The content strategy that moves AI visibility fastest over 90 days rests on five levers: schema markup, direct-answer content structure, entity signals from external sources, server-side rendering, and measurement.

Key Takeaways

Brand websites account for only 5-10% of the sources AI uses to build answers. 85% of brand mentions in AI answers come from third-party domains, per McKinsey research.¹ This is why entity signals matter more than self-published content.
Google AI Overview and ChatGPT handle language in opposite ways: AIO favours sources in the query language, while ChatGPT translates the prompt to English at the retrieval stage. Good Vietnamese content wins on AIO but needs English-language signals to enter ChatGPT’s candidate pool.²³
Evidence for schema markup directly lifting citation rate is mixed. Schema is a prerequisite for machine readability, not a standalone citation lever.⁴
Visitors arriving from AI search are 4.4x more valuable and 3x more likely to convert than other channels, even though AI traffic is still only around 1% of total sessions.⁵

Step 1: Implement Schema Markup

Schema markup is JSON-LD code that helps AI crawlers identify the content type and entities on your page. It is foundational infrastructure for AI visibility, though the evidence for its direct effect on citation rate is mixed.

This needs saying plainly, because a great deal of published advice repeats a “schema triples your citations” figure with no original research behind it. In reality, a 2026 Ahrefs study of 1,885 pages that added JSON-LD found no meaningful lift against matched controls⁶. Pulling the other way, Princeton’s GEO research found that structured data combined with authoritative citations and specific statistics produced citation rates up to 40% higher.

The practical conclusion: schema does not earn citations by itself, but without it AI systems struggle to resolve what your content is and which entity it belongs to. This is infrastructure, not tactics.

Four priority schema types:

Article: label every post. Update dateModified every time you refresh content.
FAQPage: each Question and acceptedAnswer pair is a self-contained answer unit that is easy to extract.
HowTo: each HowToStep becomes an independently citable unit.
Organization: declare your brand as a named entity, with sameAs pointing to every verified profile.

The most important technical condition: schema must be delivered server-side, not injected by JavaScript after load. Many AI crawlers do not execute JavaScript.

→ Read the full schema implementation guide with complete JSON-LD examples and a 12-step checklist

Step 2: Restructure Content for Direct Answers

AI engines extract the first 1-3 sentences of each section. Content that builds to the answer will not be cited, content that leads with the answer will.

This is the most impactful content change you can make without writing anything new. Audit your existing pages and rewrite section openings.

The before/after rule:

Before (narrative build-up):

“The relationship between schema markup and AI search is a topic that has received increasing attention over the past two years. Researchers and practitioners have noted that…”

After (direct answer):

“Schema markup is JSON-LD code that helps AI crawlers identify content type and entities on a page. It does not generate citations on its own, but without it AI systems struggle to resolve what your content is.”

Content types AI engines cite most frequently:⁷

Definition blocks: a clear, self-contained definition of a term or concept
Numbered lists with consistent sub-structure per item
Comparison tables for topics with multiple options
FAQ pairs: explicit Q&A with each answer under 60 words
Statistics with inline attribution: “Google AI Overviews appear on roughly 50% of US queries (Google, 2026)”

Content audit protocol:

For each section on your target pages: does the first sentence answer the section’s implied question?
If no: rewrite the opening sentence before doing anything else
Add data: replace every “many,” “most,” and “significant” with a specific number or named source
Frame H2s as questions where natural, “What Is AEO?” outperforms “Introduction to AEO” for AI extractability

Important: the two engines handle language in opposite ways.

This is where most AEO advice gets it wrong, and it determines how you allocate effort.

Google AI Overview favours sources in the query language. Profound’s analysis of 3.25 billion citations found that in research on the Mexican market, 96% of AI Overview citations came from Spanish-language sources. Where a quality local option exists, English sources are pushed out of the top five positions⁸. For AIO, good Vietnamese content is a real advantage.

ChatGPT translates your query into English at the retrieval stage. Peec AI’s study of 10 million prompts and 20 million fan-out queries found 43% of background searches ran in English even when the original prompt was in another language, and 78% of non-English sessions included at least one English background query. In their example, Polish users asking about local auction platforms got eBay back instead of Allegro, the dominant platform in that market⁹.

These two findings do not contradict each other. They describe different stages: Peec measured retrieval (which sources enter the candidate pool), Profound measured citation (which sources in the pool get chosen). Local sources can be filtered out at stage one, before the language-preference logic at stage two ever applies.

