Schema Markup Guide for FMCG Product Pages

An AI engine comparing products for a "best X for Y" query does not read marketing prose. It compares structured attributes: ingredients, certifications, allergen declarations, price and availability, pulled from schema markup rather than from a paragraph about how the brand was founded. A product whose ingredient list exists only as text inside a packaging photo is functionally invisible to that comparison, no matter how good the product actually is.

Why Structured Data Matters More for FMCG Than for Most Categories

One 2025 analysis of pages cited by Google's AI Mode found structured data present on roughly 65% of them, well above typical site-wide adoption rates. A separate analysis put the figure at 81% of AI-cited pages carrying at least one of five common schema types. The two figures do not agree exactly, both are vendor-sourced rather than peer-reviewed, and they should be treated as directionally consistent rather than as one precise number. What they agree on directionally is enough on its own: structured data is present on AI-cited pages far more often than chance would predict, and FMCG's core content, ingredients, certifications, nutrition values, is exactly the kind of information structured data is built to carry.

The technical prerequisite behind all of this is easy to miss. Schema needs to be server-rendered JSON-LD, because the crawlers AI systems use, GPTBot, ClaudeBot, PerplexityBot and similar, execute minimal JavaScript. Markup injected into the page after client-side hydration is frequently invisible to exactly the systems it was built for, which means a technically "correct" schema implementation can still fail in practice if it depends on JavaScript to appear.

The Schema Properties FMCG Product Pages Actually Need

Standard Product schema, name, brand, description, image, is the baseline, and it is necessary but not sufficient for FMCG. The properties that matter specifically for consumer goods extend well beyond that baseline: a full ingredients field rather than an abbreviated summary, NutritionInformation with per-serving values for food and beverage products, additionalProperty entries formatted as PropertyValue for dietary certifications such as halal, kosher or vegan status and for allergen declarations, AggregateRating drawing on real review data from verified sources, and offers with availability and retailer-specific purchase URLs so an AI system can point a consumer to where the product is actually purchasable.

For an Indonesian FMCG brand specifically, halal and BPOM registration data belongs in this same structured layer, not just on a packaging badge. The practical implementation looks like this:

Locking the same certification numbers inside a badge image instead of this kind of markup means the data satisfies a compliance checklist while remaining unreadable to the systems that could actually use it to answer a halal-specific or safety-specific query.

The Semantic Spec Sheet: A Format Built for Comparison Queries

Beyond schema markup itself, one practitioner framework worth adopting directly is the semantic spec sheet, a structured, human-readable but AI-parseable format that bridges a brand's internal product data systems and the way an AI engine actually answers a comparison query. A semantic spec sheet for an FMCG product typically includes a fixed set of fields, in a consistent order, for every SKU:

This format earns citations specifically because it structures the differentiating attributes of a product in a form that can be lifted almost verbatim into a "Brand A versus Brand B" comparison answer. A comparison page built from prose descriptions of two products asks the AI system to do the comparison work itself, reading two paragraphs and inferring the differences. A page built from parallel spec sheets does that comparison work in advance, on the page, before the AI system ever has to.

The "Last Updated" field deserves specific attention. One vendor's 2026 analysis found that adding a visible freshness timestamp to product pages lifted citation rates from 42% to 61%. That is a single study, and the exact percentage should be treated as directional rather than guaranteed, but the underlying logic holds regardless of the precise number: AI systems appear to weight demonstrably current information over content with no visible indication of when it was last verified, which makes a freshness signal a low-cost, low-risk addition worth testing on any product catalogue.

Why This Matters More in Indonesia Than the Global Average Suggests

Indonesia is not a marginal market for this shift. One analysis puts Indonesia among the global leaders in generative search adoption, with roughly 37.2% of analysed search queries triggering an AI Overview box, a materially higher rate than many other markets report. Google's AI Overviews and AI Mode now fully support Bahasa Indonesia queries, not just English, which removes what was until recently a real gap for Indonesian-language content specifically.

The consequence for organic click-through rate is significant and already measurable. Top-ranking pages have been reported losing between roughly 35% and 64% of the clicks they would previously have received once an AI Overview appears above them, with overall organic click-through rate compressing by up to 50% on desktop and 30% on mobile when an AI Overview is present on the page. That compression is exactly why the shift described earlier in this piece, from ranking for a click to being cited as a source, is not an optional strategic pivot. It is a direct response to a click-through rate that is already shrinking on the pages that used to rely on it most.

Regional marketplace-native AI features add a further layer specific to Southeast Asia. Lazada's in-app assistant, LazzieChat, built on OpenAI's models through Azure OpenAI Service, already retrieves product descriptions and review sentiment directly and inserts purchase links into its own conversational interface. A product whose data is inconsistent between its brand site, Shopee listing and Lazada listing is not just a schema hygiene problem in the abstract. It is a live risk inside a shopping assistant already operating at scale in the region.

