runs this exact process for Thai and Southeast Asian brands as its core AI Search Optimization service.
Traditional SEO ranks pages against a query using hundreds of weighted signals resolving to a fixed, ordered list. ChatGPT, Gemini, and Google’s AI Overviews generate a synthesized answer instead, pulling from a mix of trained knowledge and, increasingly, live retrieval — deciding in real time which sources are authoritative and relevant enough to cite or summarize. There’s no position 1 through 10; there’s “cited,” “mentioned,” or “absent,” and an answer can cite three sources or none.
That shift changes what “optimization” even means. A page doesn’t need to out-rank ten competitors on a list — it needs to be the clearest, most directly-answering, most structurally-parseable source available when the model is assembling an answer. That’s a different, and in some ways more binary, target than traditional rank tracking.
Answer the question in the first two sentences of the relevant section, not after three paragraphs of preamble. Both retrieval systems and human skim-readers reward the same thing here: a direct answer up front, with supporting detail after it, rather than making the reader (or the model) infer the answer from context.
Use real, descriptive headings that match how people actually ask the question, not vague section labels. A heading like “How to Rank in ChatGPT” gives a retrieval system an unambiguous match; a heading like “Our Approach” gives it almost nothing to match against.
Write in clear, complete claims a sentence can be lifted from cleanly. A model synthesizing an answer often pulls something close to a direct sentence or two from a source — content written in short, self-contained, factually complete sentences gets quoted more cleanly than content that depends on three prior sentences of context to make sense.
Include specific, checkable facts and numbers, not vague claims. “Studies show AI search is growing” gives a model nothing concrete to cite; a specific, sourced figure does.
Implement Organization, Article, and FAQPage schema markup, filled in completely rather than with placeholder or minimal fields — schema gives retrieval systems an unambiguous, structured read on what a page is, who wrote it, and what it directly answers, reducing the model’s reliance on inferring that from prose alone.
Set inLanguage correctly, especially for Thai-language pages — an explicit language signal helps a retrieval system correctly match language-specific queries, a detail covered in more depth in our guide to whether ChatGPT answers in Thai.
Make sure the page is actually crawlable by AI retrieval bots. Check robots.txt for blocks on GPTBot, Google-Extended, PerplexityBot, and similar user agents — a page blocked from these crawlers can’t be retrieved live, regardless of how well-optimized its content is. This is a five-minute check that gets skipped surprisingly often.
Keep page load speed and Core Web Vitals healthy. Retrieval systems that fetch live content on query time can time out or deprioritize slow-loading pages the same way a frustrated human visitor would abandon one.
Use a clean, semantic HTML structure — real <h1>–<h3> tags, real <table> markup for comparison data, real list elements for steps — rather than visually-styled <div> soup that looks like a heading or list to a human eye but carries no structural signal to a parser.
Publish content with a named author and visible expertise signals. E-E-A-T (experience, expertise, authoritativeness, trustworthiness) isn’t just a Google ranking concept — it shapes which sources a model treats as citation-worthy when multiple pages cover the same claim.
Earn genuine external citations and mentions, not just backlinks in the traditional link-building sense. A brand mentioned and linked from other credible sources in its space builds the same kind of authority signal retrieval systems draw on, similar in spirit to (though not identical to) traditional backlink authority.
Keep facts, figures, and terminology consistent across your own site and other places your brand appears. Inconsistent claims about the same fact across different pages — even your own — make a model less confident about which version to trust and cite.
Add a clearly dated “last updated” signal on evergreen content, and actually update it. Both AI Overviews and ChatGPT’s live-retrieval mode weight freshness signals when a query benefits from current information; stale content with a stale date is easy for a retrieval system to deprioritize.
Implementation without verification is a guess dressed up as a strategy. The testing loop:
A malformed schema block can be worse than none at all if it causes a parser to misread the page — validate with Google’s Rich Results Test or Schema.org’s own validator after every change, not just at launch.
The technical steps above apply in both languages, but content structure, keyword phrasing, and even which questions get asked at all differ by language — see the localization-specific guide linked above for that gap in detail.
Retrieval systems and model versions change on a schedule outside any business’s control; a page optimized once in 2026 without revisiting it is optimized for a model that no longer exists by 2027.
Getting cited in a low-relevance, low-intent answer isn’t the same win as getting cited in the exact comparison or recommendation query a real customer would ask before buying.
Without knowing where you started, there’s no way to tell whether a change actually helped, hurt, or did nothing — and model behavior shifts on its own regardless of what you changed.
Getting cited by ChatGPT, Gemini, and Google AI Overviews isn’t a switch to flip — it’s the compounding result of content a model can extract cleanly, technical signals that make a page unambiguously parseable, authority signals that make it trustworthy to cite, and a testing loop that proves any of it actually worked. None of these four steps is exotic on its own; the businesses that pull ahead are the ones that do all four consistently, then keep re-testing as the underlying models keep changing.
There’s no ranking in the traditional sense — instead, focus on four areas: structuring content so a model can extract direct answers cleanly, implementing complete technical schema, building genuine authority and consistency signals, and testing citation results on a recurring schedule to confirm what’s actually working.
It varies by topic competitiveness and how much retrieval-friendly content already exists, but most businesses see measurable shifts within 4-8 weeks of consistent technical and content changes, confirmed through the re-testing process described above — not from a single change made once.
Yes — strong traditional SEO fundamentals (crawlability, page speed, authoritative backlinks, clear content) underpin AI visibility too, since retrieval systems draw on much of the same infrastructure.
For most businesses, it’s restructuring existing high-value pages to answer the target question directly in the first two sentences of the relevant section — it’s the change most likely to affect both AI citation and human skim-readability, and it doesn’t require any technical implementation to start.
The four-step process is the same, but Thai-language content needs its own baseline testing, its own inLanguage tagging, and awareness that AI assistants sometimes default to English even for Thai-language queries
Track it directly with the baseline-then-retest loop in Step 4, or use a dedicated monitoring platform to automate the tracking.
A team with existing SEO expertise can apply most of the steps in this guide directly. A business that wants the Thai/SEA-specific strategy, testing, and implementation work handled end to end typically moves faster with a specialist. Book a free AI visibility consultation and we’ll assess where your brand currently stands.
Suncorp AI Growth Solutions is a Bangkok-based agency building AI Search Optimization (GEO/AEO), AI Advertising, and AI Content services specifically for the Thai and Southeast Asian market — the localization gap this guide covers is the one we built our practice around.
Get a free, scored AI Visibility report in about two minutes
— no commitment required.
September 11, 2026