What this means for effort allocation:

	Google AI Overview	ChatGPT
Quality Vietnamese content	Primary lever	Necessary but not sufficient
English-language signals on external sources	Less critical	Condition for entering the pool
Practical priority	Structure and sub-query coverage	Entity signals, PR, international directories

If your customers search mainly through Google, put your weight behind well-structured Vietnamese content. If through ChatGPT, you also need presence on the English-language sources a fan-out query might reach: LinkedIn and reputable newspapers.

→ Read the in-depth guide to Google AI Overviews and how to optimise for them

Step 3: Build Entity Signals

Entity signals are the external citations and verified profile associations that tell AI models your brand is a trusted, named authority in your field.

This is the highest-leverage step of the five, and the least invested in. McKinsey research found that brand websites account for only 5-10% of the sources AI uses to construct answers, while 85% of brand mentions come from third-party domains¹⁰. Bain reports the same pattern: LLMs reach for sources external to the brand such as reviews, earned media, and comparison sites, so being accurately represented in those venues is decisive¹¹.

Put plainly: if your entire content budget goes into your own blog, you are optimising for 5-10% of the sources and leaving 85% untouched.

Unlike schema markup (which you control) and content structure (which you write), entity signals come from third parties. This is what makes GEO slower than AEO, you cannot manufacture trust, only earn it.

The entity signal hierarchy for Vietnam brands:

Tier 1. Highest impact:

LinkedIn company page (verified, consistently updated, English and Vietnamese content)
VnEconomy, CafeF citations, editorial mentions or guest articles in Vietnamese-language industry publications

Tier 2. Supporting signals:

Forbes Vietnam, Entrepreneur Vietnam editorial mentions
Industry association memberships with web presence
Speaker listings at industry events with schema-tagged event pages

Tier 3. Baseline presence:

Google Business Profile (verified)
Wikipedia entry (for established brands)

Step 4: Set Up AI Citation Tracking

You cannot optimise what you do not measure. AI citation tracking means monitoring how often your brand or content appears in AI-generated answers for your target queries.

What to track:

Set up a list of 30-45 target prompts in the exact language your audience uses. Because AI favours content matching the query language, if your customers ask in Vietnamese then the Vietnamese prompt set is your first priority.

Metrics to record per prompt, per engine (ChatGPT, Google AI Overviews, Gemini):

Is your brand mentioned? (Yes/No)
Is your domain cited as a source? (Yes/No)
Which competitor is mentioned if not you?

Step 5: Measure AI-Referred Traffic in GA4

AI-referred traffic is the business outcome of AI citation. It confirms that AI visibility is converting to real sessions, not just impressions inside an AI response.

Why this step is worth the effort: Semrush data shows visitors arriving from AI search are 4.4x more valuable and 3x more likely to convert than other channels. AI traffic still accounts for only about 1.08% of total sessions across a 13,770-domain sample, but the unusual quality of that segment makes isolating it worth doing now rather than later¹².

By default, traffic from ChatGPT and Perplexity appears as “direct” in GA4. To separate it, append UTM parameters (?utm_source=ai_search&utm_medium=citation) to links you share through AI-accessible channels, and create a custom segment filtering for perplexity.ai or your UTM source.

The single most useful diagnostic: if citation rate rises but sessions do not, your content is being cited without being linked.

→ Read the full guide to AI visibility KPIs and how to measure them

The 90-Day Content Roadmap

Before the roadmap, one observation about the order results appear in.

In real deployments, the homepage and service pages tend to get cited by AI before new blog posts do, even when the blog posts are better optimised. Higher-level pages already carry accumulated internal linking and external mention signals, while new posts need additional crawl cycles and time to be linked to.

The planning implication: do not judge a new post’s AEO performance in its first 30 days. Optimise the high-level pages first for early results, and publish new content with a longer time horizon in mind. Invert that order and you will wrongly conclude AEO does not work, when the real issue is insufficient crawl cycles.

Month 1: Technical foundation

Add Article + FAQPage schema markup to the 5 highest-traffic pages
Rewrite section openings on those same 5 pages to lead with direct answers
Verify server-side rendering is active for core content
Set up prompt tracking: 30-45 Vietnamese-language prompts, running each prompt multiple times because AI output is probabilistic
Benchmark: record current AI citation rate as your comparison point for later months
Prioritise the homepage and service pages first, since that is where results land soonest

Month 2: Content deployment

Publish 8-10 AEO-optimised Vietnamese pillar posts targeting the highest-volume prompts in your industry
Apply schema to each new post at publication
Begin LinkedIn company page optimisation

Month 3: Entity signal building

This is the most important month, since 85% of brand mentions in AI answers come from third-party sources.

