Schema markup helps AI answer engines understand what your content is — not just that it exists. A handful of schema types (FAQPage, HowTo, Article, LocalBusiness) meaningfully raise your probability of citation. Most others — including the much-hyped llms.txt file — are essentially irrelevant today. Here’s exactly what to implement, what to skip, and why the order matters.
Quotable definition: Schema markup for AI search is structured data embedded in a webpage’s raw HTML that signals to AI crawlers — GPTBot, ClaudeBot, PerplexityBot and others — the type, author, date, and entities within a piece of content. It does not guarantee citation, but it raises the probability that an answer engine correctly identifies and attributes your content when composing a response.
Why AI Crawlers Don’t See What Your Browser Sees
Most AI crawlers — GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot — do not execute JavaScript. They fetch raw HTML and stop there. This means anything rendered client-side: React components, lazy-loaded content, dynamically injected schema via Google Tag Manager — none of it is visible to these bots. If your schema lives in a JavaScript bundle rather than the initial HTML response, you might as well have not written it.
This is the single most common mistake Kaizenaire sees in technical audits of SG SME sites: schema implemented correctly for Google, invisible to every AI crawler. The fix is straightforward — server-side rendering or static HTML injection of your JSON-LD. Check your page’s raw source (Ctrl+U in Chrome) and look for the <script type="application/ld+json"> block. If it isn’t there on page load, AI crawlers aren’t reading it.
The Schema Types That Actually Move the Needle
Not all schema is equal. The types below have a demonstrable relationship with AI citation behaviour, based on how answer engines decompose content into factual assertions. Implement these first, in this order.
- FAQPage — The highest-leverage schema type for AEO. It presents question-answer pairs in a structure that mirrors exactly how AI engines retrieve and format responses. For a Singapore SME, this means your FAQ answers are directly extractable without the engine having to infer meaning from prose.
- Article / BlogPosting — Signals authorship, publication date, and content type. AI engines weight recency and named authorship. An
Articleschema with a realauthorentity (name, URL, same-page credentials) raises citation probability, particularly for factual how-to and explainer content. - HowTo — For step-by-step content, this schema type allows AI engines to extract individual steps as discrete data points. If you publish process guides — how to register a business, how to apply for a grant — mark them up.
- LocalBusiness — Particularly relevant for Singapore SMEs serving a geographic market. Correct address, opening hours, service area, and
areaServed: Singaporesignals local relevance to both Bing (which powers ChatGPT Search) and Google’s AI Overviews. - Product / Service — For e-commerce or service businesses, structured product data (name, description, price range, currency in SGD) feeds AI engines that construct comparison responses.
The Comparison: What Helps vs. What Doesn’t
| Schema / Signal | Useful for AI Citation? | Why |
|---|---|---|
| FAQPage (server-rendered JSON-LD) | Yes — high value | Directly extractable Q&A pairs; mirrors AI response format |
| Article + named Author entity | Yes — high value | Signals recency, authorship, content type; cited in attribution |
| HowTo with discrete steps | Yes — medium value | Steps extractable as structured facts; reduces inference load |
| LocalBusiness with areaServed | Yes — for local queries | Required for geo-targeted AI responses; Bing/ChatGPT dependency |
| BreadcrumbList | Minimal | Helps Google SERP display; AI engines largely ignore it |
| SiteLinksSearchBox | No | UI hint for Google only; no AI citation value |
| llms.txt file | No — evidenced | Ahrefs found 97% of domains with a valid llms.txt got zero requests for the file |
| Schema injected via JavaScript / GTM | No | AI crawlers don’t execute JS — schema is invisible at crawl time |
The Bing Prerequisite Most SG Owners Miss
ChatGPT Search is built on Bing’s index. That’s not a rumour — it’s how OpenAI structured the integration. If your site isn’t indexed in Bing, you’re invisible to ChatGPT Search regardless of how clean your schema is. Most Singapore SME sites are indexed in Google but never verified in Bing Webmaster Tools. It takes roughly 20 minutes to submit your sitemap to Bing — and it’s free.
Check your Bing indexation first. Go to bing.com/search?q=site:yourdomain.com. If you get zero results, your schema work is upstream of a blocked pipeline. Fix the indexation before optimising the markup.
[VERIFY: Bing’s market share in Singapore versus Google as of mid-2026 — for context on relative indexation priority]
The llms.txt Myth — and Why It Spread So Fast
In late 2024, a proposal circulated that websites could place an llms.txt file in their root directory — similar to robots.txt — to signal content preferences to AI crawlers. The idea spread quickly because it sounded plausible and required approximately 11 minutes of effort, which made it enormously appealing to anyone who’d been told they needed an AI strategy by Thursday.