Perfect schema on a page nobody's crawler can reach accomplishes nothing. Before any structured data audit, a robots.txt check is worth doing first, because it is the single most common reason a technically well-built page fails to appear in AI answers at all. The crawlers behind the major conversational AI systems, GPTBot and OAI-SearchBot for OpenAI's systems, ClaudeBot for Anthropic, PerplexityBot for Perplexity, and Google-Extended for Google's AI features, are separate from the standard Googlebot crawler most sites already allow, and a robots.txt file written years ago, before these crawlers existed, frequently blocks them by default through a catch-all disallow rule that was never updated.

Consent management platforms add a second, less obvious version of the same problem. A cookie or privacy consent banner configured under Indonesia's Personal Data Protection Law that blocks all scripts and content rendering until a visitor clicks through it will also block a crawler that never clicks anything, since crawlers do not interact with a page the way a human visitor does. A page that looks completely normal to a human tester in a browser can be entirely blank to an AI crawler if the consent implementation was never tested against a non-interactive request.

A practical framework worth internalising treats every product page as serving three distinct audiences simultaneously, each with a different tolerance for the same gaps. Human shoppers, still the large majority of traffic, want keyword-rich, benefit-led copy that converts, and are put off mainly by thin titles or weak social proof. AI-assisted humans, a smaller but rising share, need depth of question coverage and citable, specific claims, and are lost by vague phrasing that never directly answers the question they asked the AI. Autonomous AI agents, currently a tiny but fast-emerging share, need structured attribute completeness with no contradictions across surfaces, and treat any data gap or inconsistency as a hard filter with no second look, unlike a human who might read past a small inconsistency without noticing.

Writing only for the first audience now means writing for one audience out of three. The practical way through is not three separate pages. It is writing to the shared rubric underneath all three: question-depth, claim citability, cross-surface consistency and comparison-readiness, which happens to satisfy a human reader as well as it satisfies the two audiences most content teams have never explicitly written for.

The Consistency Requirement Most Brands Miss

Schema on a single page is not the finish line. The same ingredient list, certification numbers and pricing need to match across a brand's own site, its Shopee listing and its Tokopedia listing. Generative systems cross-reference product details across these surfaces to verify accuracy, and a discrepancy, an ingredient listed differently, a certification number that does not match, is treated as a trust risk that can result in a product being excluded from a recommendation entirely rather than simply ranked lower. A single, well-maintained source of product data feeding every surface consistently is worth more than a beautifully structured page that disagrees with the brand's own marketplace listing three clicks away.

This technical foundation underpins everything covered in more strategic depth in our overview of GEO for FMCG. It is also the same product data layer that supports the category and product page work described in SEO for FMCG, since a clean, structured product feed benefits a category page's search ranking and an AI engine's citation decision at the same time.

Frequently Asked Questions

Do we need different schema for our brand site versus our marketplace listings?

The schema implementation differs by platform, since marketplaces control their own listing structure, but the underlying data, ingredients, certifications, pricing, needs to match exactly across every surface. Inconsistency between a brand site and a marketplace listing is treated as a trust risk by AI systems cross-referencing product claims.

Is structured data alone enough to get a product cited by AI?

No. It is a necessary foundation, not a complete strategy. Third-party corroboration, reviews, editorial mentions and community discussion, plays a significant role in AI citation decisions independent of how well a product's own page is structured. Schema makes a product legible. It does not make it trusted on its own.

How do we handle products where the ingredient list changes between batches or formulations?

Update the structured data and the visible ingredient content at the same time the formulation changes, and use the "Last Updated" field to signal the change explicitly. Outdated ingredient data creates the exact kind of cross-surface inconsistency that AI systems treat as a trust risk, and it is a compliance issue independent of any AI consideration.

Should smaller FMCG brands with limited development resources still attempt this?

Yes, starting with the priority order rather than attempting everything at once. Core product identity and offers with availability are inexpensive to implement correctly and carry a disproportionate share of the citation benefit relative to the effort involved, compared with more elaborate schema work further down the priority list.

Does adding schema markup slow down page load speed?

Properly implemented JSON-LD adds negligible load time, since it is typically a small block of text rather than a rendering-heavy element. Page speed problems on FMCG sites are far more often caused by unoptimised images and third-party scripts than by structured data markup.

A full walkthrough of building an AI-ready product data foundation, including the schema and entity-building framework referenced in this piece, is covered in Cited or Silent. The free gated edition is available now, alongside Kindle, Google Play and Apple Books editions live internationally.

Get the Free Ebook

Written by Tessar Napitupulu, Founder and CEO of PT Arfadia Digital Indonesia, Forbes Agency Council member, and author of Found Before They Search and Cited or Silent. Arfadia has positioned itself as Indonesia's GEO pioneer since 2023.