Pitch 1-2 guest articles to VnEconomy, CafeF, or a relevant Vietnamese industry publication
Update Organization schema, sameAs with all newly verified profiles
Review tracking data: which prompts now show your brand? Which show competitors? Adjust content priorities accordingly
Set up GA4 AI-referred traffic segment and begin monthly reporting

Frequently Asked Questions

Does schema markup actually affect AI citation probability?
The evidence is mixed. A 2026 Ahrefs study of 1,885 pages that added JSON-LD found no meaningful citation lift against matched controls. Princeton’s GEO research, by contrast, found structured data combined with authoritative citations and specific statistics produced citation rates up to 40% higher. Schema is a prerequisite for machine readability and entity resolution, not a standalone citation lever¹³.

Which content types are most likely to be cited by ChatGPT or Perplexity?
Definition blocks, numbered lists with consistent sub-structure, comparison tables, FAQ pairs, and statistics with inline attribution are cited most frequently. Any section that opens with a clear answer in the first 1-2 sentences is prioritised for AI extraction over content that builds to the answer.

Where do I start if I want to appear in Google AI Overviews?
Start with three highest-leverage changes: add FAQPage and Article schema markup to your highest-traffic pages, rewrite section openings to lead with direct answers, and ensure server-side rendering. These changes typically show first results within 4-8 weeks.

How do I write FAQs that get cited in AI Overviews?
Each question must be a real query a user would type into an AI engine. Each answer must be maximum 2 sentences, include at least one specific number or named entity, and answer fully in the first sentence. Implement FAQPage schema as JSON-LD, this is schema-only and does not need to be a visible section on the page.

How do I measure AI-referred traffic in GA4?
Add UTM parameters (utm_source=ai_search) to links shared via AI-accessible channels. Create a custom GA4 segment filtering for Perplexity, ChatGPT, or your UTM source. Monitor alongside AI citation rate to connect visibility to actual sessions.

Is Vietnamese-language content at a disadvantage in AI search?
No. Profound’s analysis of 3.25 billion citations found AI overwhelmingly prefers content in the same language as the query. Where quality local-language sources exist, English sources are pushed out of the top results. The Vietnamese market’s problem is not that AI ignores Vietnamese, it is that too little Vietnamese content is structured well enough to be cited.

Written by Abbie, AEO Expert at Brandgineer | brandgineer.co
Optimised for AI visibility across ChatGPT, Google AI Overviews, and Perplexity.

Ready to implement this with expert support? Book a consultation with Brandgineer →

New Front Door to the Internet: Winning in the Age of AI Search. McKinsey & Company. https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search ↩︎
How Query Language Reshapes AI Citations. Profound, 2026. https://www.tryprofound.com/blog/how-query-language-reshapes-ai-citations
↩︎
ChatGPT Searches in English, Even When You Don’t. Peec AI, 2026 (10 million prompts, 20 million fan-out queries). https://peec.ai/blog/chatgpt-searches-in-english-even-when-you-don-t
↩︎
Schema Markup and AI Citations: What the Data Actually Shows (2026). fSEO, summarising Ahrefs and Princeton GEO research. https://fractionalseo.services/blog/schema-markup-ai-citations-data/
↩︎
AI Search Referral Traffic Statistics 2026. Semrush and Conductor data. https://thestacc.com/blog/ai-search-referral-traffic-stats/ ↩︎
Schema Markup and AI Citations: What the Data Actually Shows (2026). fSEO, summarising Ahrefs and Princeton GEO research. https://fractionalseo.services/blog/schema-markup-ai-citations-data/
↩︎
Content Structure for AI Answer Engines. HubSpot. https://blog.hubspot.com/marketing/aeo-page-structure ↩︎
How Query Language Reshapes AI Citations. Profound, 2026. https://www.tryprofound.com/blog/how-query-language-reshapes-ai-citations ↩︎
ChatGPT Searches in English, Even When You Don’t. Peec AI, 2026 (10 million prompts, 20 million fan-out queries). https://peec.ai/blog/chatgpt-searches-in-english-even-when-you-don-t
↩︎
New Front Door to the Internet: Winning in the Age of AI Search. McKinsey & Company. https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search ↩︎
Your Next Customer Will Find You Using AI. Now What? Bain & Company. https://www.bain.com/insights/your-next-customer-will-find-you-using-ai-now-what/
↩︎
AI Search Referral Traffic Statistics 2026. Semrush and Conductor data. https://thestacc.com/blog/ai-search-referral-traffic-stats/ ↩︎
Schema Markup and AI Citations: What the Data Actually Shows (2026). fSEO, summarising Ahrefs and Princeton GEO research. https://fractionalseo.services/blog/schema-markup-ai-citations-data/ ↩︎

How to Increase Brand Visibility in AI Search: A 90-Day Content Strategy for Vietnam Brands