Ahrefs tested this at scale. They found that 97% of domains with a valid llms.txt file received zero requests for it from any AI crawler. Not low traffic — zero. The major AI crawlers simply don’t implement the spec. Implementing llms.txt today is the digital equivalent of putting out a “no unsolicited advice” sign in a room where no one can read English. Polite, perhaps. Effective, no.
This doesn’t mean the spec is permanently dead — but as of mid-2026, it has no measurable impact on AI citation. Don’t prioritise it.
Entity Consistency: The Invisible Work That Compounds
Schema markup doesn’t work in isolation. AI engines build what researchers call a “knowledge graph” about entities — businesses, people, places, topics. Your schema contributes to that graph, but only if the data is consistent across every mention of your brand online.
For a Singapore SME, this means: your business name in your LocalBusiness schema must exactly match your ACRA-registered name. Your address must match your Google Business Profile. Your author name must match your LinkedIn URL referenced in the schema. Inconsistency across these signals creates ambiguity — and AI engines resolve ambiguity by citing someone else.
[VERIFY: Specific percentage lift in AI citation from entity-consistent schema versus inconsistent — no reliable published figure as of mid-2026]
This is less glamorous than installing a new schema type. It’s also where most of the actual gain is.
The One Thing to Do This Week
If you implement nothing else, do this: open your site’s raw HTML source on your most important page. Search for application/ld+json. If it’s absent — add an Article schema with a named author and a FAQPage schema with your three most common customer questions, both server-rendered. That single change, done correctly, does more for your AI citation probability than any amount of meta-tag optimisation.
The honest limitation worth stating plainly: schema markup improves your probability of citation. It doesn’t move traffic directly. AI citations currently drive a small fraction of total referral clicks — if you need top-of-funnel volume this quarter, schema alone isn’t your lever. It’s infrastructure for the medium term, not a short-cycle demand driver.
Frequently Asked Questions
Does schema markup guarantee I’ll appear in ChatGPT or Perplexity answers?
No. Schema raises your probability of being correctly identified and cited — it doesn’t guarantee placement. AI engines weigh many signals: content quality, domain authority, recency, entity consistency, and Bing/Google indexation. Schema is one factor in a layered system. Any agency telling you otherwise is overselling.
My developer uses React / Next.js. Will my schema still work?
Only if it’s server-side rendered or statically embedded in the initial HTML response. AI crawlers don’t execute JavaScript. Ask your developer to verify that the JSON-LD block appears in the raw page source — not injected after load. Next.js supports server-side rendering and static generation; the issue is whether your team has configured it correctly for the schema specifically.
Should I implement schema myself or hire someone?
FAQPage and Article schema are straightforward enough for a developer-literate owner to implement in an afternoon using Google’s documentation. The harder part is entity consistency across your ACRA registration, GBP, LinkedIn, and site — that requires an audit before implementation. For a single page, DIY is reasonable. For a full site with multiple schema types, a one-time technical audit saves rework.
Is llms.txt worth implementing for future-proofing?
Kaizenaire’s view: not yet. Ahrefs’ data shows 97% of domains with the file received zero AI crawler requests. The spec may gain adoption — but as of mid-2026, it has no demonstrated impact. Spend those two hours on FAQPage schema instead. Revisit llms.txt when major crawlers publicly confirm implementation.
What’s the fastest schema win for a Singapore SME?
LocalBusiness schema with correct ACRA-registered name, Singapore address, and areaServed: Singapore — then verify Bing indexation. ChatGPT Search runs on Bing. Without Bing indexation, no schema helps. These two steps take under an hour and address the most common gap Kaizenaire sees in SG SME technical audits.
How do I know if AI engines are currently citing my site?
Run searches in ChatGPT, Perplexity, and Google AI Overviews for your primary service + “Singapore.” Check whether your domain appears in citations or source links. It’s manual, but it’s the most direct signal available. A structured audit — like Kaizenaire’s free AI-Visibility Check — maps this systematically across the queries that matter to your business.
—
If you’re not sure which of these gaps apply to your site, the free AI-Visibility Check maps your current schema state, Bing indexation, and entity consistency in one audit. It’s a starting point, not a sales call — and it tells you exactly where to spend the next hour. You can also see the full technical and editorial scope of what Kaizenaire’s AEO/GEO/SEO service covers if you want to understand what ongoing work looks like.